Regulators rewrote the ad auction rules and left the owners alone

Six days is not usually enough time to see the shape of an industry. This week was an exception, because four separate authorities looked at the same machinery from four directions and reached versions of the same conclusion.

A federal judge in Virginia found on September 2 that Google had illegally monopolised two advertising markets, then declined to take either of them away, ordering instead that the auction be run differently. The European Commission fined Alphabet 460 million euros for ranking its own services first, and gave it sixty days to change the ranking rather than the ownership. The Federal Trade Commission and 22 state attorneys general sued Amazon on August 31 over what advertisers were told about a pricing rule, not over the fact that Amazon runs the auction, sells the inventory and reports the results. The General Court in Luxembourg confirmed on September 2 that a browser shipped with Windows need not offer a choice screen, because it is not used enough.

Nobody had to sell anything. Everybody has to explain something.

The counterpart to that regulatory pattern arrived on Friday morning from the buy side, where the company that spent a decade arguing that an independent demand-side platform could compound against the walled gardens cut about 575 people, roughly 15 percent of its staff, three weeks after guiding to its first annual revenue decline. The rules changed. The competitive structure the rules were meant to protect got smaller anyway.

Underneath both stories runs a third, quieter one. In six days, at least nine separate measurements of the same commercial activity moved: television currency, impression counting, attribution windows, view definitions, attention seconds, download rules, price comparisons between two Google surfaces, token consumption and the difference between a counted view and a watched one. Almost none of those changes were negotiated. Most were announced. A few were simply noticed by somebody with a script.

What follows is the week in the order the arguments make sense, rather than the order the wires filed them.

Seven changes to the currency, in one morning

Monday opened with the least discussed and most consequential kind of event in television advertising: a measurement supplier changing its method on a date it chose.

Nielsen deployed seven methodology changes to its national Big Data plus Panel service on August 31, 2026. PPC Land had listed the schedule when it was set, covering co-viewing capture through wrist-worn wearables, a latency-adjusted DASH universe estimate, a household demographic assignment model, integrated weighting, Hispanic universe estimation, an ACR monitored tuning adjustment and householding for one big data provider. Nielsen guaranteed no ratings increase.

Each of those seven touches a different layer of the same calculation. One decides who counts as present in the room. One decides how many households the sample stands for. One assigns demographics. One applies the weights. One collapses duplicate households supplied by a single vendor. Any single change would move delivered impressions measurably. Landing all seven on the same morning makes it impossible to attribute an August movement to any one of them, which is why the change forces buyers to re-check August delivery rather than roll forward from July.

The thing being remeasured is shrinking. Nielsen's second-quarter Ad-Supported Gauge, released on August 25, put advertising-funded television at 71.5 percent of total United States viewing in the April to June quarter, down 1.3 points from 72.8 percent in the first quarter and down from 74.2 percent in the fourth quarter of 2025. Inside that bloc, streaming took 48.2 percent, up 1.6 points on the quarter. Broadcast took 26.6 percent, down 1.6 on the quarter but up 0.6 on the year. Cable held at exactly 25.2 percent.

Two features of that release deserve more attention than the topline received. The first is the population base. Those shares are calculated on persons aged two and over. Switch to adults 18 and over, the base against which most guarantees are written, and streaming falls to 44.4 percent, broadcast rises to 28.6 percent and cable rises to 27.0 percent. The gap between streaming and broadcast narrows from 21.6 points to 15.8. Same quarter, same data, two different markets depending on which line is read.

The second is the denominator. A category taking a larger share of a contracting bloc is not the same thing as a category growing. Streaming gained 1.6 points of a pool that lost 1.3 points of the whole. Both sentences are true, they describe different phenomena, and they appear in the same table.

Sport distorted the quarter. The FIFA World Cup opened on June 11, three weeks before the quarter closed, with FOX holding English-language rights to 68 of the 104 matches and Telemundo carrying Spanish coverage. The tournament generated more than 25 billion advertising impressions across FOX, FS1, Telemundo and Universo. In the June monthly Gauge, broadcast reached 19.8 percent, its first June share increase since the Gauge launched in 2021, on a 118 percent month-over-month rise in broadcast sports viewing. Peacock recorded its first profitable quarter in the same window at 189 million dollars of adjusted EBITDA, with advertising revenue per subscriber up 43 percent to 4.98 dollars a monthand Domestic Media advertising revenue up 55 percent to 2.16 billion dollars. Comcast noted that none of those events recur in the third quarter in the same form.

A caveat sits inside the Gauge release that will matter more in six months than it does now. The second-quarter report has not yet migrated to Advertising Research Foundation DASH-based universe estimates. That migration is planned for autumn 2026, and it will produce results that are not comparable with current production data. Anyone reading the second-quarter and eventual fourth-quarter reports side by side will be comparing two measurement regimes.

Then there is the lag. The second-quarter Gauge covers April to June and appeared on August 25, roughly eight weeks after the quarter closed. The June monthly Gauge appeared on August 18; the May edition on July 28. Upfront negotiations largely conclude inside that interval, which means the commitments a quarterly number might have informed have generally already been made by the time it is published.

Nine attentive minutes out of two hundred

Six days before those changes deployed, a second measurement company published the number that sits underneath all of them.

TVision announced Ad Scoreboard 2.0 on August 25. Of more than 200 minutes of daily video consumption, roughly 24 minutes represent advertising opportunity, and about nine minutes qualify as attentive advertising time. That works out at approximately one attentive advertising minute for every 22 minutes of viewing, or under five percent of the total.

The method is observational rather than modelled. TVision runs a panel of 5,000 United States households fitted with computer vision sensors, automatic content recognition and digital meters, capturing four variables per second: whether the screen shows programming or an advertisement, how it is delivered, which individuals are in the room, and what their attention level is. The central metric is viewers per viewing household, which counts people rather than devices.

That metric is where three systems collide. Co-viewing is now counted three different ways, by three organisations, in the same households, using incompatible methods. Nielsen added wrist-worn wearables to the national currency on Monday. Google applied a Total Co-view definition to seven reach metrics on June 2, 2026, expanding reported unique users without changing delivery. TVision uses computer vision. A buyer running linear, YouTube and open connected television in one campaign receives three co-viewing figures that cannot be reconciled, each authoritative inside its own environment. Video Advertising Bureau research put co-viewing at 60 percent on premium connected television against 45 percent on YouTube, a 15-point difference in how many people a single impression is credited with reaching.

Ownership complicates the independence question. Viant closed its acquisition of TVision on May 5, 2026 for 40 million dollars, 22.5 million in cash and 17.5 million in stock. Viant reported second-quarter revenue of 104.3 million dollars, up 34 percent, against a net loss of 1.8 million, and told its August 10 earnings call that integration of the eyes-on-screen attention data was running ahead of expectation. A measurement panel owned by a demand-side platform supplies data that the same platform's buyers use to justify placements, a structure the industry has objected to elsewhere.

The announcement also left the usual gaps: no measurement window, no demographic breakdown of the sample, no statement of whether the nine-minute figure derives from the panel or a third party, no Media Rating Council accreditation, no pricing and no client roster.

Attention measurement has been consolidating regardless. Lumen Research brought its methodology to Netflix advertising across five European markets in March. Adelaide integrated with Amazon DSP in June and DV360 custom bidding in late July. Teads combined Adelaide, Lumen and TVision into a single layer with CTV Ensemble in June. InMobi claimed 3.5 times more attentive seconds than YouTube equivalents in July. The Video Advertising Bureau added its own frame during the week, finding that video occupies 51 percent of American leisure time for a fourth consecutive year, at 2 hours 37 minutes daily against 35 minutes socialising and 24 gaming, out of a 5 hour 10 minute total.

Put the two Monday releases together and the picture is a market spending most of its money against a currency that changed that morning, while a second layer suggests that under five percent of daily viewing minutes are spent attentively watching the advertisements funding it.

The invented bidder

Later on Monday, the week's first legal filing landed, and it concerns the single line of arithmetic that makes a bid rational.

The complaint in United States v. Amazon.com, Inc., case 2:26-cv-03097 in the Western District of Washington, is not an antitrust case. It was brought under Section 5(a) of the FTC Act and the deceptive trade practices statutes of 22 states, and its subject is pricing logic. PPC Land reported that the filing describes a two-stage calculation in which Amazon runs a generalised second-price auction and then applies an undisclosed soft reserve on top of the result, converting the auction price into the price actually charged.

The distinction is sharper than it sounds. A conventional reserve is published before an auction runs, and a bid that fails to clear it does not compete. The mechanism described in the complaint operates after the winner and runner-up have been determined, and its only ceiling is the winning bid itself.

Amazon's own staff named it accurately. A senior scientist described it as an invented auction participant representing how much Amazon thinks a particular slot is worth. A senior vice president in the advertising organisation wrote that the second price is not set by an actual bidder but is a proxy calculated by Amazon.

The compounding shows in one figure: the first price rate, meaning the share of clicks charged at the advertiser's own bid rather than at anything resembling a runner-up. In late September and October 2020 that rate sat at 4 percent. During 2021 it ran between 30 and 40 percent. By 2022 it reached 70 percent. In 2024 it was 79.1 percent, and for Sponsored Brands in the same year winners paid their own bid roughly half the time. Digiday's explainer, published on September 1, described the same trajectory as a surcharge occurring around 80 percent of the time by 2024, up from 30 to 40 percent in 2021.

Deployment was staged. Soft reserve pricing began in Sponsored Brands during the 2018 Christmas period, extended to Sponsored Products second-price auctions in mid-2019, and reached Display Ads in 2023.

What gives the complaint its texture is the internal record of people watching the seams. A Sponsored Brands employee wrote in 2018 that price increases were masked by holiday increases in advertiser demand. The head of Sponsored Products auctions noted that surcharge constraints could be relaxed when there were enough confounders, seasonality among them, to make bid shading hard to detect. A Sponsored Products employee in 2020 spelled out the exposure: an advertiser seeing most clicks charged at exactly the specified bid would change how bidding was approached.

December 2021 supplies the clearest case study. On December 10, 2021 the auction team removed a click-level surcharge constraint and prices climbed over five days. By December 13 more than 20 agencies and advertisers had escalated. On December 17 the vice president of Sponsored Products convened a war room. On December 20 the team drafted a response attributing the movement to Black Friday and Cyber Monday patterns. When an internal summary was revised on February 9, 2022, the discussion of generalised second-price mechanics and reserve pricing had been removed. A product management colleague objected on February 1 that the revision made it sound as though Amazon Ads had no role in the fluctuations, which was not the case. A director instructed that the document should not mention generalised second price at all.

Prime Day 2023 runs the same play in reverse. On July 11 the auction team reported cost-per-click below expectations. On July 12 it obtained approval to raise the surcharge constraint, on an order form stating the purpose as Prime Day return maintenance. An account manager relayed a grocery client's complaint that cost-per-click was up more than 90 percent.

Then there is the vocabulary. In April 2024 a scientist was instructed to strike the word surcharge from an internal conference slide, with guidance to avoid it in any communication. In July 2024 the Sponsored Products team queried a chatbot for alternatives, weighing value adjustment, price refinement, inflated CPC, price exaggeration factor, performance premium and profit extraction lever, and settled on performance premium. In September 2024 a research paper on how advertisers react to changes in ad auction pricing was pulled from Amazon's public science page and from the KDD conference site.

The deception count rests on what advertisers were told instead. Since at least 2014, the complaint alleges, Amazon represented that winners pay slightly higher, or one penny more, than the second-place bidder. That representation appeared on the website, on a training channel, in Amazon Ads Academy courses, in webinars for Kindle Direct Publishing authors, in the Sponsored Brands developer guide posted through July 2026, and in pitch decks to Sony, Samsung, Harman JBL and Coach between 2020 and 2023. In February 2024 a customer service agent explained to an advertiser that a two-dollar bid against a one-dollar next bid would be charged 1.01 dollars.

