Publishers threaten to cut Google off as ad supply falls 40%

TITLE: Publishers threaten to cut Google off as ad supply falls 40%

DESCRIPTION: Publisher ad supply fell 40% in Q2 while Google Search ads grew 17% to $63.3 billion. Reddit, USA Today and Reuters now weigh cutting the crawler off entirely.

Covering July 15-24, 2026

Something broke this week, and it was not a product.

For twenty years the arrangement held on a single unspoken premise: Google took the content, and in exchange it sent back the audience. Nobody signed it. Nobody had to. The traffic arrived, the ad calls fired, and the arithmetic worked well enough that questioning it looked like commercial self-harm. That premise died somewhere between Tuesday and Thursday of this week, and it died with numbers attached.

Consider the pairing. On July 22, 2026, Alphabet reported second quarter revenue of $119.8 billion, with Search advertising up 17% to $63.3 billion. Search alone added $9.1 billion of annual revenue in a single quarter. On the same day, Reach plc, the largest commercial news publisher in Britain, reported that Google referral volumes fell 55% year on year, halved its interim dividend to 1.44p, and told investors it is no longer modelling a recovery.

Same day. Same ecosystem. Opposite directions.

That is not a coincidence of the reporting calendar. It is the structure of the thing, rendered in audited figures for the first time, and this week the industry finally said so out loud.

The supply side collapsed and nobody called it a crash

Start with the number that should have dominated every trade headline and mostly did not.

Publisher ad supply fell by up to 40% in the second quarter of 2026, Ozone benchmarking data shared exclusively with Digiday showed on July 15. The dataset tracked roughly 20 billion impressions. Between April and June, publisher ad request volumes fell 32% to 37% year on year in the United States and 39% to 41% in the United Kingdom. Ozone's publisher network includes the Guardian, News UK and Dow Jones' Wall Street Journal, which is to say this is not a measurement of the long tail. This is premium inventory, and it evaporated by roughly a third in twelve months.

Combined programmatic spend across the Ozone cohort fell 30.6% year on year across the first half. The regional split is where it gets interesting: spend fell 14.3% in the United Kingdom but 44% in the United States. Gabe Dorosz, advertising initiative lead at INMA, attributed the gap to market structure, describing the American open web as bigger and messier with a long tail of commodity supply still diluting yields.

Now the part that complicates the collapse narrative. Prices went the other way. Average eCPMs in June were about 30% higher year on year in the United Kingdom and 7% higher in the United States. Eliza Simonova, senior customer success and go-to-market manager at Ozone, framed the price movement as a market signal rather than a warning, noting that scarcity attracting higher payment is a healthy sign even as total spend declines.

Read those two figures together and the shape of the new market appears. Fewer impressions. Higher unit prices. Lower aggregate spend. The open web is not dying evenly; it is compressing into something smaller and more expensive, and the compression is partly deliberate. Dorosz noted that some publishers are actively shaping supply, cutting ad load and low-value bid requests to protect attention and price. That is a strategic response, not a passive collapse, and it deserves to be read as one.

Yield-led markets have a ceiling, though nobody knows where it sits. Luke Stillman, managing director at Madison and Wall, pointed to television as the precedent, noting that warnings about yield-led risk in TV ran for more than a decade while marketers kept paying multiples for smaller audiences. His view is that most migration into walled gardens will happen for reasons other than price, specifically data availability, measurement credibility and convenience.

Matt Barash, chief commercial officer at Nova Studio, put the mechanism plainly: the open web is shrinking not because demand for content has fallen, but because discovery has moved. Search created pageviews; social creates engagement. Every minute inside a feed is a minute not generating open-web inventory.

One channel grew. Apps were the only format up year on year in June, with spend rising about 23% and eCPMs about 42% in the United States. Web inventory rebounded 11% month on month but remained roughly 38.5% below the prior year. Out-stream video spend fell roughly 76% year on year, a number so severe it reads like a format obituary.

There is also a geopolitical layer in the Ozone data that most coverage skipped. The Iran-United States conflict, which began February 28, shows up clearly in American programmatic spend as a break from seasonal norms. In 2024 and 2025, United States spend dipped in February and recovered to or above its January baseline by May. In 2026 it slid month by month from January through April and never recovered to January levels by May, ending roughly 18% below its pre-conflict baseline while the United Kingdom sat about 10% above. Business-to-business technology and betting and gaming pulled back hardest in the United States while growing in Britain. Peace talks in mid-May produced only partial recovery, and the two largest buying pipes, DV360 and Amazon DSP, continued to decline through that window.

So the supply contraction is real, the price inflation is real, and a macro shock sits underneath both, distorting the American figures in ways that make clean attribution difficult. That nuance matters, because a lot of commentary this week collapsed all of it into a single story about artificial intelligence.

Where does displaced budget go? The week supplied partial answers, and they point away from the open web in every case.

Retail media absorbed a share. Criteo won a Massmart deal covering 280 stores across Africa, with Sponsored Product Ads already running on Makro.co.za and Onsite Display Ads to follow. Amazon DSP began naming the specific retailers its advertisements drive sales to, using the Amazon Shopper Panel to extend April 2026 category breakouts to named stores, though the rollout stays limited to the United States.

Amazon Business gained 1.8 million organizations as sales roughly doubled to $60 billion from the $35 billion recorded in December, with 97 of the Fortune 100 now buying through the platform. Amazon separately froze 2026 holiday seller fees while shortening inbound windows for every shipment type, holding the $100 upfront fee while peak rates add $0.32 per unit. Late stock loses Prime eligibility.

Out-of-home grew against the trend. JCDecaux reached 100% programmatic coverage in Latin America across 4,600 screens, adding nine markets from Panama to Paraguay and 8.2 billion monthly impressions, as outdoor advertising overtakes television in parts of the region.

Gaming opened new inventory entirely. EA turned 1 billion FC matches a month into advertising inventory, serving dynamic boards in FC and Madden through a proprietary ad server, with Visa, Red Bull and Lowe's already running campaigns.

Every one of those environments is a walled garden or a physical surface. None of them is the open web. Stillman's point about migration happening for reasons other than price looks correct: the budget is not fleeing high CPMs, it is following measurable outcomes and first-party data into places where the seller controls the whole stack.

There is a quality dimension buyers should hold alongside the volume story. IAS found made-for-advertising traffic rises 5% on Christmas Eve and Christmas Day, with impressions surging 280% during peak shopping weeks. AdExchanger reported that the percentage of advertising spend going toward made-for-advertising sites rose this year for the first time since 2023, in its July 1 roundup.

Legitimate supply contracts by a third while the share going to manufactured supply increases. Those two trends running simultaneously describe a market where scarcity is creating opportunity for exactly the wrong sellers.

