Two documents decided a great deal on Friday, and neither was written by a marketer. One was an 88-page opinion from a federal judge in Manhattan, which turned a decade of internal Google auction mechanics into questions a jury will now answer. The other was a file in a private code repository, which Apple devices call on at intervals to learn which advertising companies they are allowed to talk to. The first is public, numbered and appealable. The second has no published criteria, no notification process and no visible list of who is on it.

That contrast runs through most of the day's reporting. Pricing, counting, identity, inventory: in each case someone outside the advertising industry is now setting the terms, and the industry is finding out afterwards.

A federal judge sends Google's auction programmes to trial

Judge P. Kevin Castel of the US District Court for the Southern District of New York denied most of Google's summary judgment motions on September 30, in a ruling PPC Land detailed on October 2. The decision belongs to case 21-md-3010, the multidistrict proceeding that gathers private damages claims against Google's advertising stack. It is 88 pages long, and it clears the way for a jury to hear claims that publishers were systematically underpaid.

The damages figures are now approved for presentation. Daily Mail publishers may put $599.96 million in front of a jury; Gannett, which trades as USA TODAY Co., may put $900.77 million. A separate class of AdX publishers advances with a $1.72 billion overcharge claim, built on a damages model by Einer Elhauge calculated through March 2024. Expert economists Ali Hortacsu of the University of Chicago and Shengwu Li of Harvard were both permitted to testify, which matters more than it sounds: expert exclusion is how cases of this shape usually die.

What survives is a list of named programmes rather than a vague theory of harm. Enhanced Dynamic Allocation, running since 2014, let Google's exchange reach impressions that publishers had reserved for direct deals. Castel found evidence sufficient for a jury that the mechanism allowed AdX to cream-skim the most valuable of them. One piece of internal Google correspondence quoted in the ruling is unusually plain: "with EDA on we set the reserve price at $1, and any AdX buyer including GDN can win." Hortacsu calculated that Enhanced Dynamic Allocation alone cost Gannett $348.9 million domestically.

Project Bernanke, and its successor Global Bernanke, date from 2013. Internal documents compared the programme to conduct the authorities had already condemned. A February 2017 email stated that the company was "currently implementing a 'bidding ring' which the FTC has declared illegal." A memo cited in the ruling put the effect in numbers: AdX win rates rose from 37% to 44%, generating $290 million in additional revenue.

Minimum Bid to Win, live since 2019, had DFP tell AdX bidders the lowest winning bid after each auction, information rival exchanges did not receive. Li testified this depressed publisher prices and discouraged competitors from bidding. Project Poirot, which ran from 2019 to 2022, adjusted demand-side platform bids inside DV360 to cut spending on rival exchanges by 14.7% while raising AdX spending by 7.5%.

Google did win some ground. New York General Business Law consumer protection claims were dismissed against both Gannett and the Daily Mail, on the reasoning that neither is consumer-oriented: the court described arrangements for billions of impressions worth hundreds of millions of dollars as too complex to fit a statute written for consumers. Inform Inc., a video advertising company that operated between 2008 and 2017, lost its federal antitrust claims entirely, and with them a $479 million damages demand. The Progressive lost its AdSense claims for want of a sufficient connection to the product. Mikula Web Solutions, a single-employee business, kept its AdSense case and may proceed to a jury, which is a detail worth holding alongside the billion-dollar ones.

Google's mitigation defence partly survived too. Publishers argued they could not realistically have walked away from AdX, which accounted for 80% to 90% of their inventory. The court let the question stand: a jury may weigh whether price floors or other settings could have reduced the exposure.

Why does a summary judgment denial matter more than most procedural steps? Because it is the stage at which damages theories are tested against expert scrutiny, and because at least twelve other companies have filed follow-on suits citing the same practices. Raptive, Teads, OpenX and Dotdash Meredith have all filed on overlapping grounds, and the New York court cleared the path for private damages earlier in the sequence. The findings from the Virginia liability case of April 17, 2025 were given preclusive effect in New York on October 27, 2025, which is why the New York fight is about money rather than market definition. The Daily Mail's claimed damages exceed $380 million outside the United States, and the court held foreign revenue recoverable despite arguments under the Foreign Trade Antitrust Improvements Act.