Reporting design closed the loop. Invoices were not itemised by click, reporting aggregated at keyword level, and desktop and mobile placements were combined. A 2019 operating plan referred to auctions with highly obfuscated reporting. An internal memorandum concluded that Sponsored Brands advertisers were unlikely to adjust bids because they had no visibility into individual auctions. The head of Sponsored Products auctions observed in 2020 that dynamic bid adjustments, which can raise a bid by up to 900 percent, mean an advertiser does not even know what bid is being entered.

Scale explains the regulatory interest. Roughly 1.2 million United States advertising customers are covered, more than 500,000 of them small and medium-sized businesses, against a business that grew 26 percent year on year to 19.8 billion dollars in the second quarter of 2026. The 20 billion dollar surcharge figure arrives with average surcharge rates, reserve guardrail sizes and several dollar estimates redacted from the public version.

Amazon rejects the case in detail. The company says nobody was overcharged, that average Sponsored Products cost-per-click was flat in inflation-adjusted terms between 2019 and 2024, that conversion rates rose more than 24 percent for individual advertisers between 2021 and 2025, and that roughly 92 percent of selected Sponsored Products ads in 2024 were not the highest bid, since ranking weighs relevance alongside price. It puts the mean winning bid at around the twelfth bid by amount, estimates 58 percent higher sales and 46 percent better return against a bid-only ranking, and calculates that advertisers saved more than 8 billion dollars between 2021 and 2025 as a result. It characterises the evidence as a small handful of outdated training materials and stray internal emails, notes that reserve pricing is common across the industry, and points out that no advertiser paid above a submitted bid.

Two dates in the disclosure timeline belong together. Amazon removed the second-price representation from a marketing page in October 2024, after learning of the investigation. It informed the FTC of that removal on August 26, 2026, five days before the complaint landed. The first public reference to reserve pricing appeared on a single Support Center page on October 30, 2025, reachable by search, and was updated in March 2026 to say that the price charged may exceed the runner-up bid but will never exceed the authorised maximum.

The FTC seeks a permanent injunction under Section 13(b). The states seek injunctive relief, rescission or reformation of contracts, restitution, disgorgement, civil penalties and costs, with Maryland alone providing for up to 10,000 dollars per violation. The Commission authorised the filing two votes to nil.

Context sits close by. The FTC opened consumer protection inquiries into both Amazon and Google over search advertising pricing disclosures in September 2025. A federal court ordered Google to disclose material auction changes the same month. Teads filed a complaint in July 2026 detailing manipulation at the level of individual impressions. What is new is the venue: the first regulatory action aimed squarely at auction mechanics inside a closed retail media environment, where the seller of the inventory, the operator of the auction and the reporter of the results are the same company.

A smaller Amazon story published the same morning describes the other end of the same opacity. PPC Land documented a United Kingdom marketplace seller whose account was suspended for counterfeit violations three days after Amazon had reinstated the listing at the centre of the complaint, following a two-week pattern in late July 2026 in which coordinated buyer accounts placed orders, requested refunds and left one-star feedback carrying scam accusations. Negative feedback aggregates against an order defect rate with a one percent threshold. A forum moderator replied that he had almost zero ability to assist with buyer-related situations because of limited visibility.

Different surface, same architecture: an automated decision, an aggregated report, and no view of the inputs.

The external check has a correction factor of its own

The natural response to a platform whose numbers cannot be audited from inside is to audit them from outside, which is what marketing mix modelling exists to do. Research reported on August 31 argues that the external check carries a systematic bias of its own.

Niklas Heusch of Zalando posted structural estimation work as arXiv:2608.21128v1 on August 21. Against a synthetic dataset with a known true paid search return of 4.20 times, a standard marketing mix model returned 10.61 times, roughly two and a half times the truth, with a 90 percent credible interval of 6.56 to 14.36 that never contains the correct answer.

The construction is deliberately generous to the model. Data covers 156 weeks and three channels for a simulated online retailer, with four geo-experiments at different spending levels and periods, estimated in PyMC using the No-U-Turn Sampler across four chains and 2,000 sampling iterations, with a Beta(1,3) adstock prior and a Gamma(3,1) saturation prior.

The critical result is the control condition rather than the headline. Given oracle controls, meaning the true confounders handed to the model directly, the estimate improved only to 8.41 times, interval 6.91 to 9.81. Bias persisted. This is not a data-volume problem.

The cause is endogeneity. Marketing budgets are not randomly assigned: companies raise spending ahead of high-demand periods, bidding systems chase performance, and budgets respond to prior results. Standard specifications assume errors are independent conditional on observed controls, a condition the paper describes as virtually impossible to satisfy in practice. The model observes a correlation between spend and sales that the business created by spending when sales were expected.

The proposed alternative treats a geo-experiment as two parallel universes differing only in the tested channel's spend, subtracts control outcomes from treatment outcomes to remove the common component, and estimates response parameters from the difference. With two tests the structural estimate landed at 4.31 times, interval 3.87 to 4.75. With four tests it reached 4.14, interval 3.79 to 4.48. Adstock decay came back at 0.19 against a true 0.20, saturation at 1.96 against 1.71, effectiveness at 0.20 against 0.18.

The practical obstacle is test volume. Google cut its minimum incrementality experiment budget to 5,000 dollars in November 2025, which lowered the entry price but not the design requirement. The method needs tests at different spending levels, a minimum of four dark weeks, and multiple test periods. One or two studies a year is below what structural estimation requires to identify a saturation curve.

A second and distinct problem sits alongside it. Ryan Dew of Wharton, Nicolas Padilla of London Business School and Anya Shchetkina of Wharton ran 2,187 simulation settings with 100 datasets each in work posted in August 2024. Without a stock variable, a nonlinear data-generating process produced conflation in 81 percent of cases, 27 percent of them major; a time-varying process produced 91 percent and 40 percent; a Hill function produced 83 percent and 46 percent. With high carryover at 0.8 the time-varying figure reached 99 percent. In a coffee brand case study, the nonlinear specification recommended 35 percent of spending on spot television and the time-varying specification recommended none, a divergence the authors priced at 227,000 dollars over 14 weeks.

Heusch's problem is that spending is not random. Dew and colleagues show a problem that survives even when it is. Both arrive at the same recommendation from opposite directions, which is that incrementality testing is under-used. Marketing mix modelling has been the industry's answer to platform self-reporting since well before Google released Meridian as an open-source implementation, and the answer now carries a documented correction factor.

Snap stops marking its own homework

The platform response to a credibility problem is to hand the counting to somebody else, and on August 31 Snap did exactly that at global scale.

Unified Attribution moved from beta to worldwide availability, with Snap citing a 26 percent lower cost per acquisition for a real money gaming advertiser measured through AppsFlyer against SKAdNetwork-optimised campaigns. Mohegan Sun's online casino recorded 89.8 percent higher return on ad spend and 77 percent lower cost per install in week four. Through Adjust, the French financial app Deblock reported a 25 percent lower effective cost per install, 28 percent lower cost per completed onboarding, 43 percent more installs and 31 percent more users reached; the publisher Codeway reported an 18 percent lower cost per install at 1.8 times the scale.

The mechanism is a transfer of authority. Unified campaigns use mobile measurement partner signals for optimisation and reporting, inheriting the partner's attribution windows with no advertiser customisation. The setting is fixed at campaign creation and cannot be changed afterwards, so switching method means rebuilding the campaign. AppsFlyer and Adjust are live; Singular, Kochava and Branch are in progress. Where an app connects to more than one supported partner, the system automatically designates whichever records the highest install volume as primary, and that designation can change without advertiser action as volumes shift.

Several qualifications belong beside the percentages. Snap's own documentation states four separate times that Unified Attribution is not designed to match partner reporting exactly, and its help text and FAQ describe the optimisation logic differently. Conversion reporting defaults to impression time, with modelling applied where privacy limits data. Accounts can display three concurrent conversion counts: Snapchat's native attribution, the partner-modelled Unified metrics, and optional SKAdNetwork postbacks. Standard and Unified cost per acquisition are not directly comparable, because they are computed differently.

More to the point, every figure quoted comes from an individual advertiser case supplied through Snap or its measurement partners. No aggregate performance across the beta cohort was published and no independent evaluation was cited. The claims describe attributed outcomes under changed counting rules, which is a different quantity from proven incremental lift, and precisely the quantity the Zalando paper warns is contaminated.

Commercially the change lands on a business that needs the credibility. Snap reported advertising revenue up 9 percent in the second quarter with ad prices up 10 percent, and its finance chief flagged World Cup demand as non-recurring. The wider pattern is established: Meta aligned with partner measurement in November 2025, TikTok added real-time iOS conversion tracking through Kochava in October 2025, Reddit began displaying dual attribution in May 2026, and OpenAI connected ChatGPT Ads to AppsFlyer and Adjust in July 2026. Google, Meta, Moloco and Unity took equity positions in AppsFlyer partly to keep the measurement layer neutral. Platforms are steadily surrendering the definition of a conversion to companies they do not control, because the alternative is permanent argument with the people holding the budget.

Brussels sets disclosure rules for two very different products

Regulation approached the same question from the supply side, and two European decisions landed within hours of each other on August 31.

The European Commission designated ChatGPT a Very Large Online Search Engine under reference IP/26/1772, after OpenAI declared 159.1 million average monthly users in the European Union, well above the 45 million threshold. Reddit, at 57.2 million, and Roblox, at 46.6 million, were designated Very Large Online Platforms in the same decision, bringing designated services to 28 and covering 262.9 million declared users between the three.

Compliance is due at the end of December 2026, four months from notification. OpenAI Ireland Limited is the main establishment, putting day-to-day supervision with Ireland's Coimisiun na Mean. Reddit Netherlands B.V. falls to the Netherlands Authority for Consumers and Markets. Roblox has no European Union establishment and has appointed a legal representative in the Netherlands.

Three obligations bear directly on advertising. Article 26 requires each advertisement to disclose the advertiser's identity, the basis on which it is presented and the parameters used to select its recipient. Article 28(2) prohibits advertising based on profiling where the service is aware with reasonable certainty that the recipient is a minor. Article 39 requires a public repository holding creative content, advertiser identity, campaign period, targeting parameters and reach metrics, retained for a year after last display. Annual independent audits and vetted researcher data access under Article 40 apply on top.

Timing makes this consequential rather than procedural. Digiday reported on August 31 that OpenAI's advertising business reached a one billion dollar annualised run rate in under 200 days, a figure derived by multiplying current monthly revenue of roughly 83 million dollars by twelve, and therefore a snapshot rather than money booked across a year. Self-serve access opened the same day to eligible advertisers across the 31 European markets where ads launched earlier in August. Sustaining a path to OpenAI's stated 100 billion dollar 2030 target implies compound annual growth of 216.2 percent.

PPC Land added detail during the week from Sensor Tower data showing that financial services advertising on ChatGPT jumped from 2 percent to 13 percent of spend between April and August, with mobile advertisement density more than doubling since spring. The product also loosened a targeting constraint on September 3, when OpenAI dropped the 25,000-user floor for advertisement exclusion audiences, allowing lists above five million members and accepting the Google Advertising ID, with an impression-billed campaign type optimising to view-through conversions described as next.

So the same product now runs contextual sponsored placements to Free and Go users in Europe, reported as arriving earlier in August, and must within four months publish a searchable repository describing how each placement selected its audience. Enforcement history suggests the repository requirement is not decorative. The Commission fined X 120 million euros on December 5, 2025 over transparency failures including its advertisement repository, then fined Temu 200 million euros on May 28, 2026 over illegal product risk assessment failures, and fined AliExpress 550 million euros on July 20, 2026.

The second European decision runs in the other direction, taking data out of a gatekeeper rather than requiring it to be published. Search Engine Roundtable documented the terms of Google's European Search Dataset Licensing Program, which makes anonymised ranking, query, click and view data from European Economic Area search activity available to qualifying rival search engines. The obligation traces to Google's designation as a gatekeeper on September 6, 2023 and to Article 6(11) of the Digital Markets Act, with the Commission adopting detailed compliance measures on July 16, 2026. PPC Land covered the underlying order requiring Google to hand rivals its search data when it was set.