There is a planning consequence buried in the eCPM figures that deserves separating from the doom narrative. A market where volume falls 40% and prices rise 30% is not simply a worse version of the old market; it inverts the skills that made buyers effective in it. Fifteen years of programmatic practice optimised for finding cheap inventory at scale, on the assumption that supply was effectively infinite and the job was filtering it. Dorosz named that assumption directly, describing the fifteen-year strategy of infinite supply as something that steadily drove CPMs down and produced a race to the bottom, with AI-driven traffic declines finally killing the premise.

If supply is finite and priced accordingly, the advantage moves to buyers who can identify which impressions are worth the higher price rather than those who can process the most of them. That is a different competency, and the measurement failures documented later in this edition suggest the industry is not currently equipped to exercise it.

The seller side faces the mirror image. Dorosz framed the opportunity as sitting with publishers who can prove they are different, surfacing first-party, contextual or attention signals demonstrating their impressions meet the quality thresholds buyers now actively seek. Reach's 49% RPM increase is one publisher demonstrating exactly that, and it is worth noting the company achieved it while its audience collapsed.

Spears at Ozone described the situation as a structural reset rather than a blip, noting that publishers with subscription businesses and apps look considerably more resilient against supply decline, and characterising the market as in recovery but volatile.

Google Zero arrives, and it is named

The phrase came from Nilay Patel, editor-in-chief of The Verge, quoted in a New York Times report and carried in AdExchanger's Daily News Roundup on July 21: for publishers, Google Zero is already here.

The supporting data in that roundup deserves attention because it quantifies a behaviour change rather than a traffic number. People spend on average between one and nine more minutes in AI Mode than in standard search, according to the Times, and a Growth Memo study found that only one quarter of AI Mode search sessions result in a click through to a web link. Three quarters of sessions terminate inside the interface. That is not a decline in referral efficiency. That is a different product with a different purpose.

The Wikimedia Foundation supplies the cleanest natural experiment available. Human traffic to Wikipedia declined 8% over the past year while total site visits rose, because AI bots scraping the site to inform their models more than replaced the humans. The foundation's response has been to promote its app, build social presence, and charge AI companies for training access.

Sit with that inversion for a moment. A site's traffic went up while its audience went down. Every measurement system built for the previous internet reports that as growth.

Google's position has not changed. The company says its AI features send billions of clicks to the web every week. Patel's rejoinder in the Times coverage was that right next to that claim sits a set of publishers whose businesses are being destroyed. Both statements can be true simultaneously, which is precisely the problem: aggregate click volume can rise while the distribution of those clicks concentrates so severely that most individual publishers experience collapse.

PPC Land has documented the distributional evidence directly. A study of 44 major United States publishers found aggregate organic search traffic across the group rose 5% since AI Overviews launched, with the gains concentrating in institutional brands. Separately, Ahrefs measured a 58% reduction in click-through rates for top-ranking pages where AI Overviews appear, nearly double the 34.5% the same firm recorded in April 2025. PPC Land also examined the gap between Google's click-volume claims and independent measurement following executive Nick Fox's July 17 statements.

Two-tier is the accurate description. Not collapse. Not growth. Stratification, with institutional brands absorbing what mid-tier publishers lose.

The regional picture reinforces the point. Search referral traffic fell across most world regions in the second quarter, with three regions each shedding 5% of search pageviews while social traffic filled part of the gap. Africa moved against the trend, climbing from 41% to 48%.

Google shipped controls that respond to the opt-out complaint, partially. Site owners now get a domain-level toggle to exit AI Overviews, AI Mode and generative AI features in Discover, clearing content within one to two days subject to caching, per newly published Search Console documentation. It operates at domain level only. It does not affect conventional ranking. It does not stop model training, which remains governed separately by the Google-Extended robots.txt token. Page-level grounding controls arrive March 3, 2027, following a nine-month implementation period the CMA granted after Google said the engineering required more work.

Three separate controls covering three stages of the pipeline, each configured differently, with the default set to include. Smaller operators without teams monitoring Google product launches are least equipped to exercise a right that requires noticing it exists.

The traffic case for staying in has weakened considerably. Beyond the Ahrefs correlation figure, a randomized field experiment involving 1,065 desktop Chrome users produced causal evidence that AI Overviews cut publisher clicks by 39.8% and raised zero-click searches by 34.5%, with clicks on sponsored results staying flat.

That last clause deserves isolation. Organic clicks fell by nearly 40%. Sponsored clicks did not move. Whatever AI Overviews do to the results page, they do it to the unpaid links.

France came out of the AI Overviews blackout during the same period, with French sites joining a market where AI Overviews cut position-one clicks from 27% to 11%. Google also brought image generation into AI Overviews using the Nano Banana model, which has produced over 5 billion images since its August 2025 debut.

Twenty-seven percent to eleven percent for the top organic position. That is the number underneath every strategy conversation happening in publishing right now, and it describes the destruction of the asset that organic search optimisation spent two decades building.

Crawler infrastructure kept moving in the background. Google renamed the NotebookLM user agent to Google-GeminiNotebook following the product rename, and updated its StoreBot accessibility documentation with clarifications on user agents, robots.txt checking, IP blocking and reprocessing timelines.

Publishers building blocking rules against a moving list of user agent names face a maintenance burden that grows with every product rename. That asymmetry favours the crawler operator permanently.

Digiday has been tracking the publisher counter-tactics, reporting that publishers and ecommerce brands under siege from AI crawlers are deploying LLM honeypotting, an old security trick updated for the model era, and covering SPUR's publisher-run Content Telemetry Framework, which aims to convert opaque scraping into a transparent usage-based licensing system publishers control.

Publishers building their own telemetry because the platforms will not supply it is the same institutional response that produced third-party ad verification fifteen years ago. That industry now exists because the platforms grading their own homework proved unacceptable. The precedent suggests where this ends.

The threat becomes explicit

Which brings us to the week's dominant story, and the reason the previous twenty years of assumptions stopped functioning.

Five of the largest names in American digital publishing are now openly evaluating whether to cut Google off from their content entirely, Wall Street Journal reporting published July 21, 2026 established. USA Today, Politico, The Economist, People Inc. and Reuters are all assessing how, or whether, to continue working with Google. Reddit sits in the same group.

The technical constraint underpinning the entire dispute is narrow and worth stating precisely, because it explains why the threat escalated from opt-out requests to full delisting. Publishers can turn off crawling for training Google's AI models. They generally cannot object to AI query results while remaining in traditional search, because Google's search index and its generative answers draw on the same crawl. PPC Land reported in February 2026 that Google described building a separate opt-out for AI Overviews as a substantial engineering challenge.

A publisher who objects to AI Overviews consuming its reporting cannot object only to that. It is all or nothing, and this week several companies started saying nothing might be preferable.

