Three documents landed within thirty hours of each other, and together they describe the same gap from three directions. A federal judge in Washington threw out a publisher's antitrust claim against Google because no contract to deliver traffic had ever existed. A coalition of seven news organisations published the first version of a technical standard designed to produce the records that such a contract would need. And in Sacramento, a governor removed a cause of action that thousands of plaintiffs had been using to sue websites over the tracking code those websites had installed. The courts, the standards bodies and the legislatures were all working on the same unresolved question this week: when content moves between parties without a signed agreement, who is owed what, and who gets to count it.

A federal judge rules that an expectation is not an agreement

Judge Amit P. Mehta dismissed Penske Media Corporation's antitrust case against Google, in a ruling reported on October 1 at 11:31 Eastern time. The decision turned on a contractual question rather than a competitive one. Penske had argued that Google forced publishers into a bargain in which content was supplied in exchange for referral traffic, and that AI Overviews broke that bargain by answering queries without sending visitors onward. Mehta found the premise unproven as a matter of pleading. Publishers, he wrote, "failed to plead any actual agreement whereby Defendants promised to 'sell' Plaintiffs any specific amount of traffic." The reciprocal dealing theory collapsed on a single sentence: "an expectation is not an agreement."

The case was filed on September 12, 2025 in the US District Court for the District of Columbia, naming Google LLC and Alphabet Inc. as defendants. Penske brought it alongside fourteen subsidiary publications, among them Rolling Stone, Variety, Billboard, The Hollywood Reporter and Deadline. The 101-page complaint leaned on market-share figures of 89.2% across search and 94.9% on mobile, on the C4 dataset supplying 12.5% of LaMDA's training data, and on an Ahrefs analysis putting the click-through rate reduction from AI Overviews at 34.5%. Penske also stated that roughly 20% of searches returning its links carried an AI Overview during late 2024 and early 2025, and valued the AI content licensing market at $30 billion by 2034. None of those numbers were what the ruling turned on.

Mehta is the same judge who decided the search remedies in the government's monopolisation case, and his dismissal was not an endorsement of the arrangement he declined to police. He acknowledged "the knock-on consequences to journalists, educators, and other online creators whose content Google takes and repurposes without compensation." That is an unusual sentence to find in a dismissal. It concedes the harm and declines the remedy, on the ground that the remedy Penske asked for required a promise nobody had made.

Jason Kint, chief executive of Digital Content Next, put the trade body's position bluntly after the ruling: "Google is using its dominance to fundamentally undermine the open web." The argument now has nowhere obvious to go inside this particular case. Publishers have spent the past year pursuing the same grievance through legislation, through regulators, through the European Commission and through bilateral licensing. Mehta's reasoning explains why the contractual route was always the weakest of them. The web's central economic relationship was never written down.

SPUR publishes version 1.0 and invites the five companies it was built to measure

The absence Mehta identified is precisely what the Standards for Publisher Usage Rights coalition spent six months trying to fill. SPUR released the version 1.0 of its content telemetry standard on October 2, and pitched OpenAI, Anthropic, Google, Meta and Microsoft to join an AI licensing advisory board. The group was formed in March 2026 and counts The Guardian, the Financial Times, the BBC, Sky, The Times of London, MediaHaus and the Associated Press among its members.

The standard describes a reporting chain rather than a block. It defines processes for recording when content is retrieved, grounded, cited, presented and engaged with inside an AI system, and for reporting that activity back to the publisher that produced it. The specification covers multimodal content rather than text alone, and is designed to sit alongside existing provenance work including C2PA. A draft went out in June 2026, public comment ran through July 24, and the first technical committee meeting is scheduled for October 15. The advisory board is expected to run to roughly twenty members, with a first meeting planned for this month.