Eligibility is narrow. An applicant must qualify as an online search engine provider under Digital Markets Act definitions, operate in the European Economic Area with services directed at European Economic Area users, not be operated or controlled by a non-European Economic Area state actor, not be controlled by a sanctioned entity, and either run for two consecutive years with at least 50,000 monthly average European Economic Area users or, if founded within two years, have secured at least 50 million euros of capital. Pricing follows fair, reasonable and non-discriminatory terms limited to incremental cost plus a specified rate of return. Recipients must accept purpose limitations, data segregation, access controls, detailed logging, third-party audit and reporting, and restrictions on international transfers. Applications go through a form.

One regime forces a platform to publish how its advertising selects an audience. The other forces a platform to share the signals that make its ranking work. Both answer the complaint arriving in the Amazon filing from an entirely different direction.

The meter Google built for itself

Wednesday's story was about what gets reported and what does not.

Google rolled the Search Console generative AI performance report and the AI feature control out to all websites worldwide as of August 31, 2026, ending a phased release that began in the first week of June. Five dimensions are available. Impressions record how often URLs appeared inside generative AI features in Search and Discover. Pages identify which URLs surfaced. Countries break visibility down geographically. Devices apply to Search only. Dates run at hourly, daily, weekly and monthly granularity.

Two dimensions are absent, and they are the two that convert visibility into revenue. There is no click data. There is no query data.

The gap is structural rather than accidental. Before June, appearances inside AI Overviews were folded into aggregate Web Search totals, which made attribution impossible in a different way: a publisher could see impressions rise and clicks fall without knowing which surface was responsible. PPC Land covered the report as the first time Search Console had given AI surfaces their own view when it appeared on June 3. It does not create a separate silo; AI impressions remain inside overall performance totals and the report carves out a dedicated view. So the number that has moved most sharply since AI Overviews launched in May 2024, the ratio between being shown and being visited, is still the number the tool declines to isolate. Research cited in that coverage documented a 61 percent reduction in organic click-through rate for queries where AI Overviews appeared. An Ahrefs study put the figure at 58 percent for top-ranking pages.

The control shipping alongside it has a clearer provenance. Google did not volunteer it. The toggle arrived as a response to a United Kingdom order, the first binding conduct requirement imposed on the company under the Digital Markets, Competition and Consumers Act 2024. The Competition and Markets Authority designated Google with Strategic Market Status in September 2025, on the basis that it accounts for more than 90 percent of general search queries in the United Kingdom. The requirement obliges Google to let publishers exclude content from AI Overviews, AI Mode and AI features in Discover, either as links or as grounding material, without touching conventional rankings.

One clause anticipates the obvious move. Google shall not seek to prevent publishers from using the controls, nor penalise them for doing so by reducing conventional search visibility. Regulators wrote an anti-retaliation provision into a remedy about a checkbox, which indicates how much trust the arrangement carries.

The compliance calendar is long. Main control obligations take effect on December 3, 2026. Page-level granularity, meaning the ability to exclude specific URLs rather than a whole domain, is not required until March 3, 2027. The first CMA compliance report is due on January 3, 2027. The requirement expires on October 10, 2030. Between now and December, publishers hold a domain-wide switch, an impressions count, and no way to price the trade.

Search Console's own recent record does not help. The September 2026 Google Webmaster Report lists a performance report incident around August 12 in which impression and click figures dropped without cause, crawl stats missing two days of data, and a favicon reporting bug since resolved. It also notes that AI Mode queries are now displayed separately from the generative AI reports, which means the AI reporting surface is currently split across two places measuring different things. The same summary records the August 2026 spam update rolling out between August 18 and 21, ranking volatility on August 5 and 6 and again on August 12 and 13, AI Overviews running on Gemini 3.7 Flash, and the shutdown of Google Assistant on mobile devices scheduled for September 4.

The turnstile

If the first half of Wednesday is about what Google chooses to report, the second is about what anybody else is still allowed to look at.

Search Engine Roundtable documented a new interstitial requiring users to click a Continue button before search results appear. The mechanics are deliberate. The button renders greyed out, becomes blue after a few seconds, and only then accepts a click. Glenn Gabe, who first shared the screenshot, described the pause exactly. Alexis Rylko observed the same screen in other language versions, which suggests a test running across multiple markets rather than a single locale. The delay is the point: it gives the back end time to complete a server-side check before the page becomes interactive, defeating automation that expects a results page to be one request away.

This sits inside a pattern with an eighteen-month history. In September 2025 Google eliminated the num=100 parameter, which had let a single request return a hundred results, forcing a roughly tenfold increase in API calls for anyone tracking the top hundred positions. Semrush confirmed significant operational impact. Ahrefs eventually restored top-100 tracking at higher cost.

Then came the passthrough links. Alex Greenland observed the first google.com/goto URLs on June 23, 2026. Brodie Clark documented the pattern across organic and paid listings on July 2. By August 26, 2026 the rollout was close to complete and Google had acknowledged it. Destination URLs in results are now encoded strings resolving through a 302 redirect, and they cannot be decoded locally. HEAD requests are blocked. The only workaround is a GET request that stops at the redirect, which turns one lookup into one network round trip per link. Derek Perkins of Nozzle put the arithmetic on the record: resolving all links for a single five-page ranking takes between 500 and 1,000 requests.

The bill lands unevenly. PPC Land reported during the week that SISTRIX has been forced into an added-cost tracking workaround, with each resolved goto address costing an extra server request. Google's public framing has been consistent and unspecific: a long history of deploying technical measures against evolving forms of abuse. No target was named, no technical documentation followed, and no rate limits were published, which leaves vendors to infer the boundaries by hitting them.

Litigation runs alongside. Google sued SerpApi over scraping, lost a bid to treat search scraping as a Digital Millennium Copyright Act circumvention question, saw SerpApi file to dismiss, restricted a Light Fast API workaround to three results, and has amended the complaint to build on licensed content, including Reddit's licensing terms.

Put the two halves together and the shape is plain. Independent measurement of search results is getting more expensive, more legally exposed and technically harder, month by month. First-party measurement of AI search results has just become universally available, free, and silent on clicks and queries. Neither development is a secret. Each has a defensible justification on its own terms. The combined effect is that the number of organisations able to say what Google search actually does is falling.

Google Discover supplied a small demonstration of the resulting dependency on September 1. Articles failed to load or loaded only after long delays from roughly 6:00 a.m. Eastern, with Valentin Pletzer flagging the problem and Damien Adell posting traffic charts showing a sharp drop from 10:00 Central European Summer Time. By 7:40 a.m. Eastern, Adell reported recovery. Roughly one hundred minutes of publisher referral traffic went missing, with no accompanying statement. Publishers learned about it from each other.

Google unlinked its own products

The clearest failure of the week involved two Google systems losing track of one another.

On September 1, Local Service Ads accounts began detaching from the Google Business Profiles they depend on, without any action by the account holder. Jack Hepp reported an unlinked account requiring support intervention to restore, alongside repeated error messages, platform slowness and an inability to view Local Service Ads inside associated Google Ads accounts. Ginny Marvin, Google's Ads Liaison, confirmed a bug had been identified and was being worked on, and added that it was unrelated to the migration.

The migration is what gives the timing its edge. Google is retiring the standalone Local Services Ads dashboard and moving campaigns into a Performance Max variant inside Google Ads, starting in August 2026 with United States home service categories: plumbing, heating and air conditioning, electrical, roofing, pest control and related verticals. Service-area businesses and custom bidding accounts follow in late 2026. Non-United States accounts and remaining verticals move in 2027.

Much of the account survives the trip. Pay-per-lead billing continues. Placement stays limited to Search and Maps. Targeting remains keywordless and category-based. Lead history, contact details, message threads and call recordings transfer into a new Lead Manager interface, and verified business badges carry across without re-verification.

Historical performance reporting does not survive. Impressions, clicks, weekly spending and ad-level breakdowns disappear at transition. Administrators receive fourteen days of notice before their individual migration date and must download the data manually inside that window. Manual bidding disappears, as do vertical-level target CPA settings and Better Business Bureau callouts. Weekly budgets convert to daily amounts. Business Profile core details sync one way into Google Ads and can only be edited at the profile.

So the bug landed on accounts in the middle of a fourteen-day export window, on a platform returning errors, in a product whose historical record is about to be deleted, and it severed the link to the profile that the migration uses to establish identity. Google has said accounts without a Business Profile link would have one created automatically during migration, which is a reasonable safeguard for an account that never had a link and a less reassuring one for an account that just lost the link it had.

October brings another change to the same product. From October 1, 2026, Google begins charging for unanswered calls that last more than 20 seconds during business hours, with an exception where call routing requires a key press, and begins charging for qualifying follow-up calls from a user whose first contact did not qualify. The billing definition widens in the same quarter that the reporting used to audit billing narrows.

AdSense stops counting ads that never appeared

A second measurement change landed in the same 48 hours, with a longer fuse and a narrower audience.

Google notified publishers on September 1 that from February 17, 2027, AdSense and Ad Manager will count a display impression only once the advertisement has successfully loaded and started to render on the user's device. The current method counts at the moment the advertisement begins downloading. Search Engine Roundtable covered the notice on September 2, with Google saying the update aims to provide more consistent measurement and align with industry standards.

The technical detail matters more than the framing. The change applies to banner display inventory on web, mobile web and connected television, while native, interstitial, video and application formats already count on begin-to-render or an equivalent. The methodology is not new: it comes from IAB Technology Laboratory and Media Rating Council guidance published in October 2017. Google's own buying platforms, Display and Video 360 and Campaign Manager 360, moved to begin-to-render during 2025. The seller-side products follow seventeen months later, which means that for more than a year Google's buyers and Google's sellers have been counting the same event by different rules.

Impression counts will fall. Every download that never completed rendering, because a user navigated away, because a script failed, because a connection dropped, disappears from the total. No estimate has been published of how far. That silence is a departure. When AdSense made a comparable counting change in 2018, the notification included an explicit statement that earnings were not expected to be impacted. No such sentence appears this time.

The comparison window is the practical constraint. Historical data suitable for estimating impact begins on August 12, 2026, giving Ad Manager publishers roughly six months of baseline before implementation, and giving AdSense publishers, who lack equivalent reporting tools, considerably less. Year-over-year dashboards will carry a permanent discontinuity at February 17, 2027, and any publisher whose commercial arrangements are denominated in impressions rather than revenue will be renegotiating against a different denominator.

Two days earlier, a different Google measurement surface simply stopped working. Google Analytics failed to record September 1 traffic across GA4 properties, showing zero for the day in main charts while sites continued to receive visitors. Dana DiTomaso had flagged processing problems with August 31 data the day before. Search Engine Roundtable published the incident report at 6:13 a.m. Eastern on September 2. No official statement accompanied it.

What the court refused to do

Sixteen months after finding that Google had illegally monopolised two markets, the judge who made that finding declined to take anything away from it.

On September 2, 2026, Judge Leonie Brinkema of the United States District Court for the Eastern District of Virginia ruled that Google will not be required to sell its AdX exchange or its DFP publisher ad server. She adopted most of the behavioral remedies that the Department of Justice and Google had each proposed, rewriting how the open-web display auction works without changing who owns it. Her written opinion was sealed for fourteen days while both sides reviewed it for confidential material, which meant the industry spent the day arguing about a decision it had not read.

The liability finding came on April 17, 2025. PPC Land reported at the time that the court had found Google monopolized the publisher ad server market and the ad exchange market for open-web display advertising, under Sections 1 and 2 of the Sherman Act, with Google holding 91 percent of the publisher ad server market. Brinkema wrote then that Google had willfully engaged in a series of anticompetitive acts to acquire and maintain monopoly power. One government count failed: insufficient evidence of monopolisation in the proposed advertiser ad network market.