The executives went on the record, and their temperatures differ. USA Today Co. chief executive Mike Reed said it is time to take a stand, according to the Journal, and indicated willingness to turn off Google's access if search decline continues. People Inc. chief executive Neil Vogel was blunter, saying that blocking Google entirely is fully on the table. His portfolio includes People, Better Homes & Gardens and Travel + Leisure, and Google contributed just 25% of its traffic in the first quarter of this year, down from more than half two years earlier. That number is the whole argument: when the dependency falls far enough, the threat becomes credible.

Reuters president Paul Bascobert described the position as measuring economic trade-offs between search and AI summaries rather than declaring an intention. The Economist has drawn a line short of departure, with generative AI vice president Josh Muncke noting that traffic can be choppy and declining but that traffic is still traffic. Employees at Axel Springer's Politico have discussed limiting bot access to free articles, with a registration wall as one floated option.

Time took a stranger route, and it may be the most revealing item in the entire week. The magazine is helping advertisers place messages in front of bots, creating text-based ads that human readers never see but that AI crawlers might find, in the hope that chatbots reference the material when answering queries. Chief operating officer Mark Howard framed it as serving two audiences, humans and bots, with the latter treated as secondary but real.

Bots now make up more than half of all web traffic, according to Cloudflare figures cited in the Journal report. Cloudflare data published July 1, 2026 showed crawl-to-referral ratios ranging from 118 crawls per referral to nearly 50,000 at the extreme.

An advertising product aimed at machines is either the most cynical arbitrage of the year or an early sighting of the next medium. Possibly both.

The infrastructure making the threat executable arrived alongside the rhetoric. From September 15, 2026, Cloudflare will block crawlers it classifies as Training and Agent by default on ad-bearing pages for new domains joining its network, while Search crawlers remain allowed. The company also shifted its payment model away from per-crawl charging toward compensation tied to whether content contributed to a generated answer.

Default settings determine outcomes far more reliably than executive statements do. A policy that blocks by default at the network layer will move more inventory than any number of chief executives saying blocking is on the table.

Now the counter-evidence, which runs hard against the mood and deserves equal billing.

Publishers who blocked generative AI crawlers via robots.txt experienced a persistent 23.1% decline in log-monthly visits, with Comscore panel data measuring human browsing showing a 13.9% drop. A separate study covering November 2022 to May 2024 put the average cost at about 7% of traffic, attributing the loss mainly to reduced brand exposure inside AI answers rather than lost referral clicks.

Note carefully what those figures measured: blocking AI crawlers alone. Blocking Google's unified crawler removes a publisher from conventional search results as well, a materially larger step with no comparable dataset behind it. Nobody has run that experiment at scale, which means every executive contemplating it is estimating a cost that has never been measured.

David Buttle, chief executive of media consulting firm DJB Strategies, framed the stakes categorically in the Journal report, describing the situation as existential for some categories of publishers and noting they are examining more radical options.

The word existential is doing real work there. Companies do not evaluate self-harm as strategy unless the alternative looks worse, and the measured cost of blocking, somewhere between 7% and 23% on the AI-only version, is smaller than the 28% to 44% declines several of these publishers already absorbed without taking any action at all.

That is the calculation, stated plainly. If passive participation costs 40% and active withdrawal costs 23%, withdrawal starts looking rational. Whether those numbers are comparable is precisely the question nobody can answer, because the measurement layer producing them corrected itself by a factor of two this week.

The asymmetry that makes the threat weak is worth naming too, because it explains why nothing has actually been switched off. A publisher that delists loses its traffic immediately and completely. Google loses one source among millions, and its index degrades imperceptibly. Collective action would change that arithmetic, and collective action among competitors on commercial terms is the precise activity antitrust law spent a century restricting. Publishers are therefore negotiating individually against a counterparty whose cost of losing any single one of them rounds to zero.

Muncke's formulation, that traffic can be choppy and declining but traffic is still traffic, is the rational individual position. It is also the position that guarantees the collective outcome nobody in the group wants. Vogel's company reached a credible threat only by first reducing Google to 25% of its traffic, which took two years of building events, social referral and app businesses. The option became available because the dependency was already gone.

That sequence is the actual lesson, and it inverts the usual reading. The threat is not what creates leverage. Reduced dependency creates leverage, and the threat is merely what becomes sayable afterwards.

The correction nobody circulated

Here is where the week's most important epistemological problem surfaces, and it deserves more attention than it received.

The Wall Street Journal published corrected Semrush figures on July 22, 2026, after an earlier version of its article carried inaccurate data. The original piece stated Politico's traffic fell 23%, CNN's about 25%, and Business Insider's more than 85%. The corrected chart of trailing twelve-month averages records declines of roughly 3.2% for People, 18% for The Wall Street Journal, 27% for Time, 28% for USA Today, 35% for Reuters, 43% for Business Insider and 44% for The Washington Post, alongside increases of about 16% at The Guardian and 15% at the BBC.

The uncorrected version circulated widely first. A post by SEO consultant Glenn Gabe drew 28,000 views, and a full-text repost by Cyrus Shepard reached 8,059 views, both before the correction published.

Two of the ten publishers charted posted gains. That single fact undermines any clean narrative of universal collapse, and it is the detail most likely to be dropped from the retelling. The direction of the trend survives the correction. The magnitude for several titles does not.

There is a lesson buried in this that has nothing to do with Google. An industry making capital allocation decisions worth hundreds of millions of dollars is running on measurement infrastructure that produced an 85% figure and a 43% figure for the same publisher over the same period, one day apart. The delisting debate is being conducted on top of numbers that moved by a factor of two within twenty-four hours.

Reach converts percentages into pounds

Reach plc supplied the audited version, and audited numbers behave differently from analytics dashboards.

The company reported half year results for the six months ended June 30, 2026 showing group revenue of £232.9m, down 9.0% from £256.0m. Digital revenue fell 11.4% to £54.2m. Print revenue fell 8.3% to £178.0m. Reach owns the Mirror, Express, Daily Star, Daily Record, Manchester Evening News, Liverpool Echo and roughly 120 other titles.

Three figures describe the entire problem. Google referral volumes fell 55% year on year. On-platform page views fell 40%. Revenue per thousand page views rose 49%. A 40% audience decline paired with a 49% yield increase produced an 11.4% digital revenue decline, which means the yield improvement absorbed most, though not all, of the volume loss.

The internal split shows where damage concentrated. Indirect revenues, meaning revenue generated on social platforms or programmatically on owned sites, fell 16.2% to £31.3m. Direct revenues fell 3.9% to £22.9m, and excluding the local advertising business being deliberately reprofiled, direct revenues grew 6.0%. Roughly 33% of digital revenues are exposed to on-platform activity, and that third is programmatic.