Alex Springer, SPUR's technical lead, described the ambition as assembling the pieces so that adoption is not a research project for each AI company in turn: "The goal is to string this whole thing together into a stack of" licensing and reporting solutions. He was candid about where the large model providers currently stand. "By no means did OpenAI come and comment on our standard. They are aware of it," Springer said. David Buttle, a SPUR co-founder, framed the advisory board as the mechanism for making the thing bilateral: "This will ensure both sides of the equation have a seat at the table and can work together to drive through better, fairer and more sustainable" practices. Michael Rubenstein, co-founder and co-chief executive of Firsthand, tied the whole exercise back to supply: content quality depends on publishers building "a sustainable revenue stream out of this, and continue to invest in premium content."

The sequencing is worth noting. A standard that reports usage is the artefact a licensing market needs before prices can be argued about, and the IAB has its own measurement framework due in November covering the advertiser side of the same visibility problem. SPUR has published a way to count. It has not published a buyer. Pilot programmes with technology and AI companies are described as the next phase, with no named participants and no dates.

Newsom removes a cause of action as filings approach 4,000

California moved in the opposite direction on the same underlying question of who may enforce a rule. Governor Gavin Newsom signed Senate Bill 690 on September 30, the final day of his constitutional deadline, eliminating the private right of action under California's pen register statute at Penal Code Section 638.51. From January 1, 2027, only the Attorney General may bring those claims against website and app operators. The change applies retroactively to pending cases filed since approximately January 1, 2025.

The scale of the litigation it ends is the reason the bill moved. Close to 4,000 claims were filed after the measure's introduction in February 2025, with law firms citing a range from 600 to 4,000 for the surge, and thousands of demand letters arriving alongside the filings. The statutory damages formula of $5,000 per violation is unchanged for the CIPA sections that keep their private right of action, which include 631, 632 and 632.7. A plaintiff suing over a session-recording script or an analytics pixel now has a narrower route; a plaintiff suing over interception has the same one as before.

The legislative record shows no recorded opposition at any stage. Senator Anna Caballero filed the original and broader version on February 21, 2025. The Senate passed it 35-0 on June 3, 2025. An Assembly committee narrowed its provisions on July 1, 2026, the Assembly passed the narrowed text 66-0 on August 28, 2026, the Senate concurred 40-0 the same day, and the bill was enrolled on August 31 and presented to the governor on September 4. Newsom's signing message described the problem as the "vexatious use of CIPA lawsuits" against small businesses that had "unwittingly" installed tracking software, while conceding that the software "at times tracked and shared" visitor information. He asked the Legislature to return to the remaining CIPA provisions in 2027.

Enforcement now sits with Rob Bonta's office, which is a different institution from the California Privacy Protection Agency operating under CCPA authority. Two other statutes take effect on the same January 1, 2027 date: SB 923 extends CCPA deletion rights and requires web forms, and AB 566 mandates that browsers be capable of sending an opt-out preference signal, which in practice means Global Privacy Control. Newsom had separately signed AB 1709 on September 10, 2026, limiting addictive feed features for users under sixteen with penalties up to $50,000, also enforceable only by public authorities. The pattern across all four is consistent. California is keeping the obligations and reassigning who may sue over them.

A Lagos court reaches the opposite conclusion about private enforcement

While Sacramento was removing a private remedy, the High Court of Lagos State was granting one. Justice A.F. Pokanu delivered judgment on September 25, 2026 in case IKD/21116MFHR/2025, finding that Meta violated Nigerian users' privacy rights through behavioural advertising conducted without valid consent. The decision awarded the equivalent of $100,000 in naira in general damages, plus one million naira in costs, and gave the company eight weeks to comply, a deadline that falls on November 20, 2026.

The applicants were the Incorporated Trustees of Laws and Rights Awareness Initiative together with five individuals: Deborah Esther Orji, Abayomi Olakunle Adebayo, Olamijulo Ayomide Ogunkilede, Kitan Bankole and Abiola Owoaje. Olumide Babalola and H. Tofi appeared for them. Mofesomo Tayo-Oyetibo SAN and Lukwagh Mgbanyi appeared for Meta. The originating application was dated December 21, 2025 and the suit filed on January 14, 2026 in the Ikorodu Judicial Division.