The remedies phase ran from September 22 to October 6, 2025. Final briefs landed on November 3, 2025, and PPC Land set out what each side had asked for. The Department of Justice wanted AdX divested within twelve months, DFP's final auction logic open-sourced within twelve months under a neutral administrator such as Prebid.org, and a contingent divestiture of remaining ad server components triggered at the three-year mark if competition had not returned. It asked for ten years of behavioral restrictions on top, a six-year independent trustee with full source code access, and an escrow fund to cover publisher switching costs. Google proposed behavior alone, monitored for six years.

Closing arguments on November 21, 2025 telegraphed the outcome. PPC Land's account of that hearing recorded Brinkema's skepticism about divestiture in almost the terms she has now adopted: the industry changes rapidly, and predictions about market conditions five years out remain uncertain. Matthew Huppert argued for the government that behavioral remedies would entangle the court in endless litigation and that only divestiture could remove illegally acquired monopolies. Karen Dunn argued for Google that behavioral relief could be implemented within twelve to eighteen months. Ari Paparo, watching, said the skepticism was enough to convince him the spinout would not happen. He was right, though the ruling arrived roughly seven months later than the January or February 2026 timeline suggested from the bench.

The reasoning rests on four practical objections rather than any softening on liability. There is no obvious buyer, since Microsoft would trigger an antitrust review of its own. The market moves faster than a three-to-five-year divestiture process can anticipate. Appeals would postpone any relief for years. And small publishers currently using DFP without charge could be harmed by disrupting the service. Behavioral remedies, by contrast, can begin to be enforced within a year.

Google's response was brief and precise about what it had won. Lee-Anne Mulholland, a vice president at the company, said it was very pleased that the court rejected the proposal to break apart tools that help small businesses. Search Engine Roundtable, recording the ruling the same morning, noted the framing circulating in wire coverage: the ad exchange is a small part of Google's business, and the decision is a second symbolic defeat for the Department of Justice after the search case also ended in behavioral rather than structural relief. A parallel PPC Land account recorded the same outcome from the government's side, with structural relief dead in Virginia: no AdX sale, no open-sourced DFP auction logic, and a 30-day deadline for a joint final judgment.

The mechanisms the order takes apart

The substance of what Brinkema ordered is more specific than the headline suggests, and it maps closely onto the conduct catalogued at trial.

Real-time bid data from AdX must be made available to rival ad servers. Unified pricing rules are to be deprecated. Publishers in Google Ad Manager will be able to set different price floors for individual bidders. First look and last look are prohibited on open-web display inventory. Divestiture is off the table, and so, for now, is the idea of handing AdX integration to Prebid.org, though Google signalled it would support that step.

Each item has a history. Unified pricing rules arrived in 2019 as part of Google's shift to a first-price unified auction, presented at the time as a simplification. PPC Land covered the announcement, in which Jason Bigler, then a director of product management, said Google would be able to provide publishers with reporting on all bids submitted for their ads, with a limit of 100 unified pricing rules per publisher. What publishers lost in the same change was the ability to price different demand sources differently, which is precisely the capability the court has now ordered restored. Seven years of auction design is being wound back to a state that existed before it.

First look and last look go back further. Index Exchange, which sued Google on November 10, 2025 in the same districtunder case number 1:25-cv-02003, described first look as dating to 2010, giving AdX the first opportunity to acquire any impression before other exchanges could bid, and characterised last look as the ability to open the envelope for the winning bid, know what it was, and bid after everybody else. Its single-count Sherman Act Section 2 complaint puts the annual value of the affected ad exchange market above 104 billion dollars and names Project Poirot, the 2017 programme that cut bids on rival exchanges by as much as 90 percent through DV360 while submitting full-price bids through AdX.

Teads followed on August 3, 2026 in the Southern District of New York, case number 1:26-cv-06591, with seven counts. Its central number is arresting: between 2017 and 2023, rival exchanges would have won 6.88 trillion additional impressions had Google Ads bid through both Google and non-Google exchanges the way its competitors do. That is more than a quarter of rival exchange volume in an untethered market. The complaint states that roughly 90 percent of Google Ads demand flows exclusively through AdX, and that Google Ads accounts for around 30 percent of programmatic impressions.

Those private cases are unaffected by the ruling except in one respect: the liability finding they build on survives intact, and the remedies order now supplies a public benchmark for what the court considers adequate relief. Damages claims proceed on their own timetable.

Brussels is on a different track and may yet reach a different place. The European Commission fined Google 2.95 billion euros on September 5, 2025 in Case AT.40670, finding an Article 102 abuse spanning 2014 to 2025, with Google at 91 percent of publisher ad servers and 60 to 70 percent of ad exchanges. The public version of that decision, released as structural remedies remained on the table, documents first look, last look, dynamic revenue share, unified pricing rules and Project Poirot. Google submitted behavioral commitments on November 13, 2025 that closely mirror what a Virginia court has now ordered, and filed an annulment action on January 12, 2026. Teresa Ribera has signalled that structural remedies appear necessary. Competition lawyer Damien Geradin called the proposed behavioral package entirely useless.

The same catalogue of conduct has produced a behavioral remedy in Virginia and an unresolved structural question in Brussels, on the same evidence.

Publisher reaction concentrated on enforcement rather than on the outcome. PPC Land reported that Digital Content Next argued Google must now deliver the ad tech fixes it promised, with Jason Kint making the case that auction data rather than display revenue was always the prize, and publishers waiting fourteen days for the sealed opinion and thirty for a joint judgment. Shamsul Chowhury, senior vice president for paid media at Zeno Group, said Google can take a deep exhale. Jay Friedman, who testified for the government and co-founded the advisory firm CartographAI, framed the gap the remedies leave open as a question about what a publisher does if it wants a different ad server but still wants Google's buy-side demand. Real-time bid access is meant to answer exactly that, and whether it does will be the practical test of the order over the coming year.

Friedman's own company published something on September 2 that reads differently after the ruling. CartographAI launched an accelerator programme aimed at ad tech vendors rather than at the brands and agencies it originally served, at 1,000 to 3,000 dollars a month, launching with sixteen participants including The Trade Desk, Viant, MNTN and Adelaide. A ratings firm selling marketing services to the companies it rates carries an obvious structural point. The more interesting observation is about timing. A market that has just been told it will keep its dominant integrated player, with rule changes rather than ownership changes, is a market where vendor selection carries more weight than it would in a world where AdX had been forced into independent hands. If the structure is fixed, navigation becomes the product.

Brussels names nine Google services and gives sixty days

The most detailed regulatory document of the week is not new law but newly legible reasoning.

PPC Land reported on September 3 on the text of Commission Implementing Decision C(2026) 5358 final, case DMA.100193, adopted on July 23, 2026 and signed by Executive Vice-President Henna Virkkunen. The 153-page decision finds Alphabet non-compliant with Article 6(5) of the Digital Markets Act across nine distinct services and carries a 460 million euro penalty for search self-preferencing, part of a combined 890 million euro package that also covers Google Play steering.

The nine are worth listing because the list is the finding. Six are transactional: products intermediation through the paid product advertisement unit, products intermediation through the free products unit, flights, trains, lodging and places. Google Maps is treated as a separate online intermediation service. Three are content services: translation, sports information and games, the last covering interactive units triggered by queries such as play solitaire.

Alphabet's defence rested on a definitional argument. Search engine output, it contended, is not limited to links but includes direct answers, and Article 2(23) deliberately chose a broad definition of output. Multiple sourcing methods, whether crawling, licensed feeds, data feeds or direct uploads, are all valid ways of producing that output.

The Commission applied a three-limb test instead. First, does Alphabet offer a service distinct from or additional to Google Search, ranked alongside results. Second, do third parties offer similar services in the same results. Third, does Alphabet treat its own service more favourably.

Two criteria did the work of separating a search result from a distinct service. The functional threshold captures units whose capabilities enable filtering, interaction and comparison of competing offers on commercial parameters, or functionality enabling actions performed outside search entirely, such as playing a game or translating text. The sourcing route captures units assembled not by searching all websites but from a pre-selected pool, which the decision holds fall outside the online search engine definition regardless of functionality. The sports unit is the clearest illustration: its data comes predominantly from licensed sources including Kicker.de, lequipe.fr and flashscore.es, while those same sites are separately indexed and ranked conventionally in the results beneath.

On similarity, the Commission rejected Alphabet's equation of similarity with demand substitutability, holding that similarity is assessed on the tasks and functionalities performed and on the purpose of the service, not through market-definition concepts used in merger review. On favourable treatment, it weighed frequency of display, volume of information shown, display features such as pictures and featured units, and the engagement possibilities offered to users. One finding has broad implications for platform partnership programmes: where third-party providers are required to become Alphabet business partners to receive equal prominence, that is not equal treatment.

The lodging analysis reads like a product teardown. Comparison sites appear under a tab later renamed Place Sites; individual hotels and rentals appear under a lodging tab; query shortcut chips offer a guest selector accepting up to six guests, a calendar for check-in and check-out dates and attribute filters such as top-rated or city centre; an interactive map carries filterable pins; and up to four accommodation offers show prices assembled from crawled, licensed, user-generated and business-supplied data. The products advertisement unit gets similar treatment: only comparison-shopping services can bid, merchants remunerate the comparison-shopping service of their choice, that service decides which offers to bid, billing runs on cost per click to the comparison-shopping service, and Google Shopping competes in the same auctions.

Procedurally, the decision documents four information requests during 2024, ten recorded Commission meetings between April 19, 2024 and July 25, 2025, and data room access between March 24 and April 8, 2025. Correspondence with the United Kingdom's Competition and Markets Authority and the United States Department of Justice was withheld under Article 8(4) of Implementing Regulation (EU) 2023/814.

The breach is dated from March 7, 2024, six months after Alphabet's September 5, 2023 gatekeeper designation, and ran roughly 28 months at adoption. Alphabet filed compliance reports on March 7, 2024, March 6, 2025 and March 6, 2026, the second and third maintaining that the original measures complied.

Two passages carry the most weight for anyone reading this as precedent. Recital 680 holds that the Commission is under no duty to investigate the impact, effect or harm arising from non-compliance when determining gravity, because Article 30(4) lists gravity, duration and recurrence, not effects. Article 8(1) places the duty on the gatekeeper to ensure and demonstrate that its compliance measures are effective. Between them, the regulator need not prove damage and the regulated party must prove the fix worked.

The 460 million euro penalty represents roughly 0.13 percent of Alphabet's annual turnover, against a statutory ceiling of 10 percent, or approximately 35.6 billion euros on 2025 group turnover of 402.836 billion dollars. Payment falls due three months from notification, with interest at the European Central Bank main refinancing rate plus 3.5 percentage points after the due date. An action under Article 263 does not suspend that date.

Compliance is due 60 calendar days from notification, which from the adoption date points to September 21, 2026. Recital 713 offers two routes: extend equal prominence to similar third-party services, or remove the prominence given to Alphabet's own distinct services altogether. Either path must include measures preventing recurrence. Periodic penalty payments of up to 5 percent of average daily worldwide turnover per day are available from the deadline until compliance, though they require a further decision and do not activate automatically.

AI Overviews are absent from the decision. Commission proposals on how the principles apply to them remain under discussion with no published timeline. The reasoning assembled here, particularly the sourcing-route criterion and the functional threshold, provides a template that would transfer directly. Google has indicated it is considering an appeal and argued in September 2025 that European compliance had already cost its business 114 billion euros.

Two more European rulings on what a gatekeeper owes

Luxembourg and London produced complementary lessons about the limits of platform regulation inside the same forty-eight hours.

On September 2, the General Court of the European Union dismissed Opera Norway's challenge in Case T-357/24, confirming that Microsoft Edge is not required to display a browser choice screen under the Digital Markets Act. The decisive number is from December 2022: Edge accounted for 5.8 percent of European web page views, against 59 percent for Chrome and 22 percent for Safari. The case turned on Article 3(1)(b), which defines important gateway status, Article 3(5), the mechanism for rebutting quantitative thresholds, and Article 6(3), which imposes the choice screen obligation. The court held that comparing Edge to its rivals was permissible, because comparison between services is allowed when the question is whether a service functions as a gateway concretely and specifically.