Quarterly detail sharpens it further. Digital revenue fell 8.1% in the first quarter and 14.5% in the second. Indirect revenue fell 10.5% then 21.3%. Deterioration accelerated between quarters even as page views themselves steadied.

Chief executive Piers North addressed the trajectory during the results presentation, noting stabilisation in on-platform audiences over the previous 100 days while adding that the company must work to the assumption that referrals are unlikely to recover. That clause is the operative one. Reach is not modelling a rebound. It is modelling permanence.

Cost action is what preserved the business. Adjusted operating costs fell 10.3% to £190.4m, ahead of the 5% to 6% guided target. Labour costs fell 9.8% to £96.9m, with average employee numbers falling from 3,555 to 3,023. That is 532 people. Newsprint costs fell 16.6% to £15.9m. The gap between a 9.0% revenue decline and a 10.3% cost reduction held adjusted operating profit within £1.8m of the prior year at £43.0m, and adjusted operating margin actually improved one percentage point to 18.5%.

The statutory picture diverges sharply: an operating loss of £43.5m against a £29.7m profit, driven by £86.5m of adjusted items including a £36.1m non-cash impairment on two print site closures, £21.7m of amortisation and £18.9m of restructuring. The amortisation reflects an accounting judgement change. Publishing rights and titles had previously been treated as having indefinite economic lives; Reach reassessed that in light of structural shifts and now applies a 15 year useful economic life.

A media company formally deciding its mastheads have a finite lifespan is the sort of accounting footnote that carries more meaning than most press releases.

The dividend halved to 1.44p from 2.88p. The board cited trading performance, the £25m cash cost of print site consolidation, £57m of pension contributions in 2026, and continued strategic investment. Chief financial officer Darren Fisher tied the decision to the pension timetable and the rebalancing of capital toward organic investment.

Print still supplies about 75% of revenue, and its trajectory worsened during the period. Circulation volumes fell 22% across the half and 23% in the second quarter, against normal decline rates of 17% to 20%. Two cover price increases held circulation revenue decline to 6.6% at £134.7m. Print advertising fell 11.1% to £24.6m, outperforming the volume trend on food retail spending, World Cup activity and government spending. Deliveroo and Tesco were named as partners increasing activity as the England team progressed.

That detail is worth flagging for print buyers. A tournament produced measurable incremental demand for a format whose structural decline is thoroughly documented.

The pivots are early. Digital subscriptions launched across 15 brands have produced over 40,000 paid subscribers against a year-end target of 75,000, priced between £1.99 and £6.99 monthly for an ad-lite experience. North was candid about maturity, describing the company as having launched paywalls rather than subscriptions and identifying conversion improvement as the technology priority. Studio revenues grew over 70%, social video views rose 55%, and off-platform revenues grew more than 90%, though off-platform monetisation yields less than on-platform.

AI licensing forms the third strand, and North described a funnel with large platforms at the top, mid-sized business arrangements in the middle, and real-time marketplaces at the base. A deal with Amazon signed in March generates ongoing revenue. The stated ambition is payment at high volume every time content is used by an AI bot, with North drawing an explicit analogy to real-time bidding infrastructure.

Take that analogy literally, because it sets expectations correctly. Real-time bidding took years of standards work. The equivalent infrastructure for AI content access remains under construction, though the components are appearing: Cloudflare's pay per crawl service activated the HTTP 402 status code in July 2025, and the IAB Tech Lab convened a Content Monetization Protocols working group the following month.

Reach is not an outlier. Future plc reported a 67% profit decline in May 2026, with its chief executive sorting the brand portfolio explicitly by degree of Google dependency.

Brussels prices the same problem at 890 million euros

While publishers debated exit, the European Commission acted, and the two stories share a defendant.

The Commission issued two non-compliance decisions against Google under the Digital Markets Act on July 23, 2026, imposing fines of 460 million euros for self-preferencing in search rankings and 430 million euros for anti-steering on Google Play. The combined penalty reaches 890 million euros. Google has 60 days to comply or faces periodic penalty payments of up to 5% of total worldwide turnover.

That distinction between the fine and the penalty payments matters more than the headline figure. A fixed 890 million euro penalty is a known quantity that can be provisioned. A daily percentage of global turnover is an open-ended liability that scales with time. The 60-day window is the operative deadline, not the fine.

On search, the Commission found Google displays its own shopping, hotels, transport and sports services more prominently than comparable third-party services, including at the top of the results page or through enhanced visuals and filters. The remedy ordered is specific: treat third-party services in a fair and non-discriminatory manner by reference to its own. Teresa Ribera, Executive Vice-President for Clean, Just and Competitive Transition, framed the finding in terms of competitive merit, stating that the best products should succeed because they are better rather than because they are owned by the company running the search engine.

On Google Play, the finding has two components. Google prevents developers from freely communicating and promoting offers and concluding contracts in distribution channels of their choice. Separately, the Commission examined fees and found that the level of steering-related fees and the length of the charging period went beyond what the DMA permits. Fee level and duration, not the existence of a fee. That distinction defines the compliance question.

The decision records partial credit, which enforcement decisions do not always do. Google has proposed and started testing changes to how it presents its own free services such as shopping, hotels and flights, and the Commission characterised this as substantial progress. Changes covering shopping ads and content-related services including sports remain under assessment.

Then there is the sentence that should concern anyone modelling search visibility. Google has submitted proposals on how it plans to apply the principles of the decision to AI Overviews and AI Mode, and dialogue on those proposals continues. The Commission did not indicate a timeline.

Read that carefully. The regulator has signalled that its self-preferencing principles apply to AI-generated search surfaces without specifying how. It has not resolved the question. It has deferred it, and the deferral covers the surface where the majority of future search interaction will occur. PPC Land reported in May 2026 that Commission concern extended to whether deploying a proprietary language model to generate the dominant summary display amounted to a new form of the same self-preferencing problem.

Context for the number. The 890 million euro total exceeds the 500 million euro Apple fine and 200 million euro Meta fine issued in April 2025, covered in PPC Land's report on the first major DMA enforcement actions. Against Google's broader European liabilities it looks modest: the Court of Justice confirmed a 4.125 billion euro Android fine on July 2, 2026, and a separate 2.95 billion euro penalty landed in September 2025 over publisher ad servers and programmatic buying. The DMA permits fines up to 10% of global annual turnover for a first offence, rising to 20% for repeats. This lands well below the ceiling.

Google's response was not conciliatory. The company is weighing an appeal, with Kent Walker characterising the decision as product degradation driven by self-serving complainants. Whether Search retains instant pricing displays in Europe now rests on a September 21 deadline.