The court held that Meta breached Section 37 of Nigeria's Constitution along with several provisions of the Nigeria Data Protection Act 2023, and rejected the company's argument that behavioural advertising, profiling and cross-platform tracking were necessary to perform its contract with users. That is the same contractual-necessity defence that has failed repeatedly in European proceedings, including the CJEU ruling that narrowed Meta's permitted data use and the EDPB order covering the entire EEA. The applicants had sought exemplary damages and a far larger general award; the court granted roughly a thousandth of the naira sum requested and refused the exemplary claim outright.

The figure reads small next to the regulatory track in the same country. Nigeria's Data Protection Commission found more than sixty million Nigerian Meta users affected by its investigation, issued Final Orders on February 18, 2025, and settled with the company for a $32.8 million remedial fee agreed on October 30, 2025 and entered as a consent judgment on November 3, 2025. A regulator extracted thirty-two million dollars. Six private claimants extracted a hundred thousand. The difference is instructive for anyone trying to work out which enforcement channel actually moves money, and it is the difference SB 690 has now written into California law from the opposite direction.

Resellers took 81% of every new ads.txt line added in September

Away from the courts, the plumbing produced one of the cleaner reversals of the quarter. DataBeat's September 2026 Sellers Report, released on September 18 and covering ads.txt files at the top 50,000 US publishers, recorded 54,942 net new lines. Of those, 44,608 were reseller declarations and 10,334 were direct, a ratio of roughly 4.3 to one. Resellers took 81% of the net additions; direct relationships took 19%.

The churn underneath that net figure is larger than the figure itself. Publishers added 459,733 lines gross and removed 404,791. Low-traffic sites ranked between 2,001 and 50,000 drove most of it, accounting for 34,738 net lines, 64.5% of additions and 64.6% of removals. Mid-traffic properties ranked 501 to 2,000 added 10,522 net, and the high-traffic top 500 added 9,682. Five supply-side platforms supplied 30,397 lines between them, or 55% of net growth: Nativo, marketed in those files as "Life360 Ads Featuring Nativo", with 9,814; Risecodes with 6,856; Media.net with 4,985; Sharethrough with 4,654; and InMobi with 4,088. Life360 agreed to acquire Nativo for approximately $120 million on November 10, 2025, and the branding of those declarations is the acquisition showing up in the authorisation layer.

Duplication held roughly steady. Tier 1 supply-side platform duplication stayed at 46% against June, with an average intermediary count of 1.33. Tier 2 rose one point to 39%, which DataBeat flagged as "a small move worth watching", and Tier 3 rose two points to 36%. The headline reversal is against August, when net change was close to zero as publishers trimmed reseller paths. DataBeat attributed September's swing to buyers appearing to seek more reach, a reading the report cannot support, because it contains no spend data at all. The firm also sells a competitive intelligence tool addressed to the duplication it measures, a conflict PPC Land noted in its own coverage.

The context matters more than the monthly direction. ads.txt turned ten this year, Brian O'Kelley proposed adagents.json as a replacement on March 30, 2026, and the IAB Tech Lab has been writing rules to curb duplicate bid requests across the OpenRTB pipe. A file designed to let buyers verify who may sell a publisher's inventory is being used, at the margin, to add more sellers.

OpenWeb enters insolvency in Tel Aviv over roughly $20 million

OpenWeb, the publisher engagement and advertising company once valued at $1.5 billion, entered insolvency proceedings in Israel this week. The Tel Aviv District Court declared the company insolvent and appointed a temporary receiver, acting on a petition from Mars Growth Capital, part of Liquidity Group, which is seeking to enforce liens over approximately $20 million of outstanding debt. The dispute with the lender sharpened a cash shortage that already existed. The company works with hundreds of premium publishers and had previously raised hundreds of millions of dollars.