The result is an asymmetry with no obvious principle behind it beyond arithmetic. Microsoft is a designated gatekeeper. Windows is a designated core platform service. Edge, shipped with Windows, is the only browser from a designated gatekeeper that need not offer users a choice of alternatives, because it is not popular enough to qualify as a gateway. Regulation keyed to market share rewards the browser that lost.

In London, a different kind of limit. The Competition Appeal Tribunal, chaired by Mrs Justice Bacon, will hold a settlement approval hearing on September 15, 2026 in case 1673/7/7/24, in which Professor Barry Rodger represents roughly 2,200 United Kingdom app developers against Alphabet and Google. PPC Land set out the terms of the 260 million pound settlement on September 2. Google pays without admitting liability. Of the total, 160 million pounds is earmarked for eligible developers and 100 million pounds for third-party funder costs, insurance and legal fees, which is to say that 38.5 percent of the fund goes somewhere other than the class.

The tribunal must decide whether those terms are just and reasonable, and it controls how much of the proceeds reach the funder. Represented persons have until 4pm on September 10 to file written submissions. The hearing is listed for one day with a reserve day. The original trial, three months long, had been set to begin on September 28.

The class economics explain why a settlement was attractive to both sides. Between 70 and 77 percent of the developers had potential losses below 10,000 pounds each, while the aggregate claim was valued between 425 million pounds and 1.036 billion pounds with compound interest. Against the top of that range, 160 million pounds distributed to the class is a modest recovery. And, as with the Virginia ad tech remedies a day earlier, the settlement is a compensation event rather than a structural one. Google's commission rates and distribution terms remain exactly as they were.

A copyright suit becomes an antitrust case about robots.txt

The most consequential court filing of the week was a defensive one, and it reframes a copyright dispute as a question about who is allowed to read the web.

SerpApi filed a 63-page answer and counterclaim on August 28, 2026 in the Southern District of New York, case number 25-cv-8736, before Judge Paul A. Engelmayer. PPC Land reported the filing on September 2. Reddit had sued SerpApi alongside Oxylabs, AWMProxy and Perplexity AI over the scraping of its content. The response does not merely deny; it attacks the arrangement that made the content exclusive in the first place.

That arrangement is the expanded Google partnership Reddit announced on February 22, 2024, the same day it filed its S-1, reported at 60 million dollars a year and granting Google crawler access on terms nobody else received. From July 1, 2024, the counterclaim states, Reddit began serving a requester-specific robots.txt: Googlebot permitted, Bing, DuckDuckGo, Mojeek and Qwant disallowed. A technical test by the consultancy Merj on July 4, 2024 confirmed that different files were served to different crawlers.

The timing argument is the sharpest part of the filing. Between the July 2024 exclusion and January 2025, when Google deployed SearchGuard, no technological protection measure stood between the content and anyone who wanted it. SerpApi reads that six-month gap as evidence that the exclusion was commercial rather than protective, and asks for a declaration that no cognizable claim under the Digital Millennium Copyright Act exists for the period before January 2025. Four declaratory counts accompany the antitrust claims.

The antitrust theory rests on a market definition worth reading twice: large-scale, authentic human conversational data in online threaded discussion forums. Within that market, SerpApi alleges monopolisation and attempted monopolisation under Section 2 of the Sherman Act, and seeks treble damages, injunctive relief and a declaration against requester-specific robots.txt directives. That last request has implications far beyond either party. Robots.txt has always been a convention rather than a legal instrument, honoured because honouring it was cheaper than the alternative. A court ruling on whether it may be used to discriminate between requesters would convert an informal protocol into a regulated one.

The licensing argument runs alongside. SerpApi points to Reddit's own user agreement, privacy policy and public content policy, which establish that users retain ownership of what they post while granting Reddit sublicensing rights, and notes that users contribute 99.99 percent of the platform's content. Where its customers already hold authorisation through OpenAI and Google licences, the company argues it supplies structure rather than unauthorised access.

The supporting numbers describe the scale of what is being fought over: 121.4 million daily active users globally in the fourth quarter of 2025, more than 500 million weekly reach, over 25 billion posts and comments, and 762 million dollars of Reddit advertising revenue in the second quarter of 2026. SerpApi reports more than 1.5 million activated user accounts. One detail carries a whiff of discovery to come: Reddit has not produced the Google partnership agreement despite relying on it, and SerpApi has attached it by reference, asserting it does not authorise Google to enforce Reddit's copyrights or its users' copyrights. Another sits oddly on its own: the Internet Archive, promised continued crawling access, was cut off in August 2025. A case management conference is set for October 1, 2026.

Set that filing beside the Virginia ruling and the September 3 SISTRIX report, and a single question emerges from three unrelated proceedings. Who is permitted to observe a commercial system from outside it, and on what terms.

Google answers that question for its own API

The week's last significant document supplies one platform's answer, and it is narrower than the one that existed on Sunday.

Google published updated Google Ads Developer Policies on August 31, 2026, reported by PPC Land on September 5. The update formally prohibits programmatic proxies and requires every integration to connect through its own dedicated Google Cloud project. What was previously called the Google Ads API Policy has been renamed, and the renaming is not cosmetic: the scope now explicitly covers Model Context Protocol servers alongside the traditional API and Ads scripts.

The definition is unusually broad. A programmatic proxy, on Google's Help Center wording, is a third-party hosted interface, secondary API, wrapper service, Model Context Protocol server, proxy endpoint or similar service that solely replicates, wraps or re-exposes Google Ads programmatic capabilities as an intermediate layer. That sentence captures a wide range of existing architectures: any middle-tier service built to simplify authentication for end clients, centralise billing, or expose a stripped-down set of functions through a custom interface. Naming MCP servers marks the first time Google has folded that infrastructure into its programmatic access rules, tying the newest integration pattern in advertising technology to the oldest one, the plain HTTP wrapper.

The policy text names the prohibited practices directly. Developers may not embed credentials in middleware that obfuscate the origin of an automated action, nor operate shared developer proxy services routing automated requests from multiple independent businesses to bypass individual access. Granting end users headless programmatic access to modify accounts without Google review or per-entity authentication is prohibited, which means users of a compliant tool will need to sign in manually to make account changes rather than have a script authenticate on their behalf.

Two exceptions survive. A developer automating its own agency's accounts through the API is unaffected. So is an open-source tool that a user downloads and connects with credentials the user controls, rather than routing traffic through a hosted service the developer operates.

Google's stated rationale is security and throughput. Unaudited proxies can give unauthorised actors account access and introduce cross-tenant data leaks, where a flaw in one shared intermediary exposes one customer's traffic to another. High-volume traffic routed through a single proxy can throttle throughput, cause latency across all users, and require enforcement against every client behind that proxy when mitigating a localised denial of service, including clients whose own usage was compliant.

A formal route stays open for services that genuinely need an intermediary layer. Developer Secondary Interface Review rests on three criteria: adherence to Required Minimum Functionality, unbiased cross-channel measurement frameworks intended to prevent anti-competitive self-preferencing, and immutable transaction logs mapping every automated action to a verified end-advertiser account, plus third-party security assessment to a SOC 2 Type II standard. Those logs must be available to Google Ads Compliance on request.

Enforcement is already running. The Ads API Compliance team is reviewing existing integrations, and Google has asked developers to keep contact addresses current in the API Center so that outreach reaches them, noting that legitimate communications come from the google.com domain. Non-compliance can produce fees, downgraded access levels, additional quota limits or termination.

The change extends a pattern rather than starting one. Customer Match uploads were forced onto the Data Manager API from April 1, 2026. New offline conversion imports through the Ads API stopped for inactive developer tokens from June 15, 2026. Granular reporting data retention was cut to 37 months from June 2026, down from an eleven-year window set less than eighteen months earlier. A passkey became mandatory for generating new OAuth refresh tokens on August 5, 2026. Version 4.0.0 of Google's own Developer Assistant, released on August 25, was explicitly incompatible with all prior installations, converting a local workspace project into a globally installed plugin bundle. The API moved to a monthly release cadence in January 2026, up from three major releases a year, so the developers absorbing the proxy ban have also been absorbing a steady stream of unrelated breaking changes.

The timing sits awkwardly beside the agentic buying story from the same week. Agencies spent the week installing meters on agents that connect to advertising platforms through exactly the kind of intermediary layer this policy now restricts on one of the largest of those platforms. An MCP server that a single agency builds for its own internal account management appears to survive under the internal-use exception. A hosted multi-tenant server serving many independent businesses through shared credentials does not. The distinction is not about the protocol. Google's own open-source MCP server for the Ads API, released in October 2025 with read-only access, remains available. The restriction targets third parties interposing themselves between end users and that direct connection.

Which is, expressed as a product policy rather than a court filing, the same proposition SerpApi is contesting in New York and the same one the rank trackers have been paying for since June: access to a platform's systems is being narrowed to parties the platform can identify, audit and rate-limit individually. The commercial arguments differ. The direction does not.

Nielsen closed the week by moving in the opposite direction on a smaller stage, cutting its local television tuning threshold by 80 percent and ending what it describes as a 24 percent blind spot in local measurement, six days after the seven national currency changes that opened the week.

AI agents started buying media, so agencies started metering them

The interesting number in agentic advertising this week was not a media budget. It was 1.5 million tokens, burned by a single member of staff in a single day, at an agency that only noticed because it had built the meter to notice with.

Digiday reported on September 4 that media agencies have begun building audit tools and consumption caps around AI agents rather than waiting for vendors to supply them. The problem is specific and unglamorous: an agent that reasons in a loop consumes tokens, tokens cost money, and nothing in the standard agency operating model was designed to watch that meter in real time.

Jonathan Whiteside, global executive vice president of technology at Dept, put the failure mode plainly, saying it can get out of control very quickly, and citing that 1.5 million token day. Dept's answer is an internal gateway that routes each request to a model chosen for the commercial and legal requirements of the task rather than for whatever the user happened to open. The routing layer is the cost control.

PMG built something narrower. Dillon Larberg, its consulting and strategy director, described a tool imposing daily token caps per user, with an escalation review when someone hits the ceiling. His framing was about matching the vehicle to the journey: he does not drive an 18-wheeler to work, and he does not go cross-country in a hybrid. Model selection, in other words, is a procurement decision made hundreds of times a day by people who have never thought of it that way. Gartner analyst Nicole Greene identified exactly that as the dominant cost driver, and the firm's April 2026 survey of 1,300 senior marketers found 56 percent of companies deploying the technology without a clear usage policy. Gartner expects 60 percent of organisations using it to face cost overruns attributable to inadequate tracking.

Daniel Gilbert, founder and chief executive of the independent agency Brainlabs, was blunter, calling monitoring existential rather than optional. Brainlabs operates a tiered token allowance for staff.

The reason these tools appeared now rather than a year ago is that the agents started transacting. Klaudia Smykowska, group director at Rise, part of the Quad agency group, has been running media buying agents for a supermarket client since June 2026, using the audit log PubMatic built into its platform to watch for drift. Harry Tong, PubMatic's director of sales engineering, described what the log captures: agent changes time-stamped and stored with all the details.

That audit log is not a feature invented for this news cycle. PubMatic published a five-step guardrail framework on August 5, 2026 covering platform-level boundaries, pre-campaign business rules, pre-approved inventory and creative, authenticated approval workflows that halt an agent and demand human sign-off through a one-time passcode when a decision exceeds its authority, and the audit trail with drift detection. The passcode produces a cryptographically verified approval record, which matters for a dispute arriving months later. At the time the company said the framework covered more than 100,000 websites, apps and streaming services, with over 80 autonomous campaigns and more than 4,000 direct deals executed.