Enforcement history explains the timing. Google was designated a gatekeeper in September 2023. The Commission opened both investigations on March 25, 2024, and informed Google of its preliminary view on March 19, 2025. Sixteen months separate that preliminary finding from this decision, a delay that drew public criticism. PPC Land covered the March 2026 open letter in which eighteen European industry and consumer organisations demanded a formal decision before the two-year mark, noting Google holds over 90% of EU search market share.

Parallel action continues. The Commission adopted binding specification decisions requiring Google to open Android to competing AI assistants and share anonymised search data with rival engines on July 16, 2026. AdExchanger reported on July 17 that the EU ordered Google to scrap restrictions on how rival AI services operate on Android, opening 11 Android features to competitors, in its Friday roundup.

For travel advertisers specifically, the named inclusion of hotels and transport carries direct consequences. PPC Land noted research from Mirai finding hotels in DMA regions experienced a 30% drop in clicks and a 36% decrease in direct bookings compared with non-DMA markets, in coverage of Google's travel ad format changes. Whatever remedy Google implements will move those numbers again, in a direction not yet knowable.

Reddit debates whether it is eating itself

The most intellectually honest moment of the week came from an executive admitting he does not know the answer to his own company's central strategic question.

At Cannes, Reddit's executive vice president of ads monetization Roelof van Zwol was asked directly whether the platform's content-licensing deals with OpenAI and Google compete with its advertising business, Digiday reported on July 15, 2026. He described it as something the company is intensely debating.

The tension is structural rather than tactical. Reddit sells advertising to marketers on the premise that its community conversations offer targeting signal no other platform can replicate. It simultaneously sells those same conversations to companies training models that answer user questions without sending anyone to Reddit. If the second business succeeds completely, it destroys the scarcity underpinning the first.

The contract economics are now public. Reddit struck a 60 million dollar annual deal in 2024 allowing Google to use its material for AI training, granting Google effective exclusivity over indexing recent Reddit content while other search engines were blocked. That agreement is ending soon and under renewal negotiation.

Reddit's own disclosures put the figure in context. Data licensing revenue reached 36 million dollars in a single quarter alongside 549 million dollars in advertising. During the first-quarter 2026 earnings call, an analyst pressed chief executive Steve Huffman on whether the economics adequately reflect the value of Reddit's data. Huffman declined to comment on specifics.

Run the ratio. Licensing produces roughly 6.5% of what advertising produces per quarter. If feeding the models costs Reddit even a modest fraction of its advertising differentiation, the trade is bad arithmetic. Van Zwol's willingness to say the question is open, rather than assert a comfortable answer, is more useful to the market than confident spin would have been.

There is a second-order consequence for marketers that received almost no coverage. A substantial share of what AI systems say about brands is drawn from Reddit threads and similar community sources. If Reddit restricts access, that citation base shifts without notice to the marketers relying on it. PPC Land's coverage of Semrush's AI Visibility Index recorded that only 36 of more than 1,200 tracked brands achieved consistent visibility across every AI platform, a fragility that deepens if major sources exit.

Brand visibility work built on a citation base that one company might switch off during a contract renegotiation is not a strategy. It is an exposure.

The advertising is already inside the answers

While publishers debated whether to feed the machines, the machines started selling advertising, and the growth rates are absurd.

Adthena counted 7,378 ChatGPT advertisers as placements grew 97 times in a single quarter. Booking.com leads all three tracked markets, and the United States accounts for 60.1% of observed advertisers. Adthena reports ChatGPT ad frequency now running four times that of Google's AI Mode.

Ninety-seven times in a quarter. That is not adoption; that is a land rush.

The pricing is chaotic to match. United Kingdom ChatGPT cost-per-click figures swung 278% with no published explanation, disclosed alongside an Adthena partnership with dentsu. Three dentsu clients sit among the top nine United Kingdom ChatGPT advertisers, and a new Decision Intelligence product targets what the firm describes as a costly auction blind spot.

A 278% swing in cost-per-click with no explanation available to buyers is the defining characteristic of an immature auction. Anyone spending into that surface at scale is currently paying for opacity as well as inventory.

The competitive picture beneath the advertising is shifting too. ChatGPT is losing web share to Gemini and Claude as advertising penetration hits 26%, Similarweb data shows. The same dataset puts AI Overviews in 43% of United States searches, with generative AI web visits reaching 9.5 billion monthly, up 70% year on year.

OpenAI has been expanding the advertising business internationally, offering ChatGPT ads in new markets including the United Kingdom, Brazil and Japan.

Meanwhile Taboola expanded DeeperDive into an advertising network for AI applications and agents, inviting external conversational interfaces including utility apps, finance tools and virtual assistants to use its ad API and insert native units into their own chats. Chief executive Adam Singolda's reasoning is that running high-quality large language models at scale is expensive and consumers will not maintain dozens of separate subscriptions, so AI companies must turn to advertising and commerce.

He is almost certainly right about the economics, and the conclusion follows uncomfortably. The interfaces currently absorbing publisher traffic will monetise the same way publishers did, using the same mechanisms, with the same intermediaries, minus the content production costs.

Google's own AI advertising surfaces are past the roadmap stage. AI Mode surpassed one billion monthly active users after expanding globally in October 2025. Text ads gained contextual sitelinks based on conversation. Direct Offers, a pilot opened in January 2026 with Chewy, Gap and L'Oreal, is gaining momentum, and Highlighted Answers, which places clearly marked sponsored links inside list responses, debuted at Google Marketing Live on May 20, 2026.

Nine weeks separated the product demonstration from executives citing those formats as revenue drivers to investors. The gap between announcement and monetisation has effectively closed.

AI Max becomes infrastructure while the evidence stays contested

Half a million advertisers now run AI Max campaigns, with chief business officer Philipp Schindler stating that adopters of AI-powered campaigns see an average of 15% more conversions or value on Search at similar return on ad spend.

That figure sits one point above the 14% uplift Google claimed when AI Max entered open beta in May 2025. Both numbers come from Google's own measurement.

Independent testing reached a different conclusion. Tests published in November 2025 found AI Max produced a higher cost per conversion than broad, phrase and exact match within identical campaigns. That gap between vendor claims and practitioner data has not been resolved publicly, and it is now being resolved by default rather than by evidence: Google declared AI Max out of beta on April 15, 2026 and confirmed Dynamic Search Ads retire in September 2026 with automatic upgrades into the newer format.

Fourteen months from experiment to default infrastructure, with the performance question still open. Verification sits entirely on the buy side, and the deadline for verification passed while the debate continued.

The automation extends inside Google itself. According to Schindler, 83% of the sales team uses Gemini-assisted tools weekly, producing a 20% higher win rate when customised pitch narratives are involved, and agentic support systems autonomously resolve 75% of ads customer support queries.