Twenty million dollars is a small number for a company that carried a billion-and-a-half-dollar valuation. The gap between those two figures is the whole story of the private end of the publisher technology market in 2026. OpenWeb's product sat in the part of a publisher's page that monetises attention after the click has already happened, which is the part of the page most exposed to the referral declines that Penske's complaint tried to litigate. A conversation layer is worth building when audiences arrive in volume. The economics of one change when they do not.

AI creative production moves into the upper middle of the market

The clearest commercial pattern in the past two days came from advertisers well below the largest spend tiers. Digiday reported on October 2 that scaled AI creative production has become ordinary among companies with annual revenues under $1 billion, and the examples carried numbers rather than intentions.

Wyndham Hotels and Resorts, with a $565 million marketing budget, has worked with the AI firm Adora for a year, generating assets for paid Instagram and owned channels. Mike Shiwdin, group vice president of loyalty and guest engagement, described fifteen times the previous volume of assets with production timelines cut by 75%, and said performance followed: "We've seen a six times improvement in site visit rate over our benchmark." Opella, spun out of Sanofi in 2025, runs an 85-person team on generative tools that produce strategy briefs, assets and animated brand mascots, including an owl named Nigel for the allergy treatment Xyzal. Julien Leterrier, global head of martech, put the output at more than 20,000 pieces of content and said the team had "produced 20 times" its earlier creative volume.

Not everyone is handing over the final cut. BetMGM, with $160 million in annual media spend and a 120-person in-house creative department, uses Arcana AI for imagery and video while keeping human oversight on regulated content. Victor Gonzalez, senior director of creative, was direct about why: "We haven't found that trust in the machine yet." Will Hanschell, chief executive of Pencil, said the fastest-growing clients on the platform are the ones under $1 billion in revenue, because "they culturally move faster."

Three supporting figures explain the timing. Token pricing has fallen 41% from its March 2026 peak, which is the same input cost that drove agencies to meter AI buying by the token earlier this year. A Dentsu survey found 81% of chief marketing officers expecting to need significantly more content. Kantar recorded a 17% increase in creative tests between December 2025 and July 2026, which is what rising volume looks like on the measurement side. The tooling has arrived in parallel from the platforms themselves, through Google's Asset Studio and Adobe's GenStudio expansion.

A countercurrent appeared on the same day. Red Antler Group launched Fat Earth Studios, a creator content operation built explicitly to answer AI-generated material with user-generated and influencer content. JB Osborne of Red Antler and John Gross of Fat Earth spent a year developing it inside the group's performance marketing division. The studio matches clients to creators, briefs them on messaging and distributes across social channels, with deliberately loose creative direction. Fifteen times the assets at a quarter of the timeline is one answer to a saturated feed. Paying humans to look unlike everything else in it is another, and both bets are now being placed by the same cohort of advertisers.

Warner Bros. Discovery plugs into Prisma Direct as its second media partner

Mediaocean added Warner Bros. Discovery to Prisma Direct on September 30, making it the second media company on a platform that launched with Disney on March 31, 2026. Prisma Direct integrates five transaction stages that premium media buying still largely handles by hand: ordering, trafficking, campaign analytics, billing and reconciliation. The initial connection targets Neo, the Warner Bros. Discovery sales platform launched on May 14, 2025, with a fuller integration covering all transaction types promised without a date.

Drew Kane, chief product officer at Prisma, supplied the only quotation in the announcement: "Premium media represent some of the most valuable inventory in advertising, yet buying it still relies on too many manual, fragmented workflows." Prisma carries more than 100,000 users and roughly $200 billion in annual advertising spend through Mediaocean's systems, which is also the asset its venture arm offers to AI startups. Paramount's own research found 56% of marketers naming platform fragmentation as their top concern, a figure that reads as pre-justification for exactly this kind of integration.