For Gannett the claim now sits on the balance sheet as a gain contingency. For everyone else, the useful part is the paper trail: the mechanics of Enhanced Dynamic Allocation, Bernanke and Poirot were trade secrets for most of their operating lives, and litigation is the only process that has reliably produced numbers for them.

Apple widens the iOS 27 blocklist from five names to hundreds

What changed since the end of September is scope and method. When iOS 27 shipped roughly two weeks ago, the WebKit change blocked a short, nameable set of companies from serving on Safari: The Trade Desk, LiveRamp, ID5, Permutive and Audigent. PPC Land covered that narrow version on September 30, down to the eleven lines WebKit merged in February. On October 2, James Hercher reported that Apple replaced the short list with an expansive library of hundreds of customer data platforms, ad tech and martech companies, data sellers and identity graph operators. Two sources with direct knowledge of the WebKit updates described those entities as sitting in a purgatorial state of potential exclusion.

The method is the part that deserves attention. Rather than requiring an operating system update to change who is blocked, devices now regularly call on a remote list held in a private GitHub repository. Technical designations in the code include references to isRequestToKnownCrossSiteTracker and to fingerprinting classifications. In practice that means membership can change without a release note, a version number or any public record, and nobody outside Apple can enumerate the list. Whether Google properties including ad.doubleclick.net appear on the probationary list remains unconfirmed.

Allison Schiff set out the distinction from the 2021 AppTrackingTransparency framework in a column published the same day. AppTrackingTransparency made targeting harder and attribution less certain by gating the mobile advertising identifier behind a prompt, but, as she put it, "the pipes stayed connected" and companies adapted by modelling what they could no longer observe. Blocking adsrvr.org is a different category of problem, because there is nothing to model: the request does not leave the browser. There were published guidelines for AppTrackingTransparency, developer sessions explaining it and a compliance path. Here there is no stated criterion, no appeal route, and no clarity on whether a given block is deliberate or incidental.

Asymmetry is the live complaint. Ian Meyers reported that Google's bids went through while The Trade Desk's were blocked, which is the sort of observation that tends to end up in a regulatory file rather than a bug tracker. France, Italy and Germany have all examined the theory that Apple applies stricter standards to third parties than to its own services, and a class action in the United Kingdom seeking around 2 billion pounds makes the same argument. Safari 27 also blocks LinkedIn and Bing advertising trackers by IP address, which suggests the browser's pre-bid filtering of commercial domains is broader than any single dispute.

Set the two stories side by side. A court spent five years producing a public record of how one company's auction worked, and the remedy is a jury trial with numbered damages. Apple can change which hundreds of companies reach a browser by editing a file, and the affected parties learn about it from reporters.

SeatGeek prices seats one at a time and reports its own results

SeatGeek released TourIQ on October 2, a predictive pricing product for promoters and venues sold through its SeatGeekIQ division, covered by PPC Land the same day. The product does two things: it projects event revenue before an onsale, and it assigns prices to individual seats rather than to tiers. A companion product, PriceIQ, handles dynamic movement once tickets are live. Tour-level guardrails set minimum and maximum bounds across venues.

The data foundation is seventeen years of primary and resale transaction history, SeatGeek having launched in 2009. ArinMichelle Casey, head of entertainment partnerships, described it as "the largest blended dataset of primary and resale ticket transactions in live entertainment" and positioned the company as "an intelligence partner, not just a ticketing vendor." Noah Sarkey, quantitative research manager at SeatGeekIQ, framed the pitch as replacing instinct with evidence, noting that experienced bookers have always had a view of what a show should do but have lacked a way to prove it.