Three quarters of advertiser support queries now resolved without a human. Advertisers experiencing account problems this year might reasonably connect that figure to their support experience, particularly given a policy change covered below.

The network line quantifies the open web

Google Network revenue, the money Google earns serving ads on properties it does not own, came in at $7.303 billion, down 1% from $7.354 billion. Schindler gave the segment a single sentence in prepared remarks.

The trajectory tells a longer story. The segment fell 2% in the first quarter of 2025, 1% in the second, then accelerated to a 4% decline in the first quarter of 2026. Against that, a 1% decline represents marked softening. Whether that signals stabilisation for the AdSense, AdMob and Ad Manager ecosystem or a seasonal pause inflated by World Cup demand cannot be answered by one quarter.

Set the two lines side by side. Network produced $7.35 billion in mid-2025 and produces $7.30 billion now. Search added $9.1 billion over the same period. The open web's position inside Alphabet's income statement is a rounding error against the growth of the owned-property business.

Traffic acquisition costs rose 10% to $16.2 billion, slower than the 14% growth in advertising revenue, a spread consistent with growth concentrating on owned properties rather than revenue-shared network inventory.

The spending underneath is the part that reframes everything. Capital expenditures reached $44.9 billion for the quarter, double a year earlier, with roughly 60% going to servers and 40% to data centres and networking. Alphabet raised full-year 2026 guidance to $195 billion to $205 billion, up from $180 billion to $190 billion, the second increase this year from the $175 billion to $185 billion range set in February. Guidance has climbed $20 billion at the midpoint since January.

Free cash flow was negative $5.9 billion for the quarter. Trailing twelve-month free cash flow of $53.3 billion fell 20% year over year. Long-term debt stands at $98.2 billion, up from roughly $16 billion twelve months earlier, after an equity capital program announced June 3, 2026 produced net proceeds of $49.6 billion alongside $20.3 billion in senior unsecured notes.

A company generating $63.3 billion in quarterly search advertising is running negative free cash flow and has grown its long-term debt six-fold in a year. The buildout consuming that capital is the same buildout producing the AI surfaces that are compressing publisher traffic. The open web is being disassembled and the proceeds are being reinvested in the machinery doing the disassembling, financed partly by debt.

One further detail from the AdExchanger roundup deserves recording: investors did not raise Google's advertising business or advertising in general once during the question and answer portion of the Alphabet earnings call.

The largest advertising business in history, and the analysts wanted to discuss something else.

Advertisers lose the right to appeal

Buried under the week's larger stories sat a policy change that should alarm anyone managing accounts at scale.

Google Ads eliminated appeals for policy decisions older than six months. Policy decisions past that age become final inside the account, and Google provided zero notice period.

No notice period. A remedy that existed on Tuesday did not exist on Wednesday, retroactively, for decisions already made. Combine that with agentic systems autonomously resolving 75% of support queries and the recourse structure available to a suspended advertiser becomes difficult to describe with confidence.

Other advertiser-facing changes landed the same week. Google Ads banned blurry image assets and restricted eligibilityto accounts with 60 days of history, active text ads and Search spend above zero. Sensitive verticals stay locked out, and AI labels now bypass the overlay ban.

The API layer shifted underneath as well. Google Ads API v25 removed the CustomerLifecycleGoal and CampaignLifecycleGoal resources, breaking automated goal configuration and forcing code rewrites. Five features arrive alongside the removals, including Shorts likes and comments in reporting.

Retailers face a firmer deadline. Google is forcing local inventory ads on all Shopping campaigns from August 31, removing the enable_local opt-out. Code running on v25.1 that sets the field to false will trigger an OPERATION_NOT_PERMITTED error at runtime. Search Engine Roundtable covered the advertiser email announcing the change, noting the inventory filter as the remaining control, in its July 16 report.

Local Services Ads are folding into Google Ads, with historical reports not carrying over and manual bidding disappearing. United States home service accounts migrate first in August; non-United States accounts wait until 2027.

Reports that do not carry over means historical performance data ceases to exist for accounts that migrate. Any advertiser wanting year-over-year comparison in that vertical needs to export before August.

Search Engine Roundtable also documented community concern about an August 18 change to bidding for campaigns limited by budget, with practitioners suspecting a broader smart bidding update than Google communicated and Google denying it, in reporting published July 17.

Policy consolidation continues in parallel. Google will merge Shopping ads and free listing policies into a single document in September 2026, a change PPC Land noted adds zero new merchant restrictions.

Gambling policy moved in two directions simultaneously, which is worth noting for anyone in that vertical. Google is dropping end-advertiser certification for gambling ads across 37 markets on Authorized Buyers from August 10, meaning publishers may see more online gambling ads through programmatic demand. Simultaneously, Google is blocking mixed gambling accounts and forcing one application per country on the Google Ads side, requiring separate accounts for online gambling and social casino certification, with affiliates obliged to confirm in footers that all outbound links are licensed.

Looser on the programmatic side, tighter on the direct side, in the same month.

Regulated verticals saw geographic expansion elsewhere. Spain opened to telemedicine advertising for LegitScript certificate holders from August 5, with Spanish telehealth spending forecast to compound at 11.6% through 2033. Norway, Iceland and Liechtenstein completed the 30-country European Economic Area map for Google crypto advertising following MiCA licensing rules, with CASP licensing plus Google certification determining eligibility from August.

Fourteen new Shopping markets opened including Bulgaria, Serbia and Malta, though over-the-counter and prescription drug listings stay blocked across all fourteen.

The measurement layer is failing quietly

A cluster of studies published this week measured how badly the data underneath campaign decisions performs, and the numbers are worse than most practitioners assume.

Identity errors cut campaign return on investment by 70% in LiveRamp simulations, dropping simulated ROI from $1.50 to $0.43. The study found that just 1% of wrong data was sufficient to reverse channel rankings.

Read that again. One percent contamination reverses the ranking. A profitable campaign measured through a flawed identity layer looks like a loss, and the resulting budget decision moves money away from something that was working.

Children's data performs worse. A CIMM report found presence-of-children data proves correct just 42% of the time, wasting $590,000 of every $1 million campaign. Ten companies back a new cross-industry taskforce, and linear television has dropped to 6% of children's video time.

A 42% accuracy rate is worse than a coin flip on a binary attribute. Advertisers paying premiums for that segment are paying for noise.

The blind spots compound. AppsFlyer research covering 157 United States agency and brand leaders found 58.6% of marketers underinvest in channels AI cannot measure, with social media and connected television leading the gaps. AdExchanger has been tracking the same problem from the attribution side, asking whether effective attribution is now an illusion as match rates fall.

Automated optimisation systematically starves channels it cannot see. That is not a measurement problem; it is a capital allocation problem wearing a measurement problem's clothing.