George Forge, senior vice president of client technology and product development at Rise, supplied the most honest line in the reporting: calculating what any of this actually costs is, at this point, a very rough math equation.

What five hundred percent looks like when a bot books it

The campaign giving the cost question its urgency ran in the second quarter of 2026 and was disclosed on September 3.

Rouge Care, a direct-to-consumer red-light therapy brand, let two AI agents plan and buy a connected television campaign through PubMatic's AgenticOS, working with the agency Klever Programmatic. One agent handled planning. A second handled audience discovery and the buying decisions, operating on the sell side. Spend was 25,000 dollars. Attributable sales exceeded 125,000 dollars. Samuel Bonneau, senior copywriter at Rouge Care, said the campaign delivered 500 percent against a goal of 250.

A single-brand result at that budget proves the plumbing works end to end. It does not establish an average, and nobody involved claimed it did. What it establishes is that the buying agent is no longer a demonstration. AgenticOS launched on January 5, 2026 with live campaigns already running, built in three layers: infrastructure handling tens of millions of auctions per second on accelerated computing, an application layer of coordinated agents covering planning, forecasting, pacing, yield management, troubleshooting and measurement, and a transaction layer wired into PubMatic's Activate buying platform. Launch claims included an 87 percent reduction in campaign setup time, a 70 percent improvement in issue resolution and five times faster decisioning, against a base of 78 trillion impressions processed quarterly.

Set the Rouge Care result beside the token bill and the commercial problem becomes clear. A 25,000 dollar campaign returning 125,000 dollars can absorb a great deal of inference cost. A campaign at a tenth of that size, or one returning 1.2 times its spend rather than five, cannot. Nobody yet knows where the crossover sits.

While agencies built meters, the largest model vendor made the meter run slower. On September 1, Google shipped an agentic processing mode for video understanding in which Gemini selects which segments of a video to analyse rather than ingesting frames at a fixed rate. The reductions are large and, unusually, benchmarked. On the Minerva benchmark, consumption fell 58.4 percent, from 80,900 tokens to 33,600. On the 1H-VideoQA test it fell 88.0 percent, from 397,600 to 47,700. On LVBench it fell 88.0 percent, from 300,300 to 36,000. Accuracy rose rather than fell in each case: Minerva from 73.7 to 79.0 percent, LVBench from 85.1 to 88.6, 1H-VideoQA from 87.5 to 88.5. Google claims cost reductions of up to 66 percent without publishing the cost tables that would let anyone check the arithmetic.

Two movements, then, running against each other. Per-task inference costs fall because the models are learning to look at less. Aggregate consumption rises because agents run more tasks, more often, without a human deciding each time whether the question is worth asking. Which curve wins is the question the agency dashboards were built to answer. KPMG research found 49 percent narrowed, delayed or paused AI agent deployments when operating costs outran the value delivered, and agencies have already started folding token costs into principal media arrangements. An agency that meters tokens internally can price them. An agency that does not can only bury them. The audit log, in that reading, is less a safety feature than a billing instrument waiting for a rate card.

Three privacy regimes moved in three directions

Privacy law rarely moves in one direction at once, and this week it moved in three.

Sacramento removed exposure. PPC Land reported on September 3 that California's legislature had passed Senate Bill 690, which eliminates the private right of action under the California Invasion of Privacy Act for pen register and trap-and-trace claims brought over websites, apps and online applications. Introduced on February 21, 2025 by Senator Anna Caballero, the bill amends Penal Code Section 637.2 by adding a new subdivision (d).

The vote record is the striking part. Senate Public Safety cleared it 6-0 on April 29, 2025; Senate Appropriations 6-0 on May 23, 2025; the full Senate 35-0 on June 3, 2025. On the Assembly side, Public Safety passed it 9-0 on July 1, 2025, Privacy and Consumer Protection 14-0 on July 1, 2026, and Appropriations 15-0 on August 13, 2026. The Assembly floor vote on August 28, 2026 was 66-0, and Senate concurrence the same day was 40-0. Across eight recorded votes spanning eighteen months, not one legislator voted against.

The bill was enrolled on August 31, 2026. The governor has until September 30, 2026 to act. If signed, it becomes operative on January 1, 2027 and reaches back to claims filed within the prior two years, which means a substantial share of pending litigation would be extinguished rather than merely prevented.

What survives matters as much as what goes. CIPA Section 631 covering direct interception, Section 632 covering recording of confidential communications and Section 632.7 all remain, along with their private rights of action and the 5,000-dollar-per-violation or treble-damages formula. What disappears is Section 638.51, the provision plaintiffs' firms repurposed from telephone-era pen register law to cover browser fingerprinting, tracking pixels and analytics scripts.

The volume explains the unanimity. Filings under Section 638.51 rose from roughly 600 in February 2025 to about 4,000by August 2026, a sevenfold increase in eighteen months. Named defendants in recent cases include Ace Hardware and LinkedIn in April 2026, the latter over an alleged covert scan of more than 6,000 Chrome extensions, Perplexity AI in March 2026, OpenAI in May 2026 and the meeting notetaker Granola in July 2026. The category has expanded well beyond retail websites into artificial intelligence products.

The two largest CIPA outcomes stay untouched because both rest on surviving sections. A federal jury found in August 2025 that Meta had violated CIPA by collecting health data from users of the Flo application, and a California court signed a 50 million dollar judgment against Meta in March 2026 over Facebook data sharing. Felipe Mate, partner at Trackstars, observed that CIPA's private action has changed how websites operate more than any other law. That claim is defensible: consent management platform deployment, tag governance programmes and the entire practice of pre-consent script blocking in the United States grew primarily out of Section 638.51 exposure rather than out of the California Consumer Privacy Act.

Strasbourg added exposure. The Council of Europe's Committee of Convention 108 has circulated draft guidelines, document T-PD(2025)3rev4 dated August 26, 2026, running more than thirty pages and applying data protection principles to generative systems and autonomous agents. The Council is a 46-member intergovernmental body whose Convention 108 entered into force on January 1, 1985, sitting alongside rather than inside the GDPR and the EU AI Act. The Bureau reviews the draft in Paris on September 16 and 17, 2026, with possible adoption at the 51st Plenary on November 17 to 19, 2026.

Several provisions would land directly on advertising technology if adopted. Controllers would be required to test complete deployed systems rather than only models, before launch and after significant changes, covering prompt injection, unauthorised data extraction and cascading agent errors. They would keep operational records sufficient to reconstruct an action sequence. They would keep read permissions separate from create, modify and delete permissions for agents. They would ensure effective human review points before actions with significant or irreversible effects. And persistent memory would be disabled by default unless continuity serves a specified purpose.

That last provision has commercial teeth. A buying agent retaining context across campaigns is more useful than one starting cold; a default-off memory requirement inverts the design assumption most vendors have been building toward. The permission-separation requirement points the same way, at agents that can read a campaign but not modify it without an explicit grant. One product shipped during the week matches the shape exactly: Dreamdata added a Model Context Protocol server letting language models read its business-to-business attribution data without the ability to change the underlying numbers.

On memorisation the draft resists a common industry argument. The absence of personal data from initial input or final output does not, in itself, establish that no processing of personal data takes place within intermediate operations. The supporting research is uncomfortable: sequences appearing ten times in training data were reproduced roughly a thousand times more frequently than sequences appearing once, and training data extraction was estimated to cost approximately 200 dollars. The draft arrives into a crowded field. The European Data Protection Board adopted guidelines on web scraping for generative artificial intelligence on July 7, 2026, rejecting consent as a general basis for scraping, and EU AI Act Article 50 transparency obligations became legally applicable on August 2, 2026.

Canberra proposed to make a pixel a transaction. Australia's draft privacy reforms would require consent before personal information is traded, with cookies and pixels in programmatic advertising named explicitly, treat location data at 500-metre precision as sensitive, and hand platforms above 500 million dollars an erasure duty. Submissions close on September 18, 2026. A parallel proposal sets out a dual-track model letting organisations choose between notice-and-consent and an outcomes-based test.

Spain supplied the enforcement end of the same subject. Yoti halted its Digital ID application in Spain on September 10 after a 950,000 euro fine issued on March 10, 2026 and split across three findings. Existing Spanish users keep the application but lose updates, and only they can delete their accounts. A company selling age assurance withdrew from a market rather than change how its biometric processing works, in the year that age assurance became a legal duty across the United Kingdom, the European Union, Australia and several United States states.

The independent bet gets smaller by 575 people

Ten years ago The Trade Desk went public with a little over 400 employees and 202 million dollars of annual revenue. On Thursday evening its chief executive sent an email telling the company that by the following morning it would be smaller by roughly 15 percent.

PPC Land reported on September 4 that The Trade Desk would cut about 15 percent of its global workforce effective that day, following an internal email from co-founder and chief executive Jeff Green. Against the 3,843 full-time employees the company reported across 21 countries as of December 31, 2025, the reduction lands at roughly 575 positions. Adweek, covering the same announcement, put the figure at approximately 585 against a headcount of about 3,900.

Green's language was the standard vocabulary of restructuring: greater agility, focus, ownership and speed, an opportunity to reinvest in client outcomes, and the next phase of growth. He also acknowledged, in a line worth more than the rest, that this has not been a common practice at the company. That is accurate. The Trade Desk dismissed 39 employees in December 2025, roughly 1 percent of its workforce, and reorganised in December 2024 in what it described as the largest structural change in its history, breaking engineering into about 100 scrum teams. Neither was a layoff on this scale.

The financial context sits in three numbers. Second-quarter 2026 revenue reached 715 million dollars, up 3 percent year over year, missing analyst consensus by 37.6 million and marking the slowest growth rate since the opening months of the pandemic. Third-quarter guidance of at least 650 million implies a 12 percent decline against the prior year, which would be the company's first annual revenue contraction. Full-year 2025 revenue had been 2.896 billion dollars, up 18 percent. The gap between an 18 percent year and a quarter guided to minus 12 is not a trend, it is a discontinuity.

Markets responded accordingly. Following the August 6 earnings call, the stock fell 24.22 percent to close at 13.39 dollars. The balance sheet remains unstressed: 1.122 billion dollars in cash and equivalents plus 362.354 million in short-term investments at the end of the second quarter, an undrawn 750 million dollar credit facility expanded in April 2026 and maturing in 2031, and no debt.

Underneath the revenue line, the operating detail is more interesting than the headline. Customer retention stayed above 95 percent in the second quarter, which means the problem is not churn. Consumer packaged goods and automotive together account for roughly a quarter of platform spend, and both categories have been under pressure through 2026. The company continues to describe itself as holding about 1 percent of its total addressable market, a framing that reads differently when the addressable market is being redefined by agentic buying tools and retail media networks that did not exist when the framing was written.

The executive record over the preceding fourteen months is its own story. Nate Olmstead became chief financial officer in July 2026, the fourth person to hold the role in fourteen months, following Laura Schenkein's departure, Alex Kayyal's five-month tenure and an interim period under Tahnil Davis. Three senior executives departed in April 2026 including the chief marketing officer, and the company subsequently appointed a new chief marketing officer, chief commercial officer, chief business development officer, chief operating officer and vice president of client strategy and growth. A company that replaces its finance chief four times in a little over a year is not merely reporting slower growth; it is arguing internally about what to do next.

Demand-side relationships added their own drag. In February 2026 two large buying groups exited the OpenPath supply programme, and in March one advised clients to stop transacting on it, a dispute resolved on June 12, 2026.

What the cuts leave unresolved is the product roadmap. The May 2026 Kokai release added artificial intelligence optimisation, connected television pause advertising, Sincera supply signals and a Deal Desk. Audience Unlimited was moving toward open beta. A measurement framework remains in alpha with no partners named publicly. The company announced Zuma at the end of August 2026 with a claimed 32 percent average improvement in cost per acquisition across its Koa agents, and during the week added LoopMe brand lift studies at a 1 million impression threshold where rival vendors demand three to four million. Removing 15 percent of engineering, sales and support capacity while carrying that much unfinished work is a bet that the remaining structure, described in Green's email as smaller pods and smaller scrums, ships faster than the larger one did.