What the announcement omitted is the interesting part. No Warner Bros. Discovery networks were listed, no pricing disclosed, no go-live dates given, and nothing was said about what happens to the integration if Paramount completes its acquisition. That transaction was announced on February 27, 2026 at $31 per share, received EU approval on July 22, 2026 subject to a UIP divestiture, and carries a ticking fee of $0.25 per share per quarter for Warner Bros. Discovery shareholders while closing remains uncertain, a timetable Paramount itself described as unsettled on September 25. Warner Bros. Discovery holds roughly 1.5% of the US connected TV advertising market on its own; a combined portfolio would clear 5%. The workflow plumbing is being connected to an entity whose shape is not yet fixed, which is a reasonable thing for a buying platform to do and an awkward thing for a buyer to plan around.

Amazon gives sellers a round-the-clock agent, as 8% of Americans report using one

Amazon's Seller Assistant gained continuous pricing and restocking workflows, announced on September 23 during Accelerate 2026 in Seattle, which ran September 22 to 24. More than 90% of selling partners worldwide now reach the assistant in their native language, usage runs to hundreds of thousands of active sellers against 230,000 monthly users in 2025, and Amazon puts the acceptance rate for its recommendations above 90%. The company also states that more than 90% of selling partners already use third-party AI tools, which is the competitive fact underneath the launch. Four prebuilt automation templates ship with it, a Claude plugin sets up in roughly sixty seconds without code, and a Quick Plus subscription valued at $720 for twelve months, covering three annual seats at $240 each, is open for sign-up until December 31, 2026.

The architecture runs on Amazon Bedrock with Claude models, retains pricing patterns, inventory cycles and growth goals, and persists that memory across Seller Central, Amazon Quick and Claude. Mary Beth Westmoreland, vice president of worldwide selling partner experience, framed the shift in one line: "Sellers don't just want AI that works with them. They want AI that works for them." Jigar Thakkar, vice president of agentic AI for business at AWS, described a system "continuously learning their preferences to give them answers and take actions that reflect the full picture of their business." Eric Burns, field chief technology officer at Anthropic, noted that the assistant already runs on Claude models. James Oliver, founder of Atlas Bar, offered the practitioner version: "What took an afternoon now happens instantly."

Jon Elder, an adviser to Amazon and Walmart sellers, dissented on full automation: "This is a recipe for disaster." Amazon's own materials do not fully resolve the question he is raising. The forum summary suggests sellers choose between an assistant that recommends and one that acts, while the longer post says actions are reviewed and approved before being carried out, which describes a queue rather than unattended operation. Approval timing, delivery method and the handling of time-sensitive repricing are unspecified, as are post-beta pricing, memory retention, whether the plugin uses MCP or a proprietary protocol, and how behaviour changes during promotional events. Prime Big Deal Days falls on October 6 and 7, which is soon enough for that last gap to matter. Amazon's agent policy took effect on March 4, 2026.

Consumer willingness to delegate sits well behind the tooling. Channel V Media's survey of 7,675 US adults, fielded in June 2026 with Prosper Insights and Analytics, found that 40% would hand at least one everyday task to AI while 8% currently use agentic AI. Only 18% called AI agents a good idea and 41% were unsure. The specific tasks people would delegate are small: restaurant booking at 12.1%, grocery shopping at 11.0%, smart home management at 10.0%, teaching children at 6% and car purchasing at 5%. Generative AI use by generation ran 47% among millennials, 44% Gen Z, 42% Gen X and 32% boomers. Sixty per cent were very or extremely concerned about AI privacy, 40% worried about wrong information and 29% distrusted AI's motives. Preference for a human channel held above two-thirds everywhere measured, at 84% in healthcare, 83% in banking, 78% in travel, 73% in telecoms and entertainment and 69% in online shopping. Executives reported 18% agentic use against 7% for employees.

The release publishes no margin of error, no question wording and no response options, and its own figures for current agentic use differ between the pitch email at 7% and the release at 8%. Channel V Media is a public relations agency that sells AI visibility services, which is worth holding in view when reading a survey about AI adoption. The direction nonetheless matches independent work: Zeta found 43% of parents letting AI shop to a set budget, and RTB House found AI lengthening purchase decisions for 42% of US shoppers. Willingness in principle is wide. Practice is narrow, and concentrated in low-stakes errands.