Three results were offered. At Nissan Stadium, for a football game between Ole Miss and Louisville, ticket volume ran 29% ahead of projection and gross revenue 13% above plan while the average ticket price came in 13% lower than forecast. A 2026 stadium tour opener reached 93% sell-through, reported as six percentage points above forecast, with 49% fewer unsold seat pockets after prices were redistributed; early sales had shown gaps in the lower bowl and slow movement upstairs. A third stadium show was modelled at roughly $1 million in additional revenue against tier-based pricing.

The disclosure gaps are substantial, and PPC Land itemised them. All three examples come from SeatGeek, with no independent audit. Only one names a venue. The $1 million figure is modelled potential rather than box office. The three major promoters said to be using the tool are unnamed, as are the artist and venue in the second example, and no total count of shows that have run through TourIQ was given, which means the published examples cannot be placed against the ones that were not published. The arithmetic in the first example also sits awkwardly: 29% more tickets at a 13% lower average price does not obviously produce 13% more gross.

The move from rate cards to per-unit price discovery is familiar territory for anyone who watched programmatic advertising go from fixed sheets to impression-level bidding. So is the regulatory tail. The FTC's Unfair or Deceptive Fees Rule took effect on May 12, 2025; StubHub paid a $10 million refund on April 9, 2026 over deceptive pricing; the UK Competition and Markets Authority opened investigations into StubHub and Viagogo on November 18, 2025; the FTC proposed an enforcement policy on undisclosed personalised pricing on August 19, 2026, having already probed eight companies over surveillance pricing; and the Senate held a surveillance pricing hearing on August 4, 2026. Seat-level pricing driven by a self-improving model is precisely the shape of thing those proceedings are circling, and the same debate has followed shopping AI.

YouTube removes Indonesian under-16 channels from view

YouTube will stop Indonesian users under 16 from uploading videos, creating Shorts, posting text or images, or leaving comments, PPC Land reported on October 2. The notice came from community manager Tammy Wi on the YouTube Help Community forum. The consequential line is retrospective: existing channels, videos and comments from under-16 Indonesian users will no longer be visible to other users.

The legal driver is Government Regulation No. 17 of 2025, known as PP Tunas, signed by President Prabowo Subianto, with enforcement beginning on March 28, 2026 after a one-year transition. It requires platforms classified as high-risk to set a minimum account age of 16. Ministerial Regulation No. 9 of 2026, issued on March 6, 2026, named eight such platforms: YouTube, TikTok, Facebook, Instagram, Threads, X, Bigo Live and Roblox. Lower-risk services may still admit users from 13.

The enforcement sequence is instructive. On March 30, 2026 Indonesia summoned Google and Meta, having found only X and Bigo Live compliant. An inspection on April 7 found YouTube non-compliant and produced a formal warning. Google submitted a compliance letter to the Digital Space Supervision Directorate on April 17, and on April 22 the ministry confirmed a 16-year minimum and a staged deactivation plan. Platforms filed risk self-assessments in June. By June 25, Communications and Digital Minister Meutya Hafid said YouTube had deactivated 600,000 under-16 accounts and TikTok 4.1 million. Google had argued on March 27 that blanket restrictions would eliminate the protections built into supervised accounts.

Scale explains why this is not a minor market adjustment. Indonesia held 3,000 of the 7,600 Southeast Asian YouTube channels above one million subscribers as of June 2025, and the country has roughly 70 million users under 16 with average daily screen time of seven to eight hours. YouTube had been building monetisation there, including Gifts powered by Crystals for Indonesian creators. The practical effect for buyers is a quiet reduction in contextual inventory, plus engagement metrics on videos with young Indonesian audiences that may decline retrospectively as comments vanish.

Several things the notice does not say are as significant as what it does. There is no start date beyond "segera", meaning soon. There is no statement on whether affected users can still sign in to watch. No age-verification method is described. Supervised accounts, mentioned in an April notice, go unaddressed. Indonesia has been an active regulator on adjacent fronts, having set 2% revenue fines under its data rules and previously blocked DuckDuckGo.