Geographically, the reduction falls across a workforce that was roughly 63 percent North American, 20 percent EMEA and 17 percent APAC. No breakdown by region or function has been published.

The supply side went the other way

The uncomfortable symmetry of the week is that connected television supply numbers moved in the opposite direction to the buy side.

Magnite said on September 3 that it had added more than 5,800 advertisers new to live sports inventory and that global live sports advertising spend on its platform rose 56 percent year over year between January and July 2026. Its Live Scheduler tool, launched on November 18, 2025, is now used by 37 media owners globally and has scheduled more than 4,000 live events. Across 289 days that averages roughly fourteen events a day, or about 108 per media owner. Named events include the FIFA World Cup, NFL Monday Night Football, NHL playoff games and the Academy Awards.

Three capabilities carry the growth. Precision Activation creates sport-specific, daypart-aligned high-priority line items with pacing algorithms attached. Live Stream Acceleration distributes traffic automatically when viewer numbers surge, which is the technical problem live advertising has always had and the reason so much live inventory historically went unsold programmatically. The third is buying access through conventional programmatic channels plus the company's Buyer Agent, launched on April 27, 2026 with Disney, Spectrum Reach, Kepler and MiQ.

The company's second-quarter connected television contribution excluding traffic acquisition costs reached 97.1 million dollars, up 36 percent year over year, representing 51 percent of total contribution. March Madness revenue in 2026 grew more than 80 percent. Live sports revenue is not disclosed separately, no third-party auditor is named, no absolute dollar figures accompany the percentage growth, and the 5,800 figure counts advertisers new to the category rather than new to the platform.

Two adjacent measurements complicate the enthusiasm. DataBeat reported on August 4, 2026 that United States programmatic cost per thousand impressions rose 51 percent year over year while connected television rates fell 12.3 percent against 2025 levels, and found a 13.4 percent cost per thousand disadvantage for agent-mediated buying against conventional programmatic. Gracenote reported on September 1 that 26 percent of sports fans could not locate the games they wanted to watch, rising to 36 percent among adults aged 18 to 34, with 38 percent now preferring highlights to full games. Inventory is easier to buy than it has ever been. Finding the game remains a problem for roughly a quarter of the audience.

New inventory arrived from an unexpected direction. AdExchanger's Tuesday roundup reported that Sony launched Live TV on PS5, a free ad-supported service carrying more than 100 channels. The asset is the install base: roughly 125 million logged-in users, arriving with authenticated identity, a payment relationship and a device sitting on the main television in the house. Programming spans sports, news, action, horror and community channels, with two new anime channels for Crunchyroll, which Sony owns. Ad serving runs through Publica and PubMatic. The motivation is margin: console manufacturing costs have risen, and chief executive Hiroki Totoki has said that marketing to new customers ranks below monetising the existing base.

Console gamers have historically been among the most advertising-averse segments in media, and they are underrepresented in mainstream connected television panels, which means buyers will evaluate this inventory without much comparable measurement history. Sony published no launch markets, no timing, no CPM guidance, pod structure, frequency rules or measurement partners, and no statement of whether the inventory clears programmatically, through direct deals, or both. Publica is owned by Integral Ad Science and handles ad pod decisioning, unified auctions and server-side ad insertion; its Samsung partnership was renewed as an exclusive global agreement on September 30, 2025. PubMatic reported revenue up 11 percent after three quarters of decline. The free ad-supported category has been growing quickly, with channel counts up 42 percent as programmers moved beyond back-catalogue material and European penetration reaching 27 percent.

Radio joined the same architecture. PPC Land reported on August 31 that Triton Digital connected its programmatic marketplace to RCS playout software, opening broadcast inventory at stations across 120 countries to automated buyers. RCS supplies the software deciding second by second what a station transmits, including where the commercial breaks fall. The commercial logic is a gap between consumption and money: audio accounts for roughly 31 percent of consumer media time against approximately 9 percent of advertising budgets. iHeartMedia is furthest along in pursuing it, targeting roughly 200 million dollars of programmatic revenue in 2026 against 135 million in 2025.

That last detail carries a disclosure of its own. Triton Digital and RCS are both iHeartMedia subsidiaries, and the announcement describes the arrangement as a partnership without noting the shared parent, which also operates more than 860 stations competing for the same demand. The build-out has been methodical: Viant became the first buyer of iHeartMedia over-the-air inventory in December 2025, StackAdapt integrated broadcast radio in November 2025, and Amazon DSP added Triton audio supply across more than 80 countries. What has not been published is the part a buyer would need: no station count, no launch markets, no revenue split, no list of which demand-side platforms can reach the inventory, no floor prices or auction structure, and no statement of whether this supports real-time bidding against individual breaks or scheduled allocations, which are different products wearing the same word.

Four demand-side platforms already held 85 percent of global programmatic spend in the first quarter of 2026. A new inventory class arriving inside that concentration, priced by mechanics nobody has described, is the situation the Amazon complaint documents at the far end of its life cycle.

Two views of the same catalogue, 21.6 percent apart

An independent study published on September 1 measured something nobody had quantified: whether the products Google surfaces in AI Mode cost the same as the products it surfaces in conventional search.

Hugo Huijer, founder of Productrise, collected more than two million product listings across over 100,000 results pages and AI Mode responses in the United States and United Kingdom between August 9 and 31, 2026, matching products using Google's stable product identifier and comparing the same query on the same calendar day. The finding: identical products in AI Mode averaged 21.6 percent more expensive than in traditional search. The median AI Mode product carried a price of 149 dollars against 100 dollars in traditional results.

The distribution matters more than the average. Prices differed on 38.1 percent of matched products. When they did differ, AI Mode was the pricier surface 68.4 percent of the time, with a median difference of 22.2 percent. The lead seller differed on 49.6 percent of matched products, meaning that on roughly half of overlapping items the two surfaces route the shopper to a different merchant entirely.

The overlap figures are the quiet headline. Only 1.28 percent of traditional search products appeared in AI Mode at all. An average query returned 27.8 traditional products, 3.9 AI Mode products, and 0.94 in common. AI Mode showed 12.3 percent of all tracked products. Two surfaces on the same domain, answering the same query on the same day, are drawing from what amounts to different catalogues.

Methodological limits are stated openly. The traditional-search side was limited to the popular products carousel. The measurement captured the first or lead offer visible rather than the cheapest available. Brodie Clark made the obvious counterpoint: the retailer with the lowest price gets the click in the end. Productrise sells tracking products addressing exactly the visibility problem the study describes, which is worth holding in view. Google was not quoted and has published no response.

Context sharpens the finding. Shopping advertisements launched in AI Mode on February 11, 2026 against a base of more than 75 million daily users, and the surface reached one billion monthly active users on May 19, 2026. The United States Federal Trade Commission approved an enforcement policy statement on undisclosed personalised pricing on August 19, 2026. A price gap between surfaces is not personalised pricing, and nothing in the study suggests it is. But a measurable, systematic 21.6 percent differential between two views of the same catalogue is the kind of finding regulators read closely.

Two measurement companies moved into the same territory during the week. Comscore expanded its AI Intelligence product on September 3 to track sponsored placements inside chat responses, finding that hotel-related prompts carrying sponsored advertisements rose from 6 percent in March 2026 to 24 percent in May 2026. The measurement uses real user panel activity rather than synthetic prompts, capturing prompt-response pairs, cited domains, conversation identifiers and event timestamps. Adjacent estimates diverge widely: Similarweb data from August 2026 put 26 percent of all ChatGPT responses as carrying advertisements across free and mid tiers, while Adthena measured 4.47 percent of United States queries served advertisements between March and May 2026. Those are not reconcilable numbers, and the gap between them is itself the argument for panel-based measurement.

NIQ and Similarweb announced a partnership on September 2 for agentic commerce measurement, with an initial product due in the fourth quarter of 2026. Five areas are named: consumer intent, agentic shelf visibility, product content readiness, AI-driven traffic and AI-driven conversion, across ChatGPT, Gemini, Google AI Mode, Perplexity and Claude. Initial categories, markets, pricing, launch sequencing, revenue sharing and exclusivity are all undisclosed.

That arrangement is distinct from the one NIQ announced with OpenAI on August 25, where the most precise language concerned exclusion: the company's proprietary data remains governed and available only through authorised NIQ products, client agreements, permissions and applicable security controls. The asset being fenced is substantial: a 160-petabyte data engine spanning roughly 260 million product items and 10.5 billion product attributes, ingesting 4.3 trillion consumer transaction records weekly, covering more than 90 countries and 7.4 trillion dollars of consumer spending. One arrangement concerns governed transaction data feeding a model. The other concerns measuring what happens on the assistant surfaces. Both were announced within nine days.

One figure recurs across all of it and deserves repeating: only 16 percent of brands systematically track their visibility in artificial intelligence surfaces. A related finding from Contentsquare puts a price on the gap at the other end of the journey, with 23 percent of AI-referred shoppers switching to a competitor after a poor site visit and 21 percent finding AI-supplied details that contradicted the brand's own page.

The gap between a view and a viewer, measured

YouTube changed how it counts on August 24, 2026, making a view register from the very first frame of playback across all formats. This week somebody measured the consequence.

Podtrac published analysis on September 3 finding that engaged views trail counted views by 36 points on podcast episodes, drawn from 84.2 million views across hundreds of video podcasts and channels through the YouTube Analytics and Reporting APIs on August 27 and 28, immediately after the change took effect.

The mechanics are simple and the implication is not. A counted view registers exposure from the opening instant with no minimum duration. An engaged view registers either that the viewer let playback continue past that instant or that they actively clicked to start it. On podcast playlist episodes, engaged views amounted to 64 percent of counted views. On other content from the same channels the figure was 58 percent, a 42-point gap and an eight-point spread between the two categories.

For advertisers buying against podcast video inventory, that is the difference between a denominator and an audience. Podtrac's sample is two days long and drawn from a self-selected set of channels, which limits how far the ratio generalises. It is nonetheless the first published attempt to quantify the effect of a change that altered every view number on the platform simultaneously.

The context keeps moving. IAB Tech Lab published Podcast Technical Measurement Guidelines version 2.3 on July 21, 2026, bringing video podcasts under the same download counting rules as audio. YouTube doubled Partner Program entry thresholds on August 10, 2026, taking qualified watch hours from 4,000 to 8,000 and the Shorts requirement from 10 million to 20 million views, effective February 1, 2027. And Netflix began distributing Martha Stewart podcast content from iHeartPodcasts on August 25 and 26, 2026, in a deal PPC Land covered when "Martha Cooks" launched the same day and "The Martha Stewart Podcast" followed on August 26, with no CPM, no named sales owner, no advertising format specification and no measurement methodology disclosed. Episodes distributed inside Netflix fall outside a framework built for open RSS feed delivery. A metric that inflates uniformly across a platform is survivable. A metric that inflates by 36 points in one content category and 42 in another, on the same channels, is a planning problem.

Two further YouTube findings from the week complete the picture of what a subscriber number means. A platform director told creators that subscribers skip roughly 90 percent of uploads in their feed, with probably all of a channel's videos carrying a click-through rate of about 10 percent or less. A separate German-language FAQ confirmed no relationship between the length of an upload break and changes in view counts, along with no ranking weight for monetisation status, channel location or publish timing.

YouTube starts labelling brand deals creators do not declare

YouTube's disclosure regime changed on September 3, in a community announcement covering four changes at once.

The platform refreshed its paid promotion label, added manual geographic and age controls, rewrote its policy vocabulary around the term branded content, and said it will begin automatically classifying undeclared branded content in newly uploaded videos over the coming months. The original paid promotion declaration dates from October 4, 2016; this is the first structural revision in a decade.