Yahoo DSP responded by adding Truthset-rated audiences scored against 30% to 60% data waste, with independently scored AAA-B segments matching Yahoo ConnectID on a one-to-one basis across connected television, mobile and open web. The VAB added three vendors to its measurement directory, now cataloguing 20 including datafuelX and VideoAmp, arriving as Nielsen's currency position and television's self-grading model both fracture.

Fraud sits alongside inaccuracy. Ofcom published a report on best practices for countering platform advertising fraud, covered by AdExchanger on July 21. Rob Leathern, who led advertising integrity and privacy teams at Facebook and Google, argued the most important mandate would require platforms to provide trusted third parties with access to large representative random samples of all active ads plus the total number running. No major platform reports this.

Leathern identified the incentive problem directly: platforms benefit from fraud, because advertisers still pay when ads run on fraudulent publishers, and refunds often arrive as advertising credits rather than cash. Sometimes the advertiser is the scammer, which also works out well for platform revenue. Victims of search or social fraud are not repaid by the platform either.

For platforms to take fraud seriously, it must be something they pay for rather than profit from. That sentence is the entire regulatory theory of the case.

Scale gives it weight. A VAB analysis found Facebook banned 3.5 billion fake accounts in 2025, equivalent to 43% of the global population, yet Meta allows advertisers 8 to 32 fraud strikes before an account ban.

Thirty-two strikes. Most disciplinary systems stop at three.

AdExchanger also reported that a demand-side platform refused advertising watchdog Check My Ads a seat on its platform even after both parties cosigned a master service agreement, citing concerns about protections for vendor and supply partners, in its July 21 roundup.

An industry that cannot tolerate an auditor is telling the market something about what an audit would find.

What AI actually delivers, measured

The gap between AI adoption claims and AI results narrowed into hard numbers this week, and the results are unflattering.

WARC and TikTok research surveying 400 marketers across four markets found generative AI lifted creative volume for 88% but improved quality for only 45%.

More output, roughly half as much better output. Every incremental unit of volume without quality improvement is inventory competing for the same finite attention, which is a description of commodity production rather than competitive advantage.

Google's own research pointed the same direction. AI touches 68% of jobs but only 21% of their tasks, with workplace adoption reaching 88% of United States employment while median task coverage stalls at one fifth. Adobe found workers estimate 32% of their tasks could improve with creative AI, yet only 9% use it, with only a third having received training and 400 hours wasted.

Adoption is near-universal. Depth is shallow. The displacement thesis and the productivity thesis are both weaker than the discourse suggests.

Quality failures are becoming visible to advertisers. AdExchanger reported that Meta drew advertiser anger for pushing AI tools despite clear errors, including an ad for outdoor brand REI featuring a bike with two handlebars and a creative centring a man in a campaign for a women's networking group, in its July 14 roundup. Less visible errors include altering a brand's actual products, which is the category that should worry brand teams most.

The counter-movement is measurable too. Inspired Taste is paying three home cooks $2,000 each to test recipes, with survey data showing a 300% blog preference over AI. Human verification is becoming a product feature precisely because generated content is abundant.

Google addressed the indexing consequences directly. John Mueller and Martin Splitt discussed how quality issues including low-quality AI-generated content can result in pages not being trusted and not indexed, showing as crawled but currently not indexed status. YouTube separately clarified its policies on monetisation of AI slop and upsetting videos, AdExchanger noted in the same roundup.

The naming convention matters. When platforms adopt the word slop in policy documents, the volume problem has become an operations problem.

On the regulatory side, the EU set a 200-token watermark threshold and named deployers liable from August 2, with penalties reaching 3% of turnover. Non-signatories must prove their own methods match the Code text.

Ten days from the compliance date at time of writing, and the labelling infrastructure at most publishers is not built.

Buyers are hedging on the naming question too. Fractl research found 81% of marketers reject the GEO pitch and still call it SEO, with buzzword-heavy pitches ranking as the number one vendor red flag for 36% of buyers and just 27% of teams using any label beyond SEO.

Vendors renaming the category faster than practitioners will accept the rename is a reliable indicator of a market in its selling phase rather than its proving phase.

Semrush research surveying 519 United States professionals, 84% of whom use AI on deals above $1,000, found brand recognition sways just 7% of business-to-business AI buyers, with precise use-case fit mattering more than name recognition inside AI answers.

Decades of brand-building equity converting to 7% influence inside the interface that increasingly mediates purchase decisions is the kind of finding that should restructure budgets. It probably will not, quickly.

The agency layer is absorbing the same pressure from two directions. Digiday reported that agency bosses say the AI gap with clients keeps widening, and that agencies chasing an edge from AI agents must simultaneously fend off brands looking to in-house with the same technology. Omnicom, meanwhile, has been testing AI agents to cut out advertising technology intermediaries.

Agencies selling clients the capability to replace agencies is an unusual commercial position, and the resolution will not be comfortable for everyone currently holding a seat.

Production economics are compressing accordingly. AI Digital bought a Barcelona agency to feed a studio that has cut an eight-to-twelve-week production process to about one week since May. GumGum's Mindset Agent converts an uploaded brief into custom targeting on Trade Desk, with Heineken, e.l.f. and the BBC testing the underlying graph. HubSpot's agent cut 350 hours a year for Ignite Reading across 25 states, reducing a 20-minute district task to seconds, though pricing for the Professional and Enterprise public beta remains unstated.

Set the WARC finding against these deployments and the picture resolves. The time savings are real and documented. The quality improvement is not, for 55% of marketers. Compressing twelve weeks to one week produces twelve times the output at roughly the same quality, which is only an advantage while competitors have not done it too.

Ownership concentration underneath all of this went largely unremarked. Northeastern data covering 4,608 United States users showed Alphabet takes 35% of desktop time, with Gmail alone beating 300,000 sites and five firms holding half of all desktop attention. AI tools have now edged past news as a browsing destination.

Five companies hold half of desktop attention. That is the market structure inside which every publisher delisting decision, every measurement debate and every regulatory remedy this week actually operates, and it is the variable none of those interventions changes.

The consolidation continues on the vendor side too. AdExchanger reported that Vista Equity Partners and Quinti Capital submitted a bid to acquire Criteo at roughly 50% above a market capitalisation of about $1.1 billion, in its July 8 roundup. Digiday framed the takeover interest as further proof of private equity's continued appetite for advertising technology. Digiday separately reported that OpenX is hunting a new chief executive after parting ways with Matt Sattel.

Independent advertising technology going private while the supply it monetises contracts by a third is not a coincidence of timing. Public markets price declining supply immediately; private owners can restructure without quarterly disclosure.