Google loosens advertiser business names from October 16

Google will allow some advertisers to display business names that do not match the destination domain, with the policy taking effect on October 16, 2026. Barry Schwartz published the change on October 2 after Arpan Banerjee shared the email notification.

Three conditions apply together. The business name must accurately reflect the recognised name or brand of the advertiser. A verified direct relationship must exist between advertiser and domain owner. The advertiser's products or services must be offered directly through the destination domain. Third-party resellers, independent booking intermediaries, affiliate distributors and secondary sellers are explicitly barred from the provision, and may not use standalone brand names of goods or services as their business names.

The exclusion list is the substance. Google has spent two years tightening the link between a visible advertiser name and a verified entity, from showing who actually pays for ads to extending identity verification to all advertisers. This change runs in the opposite direction for a defined set of brand owners while keeping the door shut on the intermediaries that the original rule was written to catch.

Two smaller interface items from the same morning sit in the same territory. Local Service Ads now hide the phone number on load behind a "Get phone number" button, with the number appearing on hover, which Schwartz documented with a recording while noting he could not establish whether it was new. That unit has already been folded into Google Ads with historical reports cut and had its reviews moved to Business Profiles, so the surface has been in motion for months. Separately, Arpan Banerje reported sponsored product carousels rendering without product images; Schwartz tested shortly afterwards, saw images display normally, and concluded a temporary bug was likelier than a test. Google issued no statement on either.

Google rewrites what counts as main content

The second documentation change of October 2 is the more consequential one. Google added two sections to its helpful content guidance, as Schwartz reported at 06:00 Eastern. The first defines main content as "any part of the webpage that directly helps the page achieve its purpose", and enumerates what qualifies: primary text, articles, images, audio or video; interactive features such as calculators, tools and games; user-generated contributions including reviews, forum posts and comments; tabbed or expanded sections; and page titles and headings. Google described main content quality as "one of the most critical factors for assessing page quality."

The second addition is a four-part framework familiar to anyone who has read the search quality rater guidelines: effort, originality, talent or skill, and accuracy. Effort is defined as "the extent to which human work went into creating the content or the systems powering it", a phrasing that explicitly admits automated systems provided a person built them. Originality asks whether the page offers "unique, original information or perspectives" unavailable elsewhere. Talent or skill asks whether the content demonstrates the expertise the topic requires. Accuracy carries heightened standards for topics that affect money or health.

Two features of the wording repay attention. Including user-generated contributions and interactive features inside main content means a page can fail the assessment on parts its publisher did not write. And counting "the systems powering it" as effort moves the test away from whether text was generated and towards whether anyone did work worth measuring, which is the same distinction underlying the scaled content abuse policy. Images and video now sit inside main content as well, which gives practical weight to housekeeping such as alt text and IPTC metadata.

The timing is not accidental. The same day Google inserted a line telling sites to manually factcheck and review all AI-generated content, with language explaining that generative models predict likely sequences of words rather than retrieving facts. Google has been assembling this apparatus in pieces: AI content evaluation criteria entered the rater guidelines, eleven pages on spam policies were added, and spam policies were extended to cover AI Overviews and AI Mode. The E-E-A-T vocabulary now has an effort axis bolted to it.

Here is the day's most instructive false alarm. Tom Pool posted on LinkedIn that links inside AI Overviews had begun carrying a parameter, st_source=ai_overview, and asked whether it marked "the start of actual proper attribution reporting?" Schwartz wrote it up on October 2 at 07:46 Eastern, with screenshots across several examples, and tried to replicate it in multiple browsers without success.

By midday the post carried two corrections. The parameter was not Google's. Pool traced it to a Chrome browser extension rewriting URLs to insert it, and offered the obvious reading: "Presumably to sell the data on to brokers..."