The detection change is the substantive one. YouTube says it may automatically apply the disclosure label on a creator's behalf. Creators who believe a label was applied in error may have the option to certify that a video does not contain branded content, phrasing that leaves availability conditional. Responsibility does not transfer with the label: creators remain ultimately responsible for ensuring disclosures comply with applicable legal and regulatory requirements.

The definition is broad. Branded content is any content influenced by a brand partner in exchange for something of value, including featuring or promoting products or services in exchange for payment, free products or sponsorships, and it applies to videos, descriptions, comments, live streams, Shorts and other surfaces. The disclosure covers the entirety of a video, not just the sponsored segment. Branded content is excluded from the YouTube Kids application. Static title cards are limited to five seconds and end cards must appear within the final thirty seconds. The policy also imports the Google Ads prohibited categories wholesale.

The new geographic and age controls are presented as commercial enablement rather than restriction. Three manual settings appear at the point of declaration: the geographic locale where content displays, an overall minimum viewing age, and a minimum age specific to individual countries. YouTube frames this as making possible deals creators previously declined for regulatory reasons, citing the United Kingdom's high fat, salt and sugar food marketing restrictions implemented on January 5, 2026.

The compliance baseline explains the timing. Advertising Standards Authority research cited in May 2026 found only 57 percent of influencer advertising met disclosure requirements. Enforcement has been picking up: Australia's competition regulator fined PhotobookShop 39,600 dollars in March 2026 for concealing paid influencer partnerships, Sweden's Konsumentverket documented disclosure failures the same month with court penalties reaching 1.5 million kronor per violation, and Croatian trade body guidance from November 2025 cited research finding only 20 percent of consumers recognised influencer content as commercial.

What the announcement does not specify is the machinery. No documentation describes what signals the classifier reads, what false-positive rate is expected, whether back catalogues will be scanned, or how the certification override surfaces. Roughly three million channels are currently in the Partner Program.

A repost, a caption, and a claim that survives

A short procedural ruling on September 2, reported September 4, gives corporate social media teams something specific to think about.

Judge Josephine L. Staton of the United States District Court for the Central District of California denied Tesla's motion to dismiss a willful contributory copyright infringement claim in case 2:26-cv-02409-JLS-JDE. The plaintiff, James Quilici, is representing himself.

The sequence is short. Quilici published a drone video on July 14, 2025. On July 15, Sawyer Merritt uploaded it to X with a caption crediting WC Drone Productions. On July 21 the account belonging to Tesla's chief executive reposted it, and Quilici made contact the same day. He sent a formal infringement notice on November 21, 2025, filed an original complaint on March 6, 2026 on a direct infringement theory, and filed a first amended complaint on May 27, 2026 recasting the claim as contributory infringement under 17 U.S.C. sections 106 and 501.

Three holdings survive the motion. Reposting can constitute material contribution to infringement, under the Ninth Circuit standard that a defendant materially contributes where it assists in distribution by making infringing content fast and easy to locate. Attribution in a caption can establish knowledge of third-party ownership, since the caption naming the original producer put the reposting account on notice. And an executive's personal account can be attributable to a corporation through agency principles.

That third point has the widest reach. Denial of a motion to dismiss establishes only that the pleaded facts, taken as true, state a claim; nothing has been proven and no damages figure appears in the order. But the theory now proceeds to discovery, and it concerns conduct occurring thousands of times a day: an executive account resharing a third party's media with the original creator credited in the text. The pattern is not isolated. Sony Music has pursued Kroger over 392 songs in a 58.8 million dollar claim and DSW over social media posts, both resting on the premise that a brand's organic social output is a distribution channel carrying the same licensing obligations as a paid one.

Beijing tells its companies not to answer Brussels

The regulatory story with the longest tail concerns a retail transaction that has not yet closed.

PPC Land reported on September 3 that China's justice ministry had ordered domestic entities not to cooperate with the European Commission's Foreign Subsidies Regulation investigation into JD.com's 2.5 billion dollar acquisition of Ceconomy, the German parent of MediaMarkt and Saturn. The order was issued on August 19, 2026. A Commission decision is expected on October 2, 2026.

The ministry characterised the probe as improper extraterritorial jurisdiction, described the information demands as extensive and unnecessary, and called them a serious violation of the international rule of law, warning that China would retaliate if the European Union persisted. This is the second such non-cooperation order, following one in May 2026 concerning an investigation into the security screening manufacturer Nuctech. China introduced its extraterritorial jurisdiction countermeasures framework in April 2026.

The transaction is well advanced. JD.com announced its offer on July 31, 2025 at 4.60 euros per share, a 42.6 percent premium to the three-month weighted average, valuing Ceconomy at an enterprise value of 4.0 billion euros against fiscal 2023/24 sales of 22.4 billion euros. Germany's Federal Cartel Office cleared the acquisition on September 18, 2025. By early December 2025 JD.com had secured 60 percent of the company. Ceconomy operates more than 1,000 stores across eleven European countries.

The advertising dimension is not incidental. MediaMarktSaturn launched its first offsite retail media programme with Unlimitail in late September 2025. JD.com launched its Joybuy platform in Germany on August 23, 2025 and has since been competing in Google Shopping auctions across European markets. A Commission decision that blocks or conditions the acquisition would put a European retail media network with a thousand stores behind it into a different ownership structure than its current commercial roadmap assumes.

The cross-border trade backdrop moved in the same direction earlier this year. The European Union ended its 150 euro customs exemption and imposed a 3 euro per-item charge from July 1, 2026, a change that has already reduced Temu advertising spend and pushed SHEIN toward exit from parts of the market. A second European cost regime landed on the same sellers in August: Regulation (EU) 2025/40 applied in full across all 27 member states from August 12, 2026, and PPC Land documented one Dutch tabletop game shop whose annual packaging compliance costs across nine European Union markets run between 2,400 and 4,300 euros against roughly 130 euros of actual recycling fees, a multiple of 18 to 33 times the underlying obligation. Authorised representative fees make up 70 to 90 percent of that burden and are entirely insensitive to volume, so a shop shipping one parcel pays what a shop shipping ten thousand pays. Shipping eligibility and advertising availability are the same variable inside a product feed, which is why a compliance decision surfaces as feed disapprovals rather than as a budget decision.

Australia grows 14 percent, and the composition explains where

Australia published full-year numbers on August 31, distributed more widely on September 1, and they are the clearest read on channel mix available anywhere this week.

The PwC Australia report commissioned by IAB Australia put internet advertising expenditure for the year to June 30, 2026 at 19.8 billion Australian dollars, up 14.0 percent. The June quarter alone reached 5.4 billion dollars, up 16.3 percent, so the market accelerated through the period rather than decaying. The six-year average annual growth rate stands at 13.7 percent, and the fiscal series runs 15.6 billion dollars in FY24 up 9.7 percent, 17.2 billion in FY25 up 10.6 percent, and 19.8 billion in FY26.

Composition tells the more useful story. Search took 8.6 billion dollars, up 13.2 percent, holding 43 percent of the market. Video reached 5.9 billion, up 18.8 percent, at 30 percent. Classifieds took 3.1 billion, up 12.6 percent. Display excluding video managed 2.2 billion, up 6.9 percent, the weakest major format. Audio took 353 million, up 6.6 percent, with podcasting growing 10.1 percent and holding 40 percent of the audio total while streaming audio grew 4.3 percent.

Inside video, social video reached 2.4 billion dollars, up 29.5 percent, and now represents 41 percent of all video spend. Among publisher video specifically, connected television climbed to 60 percent of spend from 51 percent, desktop fell to 25 percent from 37 percent, and mobile rose to 15 percent from 12 percent. A nine-point transfer from desktop to connected television inside a single fiscal year is a faster reallocation than most markets have recorded, and it maps directly onto the supply-side growth Magnite reported in live sports.

By industry, the top five verticals took 56 percent of general display spend, led by retail at 17.8 percent and automotive at 14.2 percent, with fast-moving consumer goods slipping to 6.1 percent from 6.4. In the June quarter, government entered the top five for the first time at 6.3 percent, displacing FMCG and health and beauty. Gai Le Roy, chief executive of IAB Australia, attributed the pattern to new spend arriving from outside the usual base, much of it always-on rather than tied to traditional campaign cycles. The 19.8 billion fiscal figure raises the stakes of the privacy consultation closing on September 18.

Where the traffic went, and where the entry-level jobs went

Two findings published during the week describe the same displacement from opposite ends.

The first is about traffic. OneLittleWeb measured 9,531 tools through April 2026 and found that Chegg lost 61 percent of its web visits while chatbot visits rose 90.82 percent year over year, from 45.1 billion to 86.0 billion. Quizlet fell 29 percent, QuillBot 25 percent and Grammarly 15 percent in the same study. Translation tools lost 22.5 percent overall.

The second is about people. A peer-reviewed paper by Nolan Lovett, published in Human Resource Development Review on July 26, 2026, cites Stanford Digital Economy Lab research finding that employment for workers aged 22 to 25 in artificial-intelligence-exposed occupations fell 16 percent between October 2022 and September 2025, while employment for workers aged 35 to 49 in the same occupations grew more than 8 percent. The underlying analysis covers more than 25 million United States workers. A separate National Bureau of Economic Research analysis of 58 million LinkedIn profiles confirmed systematic removal of developmental roles.

Lovett's framework distinguishes internalised mastery, built through sustained practice, from distributed mastery, meaning fluency with artificial intelligence systems and the ability to review their output. The connection between them he calls the validation tether: substantive validation of a machine's work requires internalised expertise, while surface validation does not. The supporting behavioural evidence is consistent. A Microsoft survey found 60 percent of employees skip accuracy checks on machine output. Harvard Business Review research found 40 percent received flawed material monthly. A study of 52 developers found those assisted by artificial intelligence scored 17 percent lower on comprehension tests despite completing tasks faster.

The advertising-specific data points carry furthest. IAB Australia's 2026 Talent Review, published June 2, 2026, found entry-level vacancies had collapsed to 1 percent of the available job pool. The Basis 2026 Advertising Agency Report, published April 20, 2026 from 213 respondents, found 87.3 percent believed the traditional agency model is broken or breaking within three to five years, 39.9 percent reported layoffs in the prior twelve months, and 77.7 percent of vice presidents and above planned increased artificial intelligence spending despite contracting headcount.

Set those numbers against the 575 positions removed at The Trade Desk on Friday, and against the token meters agencies spent the week installing, and the week resolves into a single sentence. The machinery is being audited more closely than at any point in its history, by courts, regulators, measurement vendors and the agencies' own dashboards, while the number of people employed to operate it falls.

Also noted

• September 3: Google deployed WeatherNext 3 across Search, the Gemini app, Maps, the Maps Platform Weather API, Earth Engine, BigQuery and Cloud Storage, cutting forecast refresh from six hours to one and resolution to 5 kilometres for surface variables, which moves weather-triggered creative from daily to hourly cycles. PPC Land
• September 4: Apple released the Apple Ads Platform API version 1.0 and set January 26, 2027 as the sunset date for the Campaign Management API version 5, giving developers roughly five months to migrate to a REST interface consolidating App Store and Apple Maps campaigns across about eighteen endpoint clusters. PPC Land
• September 4: Google expanded Waze advertising from the United States to 42 countries, confirmed through a Performance Max documentation change spotted by Barry Schwartz, adding Canada, Brazil, Mexico, Australia, Israel, the United Kingdom and thirty European markets for store goals campaigns. Search Engine Roundtable
• September 3: RedVest Media, the network Ace Hardware launched in late 2025, used an upfront event to add weather-triggered programmatic advertising, control-group incrementality measurement, a creator partnership with Influential and a DoorDash storefront integration, built on more than 80 million loyalty members and over 5,300 United States stores. Digiday
• September 3: Barilla, which reported 5.6 billion dollars of revenue in 2025, acquired the premium mac and cheese brand Goodles on undisclosed terms, buying into a 5.5 billion dollar American category growing 4.5 percent a year where Goodles sells at 3.79 dollars against Kraft's 1.39. Adweek