Publisher-side experimentation continues in the meantime. Bauer Media France exited advertising sales entirely as Azerion absorbed its whole team, leaving the publisher of Tele 7 Jours and Maxi with no in-house sales staff after a January 2026 pilot. German regional publishers gained 30.9% while United States news dropped 17.1%, with German news logging seven straight months of growth.

A nearly 25-point gap between German regional publishers and the American news market, measured over the same period under the same AI conditions, suggests that market structure and reader relationship matter more than the technology narrative allows. Something is working in Germany that is not working in the United States, and it is not a different Google.

Measurement tools arrive for the surfaces nobody can measure

Microsoft moved on visibility infrastructure while the debate continued. Clarity's beta now sorts AI citation queries into ranked topics by citation count and share of authority, with export support for publishers. It is the second Clarity AI visibility release in two weeks.

Attention measurement moved into live bidding as well. Google put Adelaide attention scores into DV360 bidding from mid-July, shifting Attention Units from post-campaign reporting into live bid decisions. Amazon DSP received the same signal in June.

Attention as a bidding input rather than a reporting output is a meaningful change in what the auction optimises toward, and it arrives precisely as impression volume contracts. Scarcer inventory judged on attention rather than viewability is a different market with different winners.

Digiday reported that media agencies are tentatively exploring vector-based planning as a future of targeting, and that the industry is not yet ready to let AI agents spend advertising dollars.

Caution about autonomous spending, alongside enthusiastic adoption of autonomous campaign optimisation, is a distinction the market is drawing on control rather than capability.

Television and audio consolidate while the web contracts

Away from search, the video and audio markets spent the week consolidating, and the direction is consistent: fewer, larger sellers.

Peacock reached its first profit at $189 million as Comcast domestic advertising revenue jumped 55% to $2.16 billion on World Cup and NBA playoff inventory. Streaming buyers now face a scaled seller heading into a corporate split. AdExchanger noted Comcast co-chief executive Mike Cavanagh describing meaningful profitability just as the company decided to let Peacock leave the nest, in its July 21 roundup.

Netflix opened its 250 million monthly user inventory to Trade Desk buyers and dropped spend minimums, removing the private guaranteed and private marketplace gating. AdExchanger separately reported Netflix considering live channels streaming certain programs or genres continuously, in its July 13 roundup, with the company still expecting to double advertising revenue this year.

Netflix rebuilding linear television inside streaming, twenty years after streaming was sold as the alternative to linear television, is the sort of circularity worth noting without comment.

Regulatory friction hit the studio consolidation track. Paramount must quit UIP within 13 months to win European clearance for the Warner deal, with clearance covering 19 EEA countries and barring joint distribution with Universal or Disney for ten years. AdExchanger reported that a federal judge halted the Paramount Skydance and WBD merger after hearing arguments from a coalition of states, in its July 21 roundup.

Publishers built their own video infrastructure rather than renting it. Business Insider launched its own FAST channel and picked Magnite to run connected television advertising sales, giving the 38-million-subscriber publisher control over ad serving and demand. The Athletic gained its first connected television home as Fubo opened six shows to sponsors, including Rates & Barrels and NBA Daily.

News brands building television distribution because search distribution stopped working is the clearest strategic response of the week, and considerably more constructive than the delisting threats.

Amazon expanded location-based interactive video advertising to Canada, Mexico and Brazil, nine months after the United States launch, restricted to first-party ad server video. MediaPost covered the same expansion, noting that a viewer in Toronto sees a Canadian advertisement while a viewer in Sao Paulo sees a local storefront address, and that Amazon business division sales rose nearly 30% globally this year, in reporting published July 22.

In audio, SiriusXM gains 42 Audacy local stations across 29 United States markets from September 1, with 23 sports stations arriving first for football season and 19 news and talk outlets later in 2026. Nine reach all subscribers; 14 require 360L hardware.

Comscore extended transcript-level targeting to five audio platforms, integrating Proximic classifiers with Spotify, SiriusXM, Triton, Acast and Libsyn, with weekly podcast listening reaching 45% among 25 to 34 year olds. MediaPost covered the same integration, quoting Proximic head of partnerships Jessica Trainor on podcasts having a cultural moment, in reporting published July 23. AdExchanger framed the move as Comscore wanting podcast advertising to feel more like buying connected television or display, in coverage of the announcement.

Buying audio at episode level rather than show level is the same shift search made from site to page, roughly fifteen years later.

The enthusiasm has limits. Podcast adoption reached 44.9% among marketers, but just 16.8% track formal return on investment and 41.8% spend more on production than promotion.

Spending more on making the thing than on ensuring anyone hears it is a recognisable failure pattern, and it predates podcasting by about a century.

Measurement standards are catching up. The IAB Tech Lab is forcing video podcasts under the same download counting rules, with version 2.3 open for public comment until August 19 and the term listener replaced by podcast consumer. A v3.0 guideline is expected in 2027.

Also noted

• July 20, 2026 - A federal judge dismissed both counts against SerpApi, finding Google's SearchGuard protects advertising revenue rather than copyright, denying Google's attempt to treat search scraping under the DMCA as it would DVD piracy, with 21 days granted to refile a narrowed claim.
• July 22, 2026 - Advertising industry groups including the ANA, 4As and AAF wrote to lawmakers arguing that the Childhood Diabetes Reduction Act of 2026, introduced by Senator Bernie Sanders, raises significant First Amendment concerns over its junk food advertising restrictions, MediaPost reported.
• July 22, 2026 - France banned under-15s from social media but stripped the age verification clause, leaving enforcement reliant on EU Digital Services Act rules, with existing accounts given four months past the September 1 start date.
• July 21, 2026 - Adweek examined how four retailers are pitching proprietary creative advertising formats, with Albertsons launching a 22-episode branded content series with Procter & Gamble and Instacart opening a creative services arm, though buyers describe the offerings as experimental.
• July 23, 2026 - Acxiom cut identity build time from 24 months to weeks using zero-copy architecture with Salesforce, keeping customer data in place as the Omnicom-owned unit deepens platform ties, raising neutrality questions for buyers.

The week in one sentence

The open web lost roughly a third of its advertising inventory, the company that organised it reported record search revenue and negative free cash flow, a regulator fined that company 890 million euros while deferring the question that matters most, and the publishers who supply the content spent the week discussing whether to stop.

None of those four facts caused the others directly. All four describe the same reorganisation, and the reorganisation is not finished. What ended this week was the assumption that the arrangement was permanent, mutual, and not up for discussion.

Mike Reed set a condition rather than a date. Neil Vogel described an option rather than a decision. Josh Muncke ruled departure out for now. Nothing has been switched off.

But the calculation is being made out loud, by named executives, at companies large enough for the outcome to register in the index. That is new, and it is the part worth watching through August.