Two things are worth drawing out. The first is why the claim was believable for a morning. Publishers and SEO practitioners have no referrer-level way to separate traffic arriving from an AI answer from ordinary organic clicks, so a source parameter appearing in the wild looked like the missing measurement arriving quietly, which is how several genuine Google changes have in fact shipped. The second is the mechanism that produced it. A browser extension sitting between a user and the result page can rewrite links in a man-in-the-middle position, and the resulting data has commercial value; PPC Land has documented that ad blockers sell browsing data lawfully. An attribution signal that no platform publishes will be manufactured by somebody, and the somebody will not necessarily disclose it.

AudienceProject takes Pinterest measurement from two markets to nine

AudienceProject, the Copenhagen measurement firm, published a fourth-quarter roadmap on October 2 that extends Pinterest cross-media reach reporting from two countries to nine, set out by PPC Land the same day. France, Italy, Spain, Sweden, Denmark, Norway and Finland join the United Kingdom and Germany, where Pinterest measurement launched on June 18, 2026. Those nine mirror the firm's existing Netflix and Disney+ footprint. Poland, operational since March 2, 2026, is absent, as are Mexico and Canada.

Several changes are already live and matter more to day-to-day reporting than the map does. Reports can now span up to 98 days rather than 84, effective for reports beginning June 7, 2026, which allows a full quarterly campaign to sit in one document instead of being split across three. Reports may start and end on any weekday, matching flight dates rather than calendar grids. The Reports page now shows reporting period, channel, target group, reach, frequency and hitrate without opening individual campaigns, and the Items page gained five fields: tracking start date, campaign start and end dates, video completion rate, and clicks.

The roadmap items are structural. A persistent campaign entity will group related advertisements across channels, audiences and timeframes into what the company calls one clear source of truth, which is a deduplication problem before it is a reporting one. Bulk export will let several reports leave as a single structured Excel file. Australian availability is set for early 2027 with the Video Futures Collective, following partnerships with DAZN Media+ on September 3, 2026, and AudienceProject Next 3.0 is promised before December 31, 2026 without disclosed features. Worth noting: the previous fourth-quarter roadmap promised a Trade Desk integration, Polish expansion and a report builder redesign, all of which slipped into the first quarter of 2026 and went live between February and March.

Taboola puts a page view count on a television format

Taboola reported that Strictly Come Dancing coverage drew 2.2 million page views across its publisher network in a single week, reported by PPC Land on October 2. Readership peaked on Monday, September 28 at 680,000 page views, roughly 31% of the week's total. Emmerdale came second in the ranking at 527,000, which is where the quadruple comparison comes from.

Presenter-level figures cover a 90-day window. Emma Willis drew 1.7 million page views, described as a 7,302% rise from the prior 45 days. Josh Widdicombe drew 680,000 and Johannes Radebe 121,000, putting Willis at roughly 2.5 times Widdicombe and fourteen times Radebe. Dave Struzzi, communications lead at Taboola, said the numbers "prove once again that the popular dance show remains the nation's favourite fix for reality TV."

Against that, broadcast viewing fell. The first live show on September 26 averaged 4.8 million viewers and peaked at 5 million, around 23% below the prior year's 6.2 million average. The 24th series launch show had aired on September 19. So readership of articles about the programme is being offered as a measure of cultural pull at a moment when the programme itself is being watched by fewer people, which is a claim about where attention lives rather than how much of it there is.

The methodology limits are real, and PPC Land named them: no disclosed sample size, no way to separate a page load from an article actually read, mismatched comparison windows of 90 days against 45 for the presenter figures, and no prior-year readership baseline. The network in question includes The Independent, National World's sixty-plus titles, Reach titles such as the Express and the Daily Star, and HuffPost UK, which adopted Taboola's DeeperDive as AI cut into publisher clicks. The same series of releases has previously put numbers on Tesco Clubcard voucher coverage and a Ferrari launch. A supplier of recommendation widgets publishing readership league tables is doing something useful and something self-interested at once, and the numbers have no external audit.