Google tells publishers to factcheck AI text, and Microsoft opens up experiments

Google rewrote its guidance on generative AI content on October 1, inserting a sentence it had not used before: "It is critical to manually factcheck and review all AI-generated content" before publishing. The document went on to explain the mechanism, in language closer to a technical note than a policy: "generative models don't retrieve facts, but predict a likely sequence of words based on their training data. Because of this, generative AI outputs may contain inaccuracies (also known as hallucinations)." The page had last been updated in October 2025, and Google said the change brought the documentation in line with presentations given at its developer events. Barry Schwartz observed that "critical" is a word Google rarely reaches for in its documentation.

The timing invites a reading. Google has spent the autumn running a spam update, and the gap between permitting AI-assisted publishing and penalising its output has never been documented as precisely as site owners would like. A sentence instructing manual review of every generated passage does not change the ranking systems. It does change what a publisher can later claim about process.

Microsoft Advertising, meanwhile, made optimization experiments generally available across Search, Shopping, Audience and Performance Max campaigns on October 1. The feature builds a treatment campaign from an eligible control campaign, splits budget and traffic between the two, and reports both on a single results page, with the option to apply a winning treatment to the original campaign or spin it into a new one. Controlled A/B tests of bidding, targeting and creative are not a new idea in paid search, and that is the point: the gap being closed is a parity gap, arriving after Microsoft extended AI Max to all accounts globally and added customer goals across Performance Max. As automation absorbs more of the levers, the ability to hold one campaign still while another moves becomes the only way to attribute a change to a decision.

That is the thread running from Mehta's courtroom to a Microsoft results page. Every party this week was trying to establish a record: of what was promised, of what was used, of who may sue, of what changed and why. SPUR has a specification and no signatories. Penske has a grievance and no contract. DataBeat has a count and no spend data. Channel V Media has a percentage and no margin of error. The documents that would settle these arguments mostly still do not exist, and the week's clearest lesson is that the parties who will eventually write them are not yet at the same table.

Also noted

  • October 1: Hallmark Media and iSpot agreed an outcomes-based measurement partnership covering Hallmark's non-linear environments ahead of the Countdown to Christmas launch on October 16, linking cross-platform exposure to site visits, box-office sales and in-store purchases, with Casey Gould and Stuart Schwartzapfel named. Adweek
  • October 1: Sallie, a sister site of the student loan lender Sallie Mae, began selling onsite placements through Backpack Media, the retail media unit launched in March 2026 and led by former Sephora executive Marco Steinsieck, with Munif Jaafar as senior director of strategy and revenue. Adweek
  • October 1: Paula Despins, vice president of measurement for Amazon Ads, argued that attribution should act as auditor and delegator for AI models rather than be displaced by them, citing long-term sales projections over twelve months and accumulated sales tracking against them. AdExchanger
  • October 1: Google Local Service Ads were reported serving leads from the wrong locations, a delivery bug affecting advertisers whose Google Business Profile service areas are tightly drawn. Search Engine Roundtable
  • October 1: Google tested removing the I'm Feeling Lucky button from the search bar, one of the oldest surviving elements of the homepage. Search Engine Roundtable

By the numbers

  • 4,000 Claims filed under California's pen register section since February 2025, a route Newsom has now closed to private plaintiffs. PPC Land
  • 44,608 Reseller lines added to US publisher ads.txt files in September, against 10,334 direct. PPC Land
  • $100,000 General damages a Lagos judge set against Meta over consentless behavioural advertising. PPC Land
  • 8% Share of US adults already using agentic AI, against 40% who say they would delegate a task. PPC Land
  • November 20, 2026 Date by which Meta must bring its Nigerian data practices into compliance. PPC Land