Zoom runs at 99% unaided awareness and cannot use most of it

Josh Reed, vice-president of brand and content at Zoom, described the company's position at Programmatic IO NYC on September 28, in remarks Allison Schiff wrote up on October 2. Zoom carries 99% unaided brand awareness. Reed called it a gift, then immediately qualified it: "it's a gift, and it also isn't." Almost everyone knows the name, and almost everyone associates it with one pandemic-era behaviour.

He called it an iceberg problem. Above the water is video calling. Below it sit a contact centre product, an events platform that runs operations for the MLB Draft, a VoIP phone system and a set of AI productivity tools, none of which the awareness figure helps sell. High awareness of the wrong attribute is a harder starting point than low awareness, because the category association has to be displaced rather than built.

The organisational response came from Kim Storin, hired as chief marketing officer around eighteen months ago. Paid media had been running in silos across business units, each with separate measurement and attribution. That consolidated into a centre of excellence, and spending moved into linear television, creator partnerships, affiliate marketing and experiential work. Zoom's first creator programme launched in the fourth quarter of 2024 around office-humour content on LinkedIn; earlier in 2026 it introduced the Solopreneur 50, a recognition programme that drew thousands of applications for fifty places, with participants testing products and advocating for the brand.

Measurement is explicitly unfinished. Zoom still runs last-touch attribution and plans a move to multi-touch on Snowflake infrastructure, and the team has only recently moved from slide-based reporting to live dashboards. Reed's summary was blunt: "We're in the middle of our rebuilding year right now." He also described buyer research shifting from search engines to language models, and argued the company has to embrace what he called the dark funnel, the word-of-mouth, Reddit threads, podcast mentions and chatbot conversations that happen before any click exists to attribute. LinkedIn has been making an adjacent argument about building owned prominence rather than renting attention.

It is a fitting place to end a day like this one. A court spent 88 pages establishing what Google's auctions did, because the documents existed and could be compelled. Apple's blocklist cannot be read at all. SeatGeek, Taboola and AudienceProject each published numbers with no independent audit behind them. And a brand with the highest awareness figure in its category is rebuilding its measurement stack because the places its buyers now form opinions leave no record. The quantity of measurement in circulation keeps rising. The share of it that anyone outside the publishing party can verify does not.

Also noted

  • October 2: Google's loyalty programme documentation now lists 14 countries and newly describes AI Mode and Gemini integration, Merchant API support and eligibility for non-advertisers, catching the help pages up with the expansion announced in March and signalling a Merchant Center rollout; Search Engine Roundtable logged the changes.
  • October 2: PlayKids Learning, owned by London-based Sandbox Group, entered Disney+ Perks in Brazil with 230 hours of Portuguese-language content, 30 originals and 124 games, and joined Roku's premium storefront in Mexico with nine titles and 17 more from October 15, with no pricing disclosed for either deal.
  • October 2: Bing is testing a "Recommended by" label on product results, spotted by Sachin Patel, with no link to the recommending source and thumbnails pointing to retailers instead, leaving the endorsement impossible to verify.
  • October 2: Anthony Higman posted a screenshot of what appears to be a sitelink format inside AI Mode ads, which Schwartz published while saying the format looked off to him.
  • October 2: Digiday published its guide to what is in and out at Advertising Week New York 2026, ahead of a week in which agentic buying and AI measurement are expected to dominate panel titles.

By the numbers

  • $1.72bn - the AdX publisher class overcharge claim that Judge Castel cleared for a jury, built on Einer Elhauge's model through March 2024. Source
  • 29% - how far ahead of projection ticket volume ran at the Nissan Stadium game SeatGeek cites for TourIQ, on a 13% lower average price. Source
  • 600,000 - under-16 YouTube accounts Indonesia's minister said had been deactivated by June 25, against 4.1 million on TikTok. Source
  • 2.2 million - page views Taboola counted on Strictly Come Dancing coverage in one week, with Monday September 28 supplying 680,000. Source
  • October 15, 2026 - the date 17 further PlayKids titles begin arriving on Roku's Mexican premium storefront, after nine went live in September. Source