Antitrust law will not make Google pay for AI Overviews

I.An expectation is not an agreement

In an opinion dated 30 September and reported on 1 October, US District Judge Amit Mehta dismissed the antitrust suits that Penske Media and Chegg brought against Google over AI Overviews, the generated summaries that sit above search results. Chegg filed in February 2025; Penske, owner of Rolling Stone, Variety and Billboard, followed on 12 September 2025.

The publishers' theory was a bargain. For two decades, they argued, sites let Google crawl their pages in exchange for referral traffic, and AI Overviews keep the content while cutting the referrals. Mehta rejected the premise. The plaintiffs had pleaded only an expectation of traffic, he wrote, and “an expectation is not an agreement”: it is simply how a general search engine works.

According to Search Engine Journal's reading of the opinion, every claim failed. Reciprocal dealing failed for lack of an agreement. Tying failed because search and AI Overviews were not shown to be separate products. Monopoly maintenance failed on antitrust standing. Attempted monopolisation failed for want of plausible publishing markets. A California unjust-enrichment claim went once the federal claims fell.

The judge did not find that publishers were unharmed. He found that antitrust law is the wrong tool. He wrote that the court did not treat the harms lightly, and that the antitrust statutes are no substitute for legislatures addressing economic dislocation from new innovation, as FourWeekMBA reported.

II.The US closed a courtroom route as Europe opened a regulatory one

Outside the US, the direction runs the other way. On 3 June 2026 the UK Competition and Markets Authority imposed conduct requirements on Google that let publishers keep their content out of AI search features, opt out of model fine-tuning, and receive attribution links. Google has nine months to comply and must report every six months in the first year.

In Brussels, the European Commission opened an investigation in December 2025 into whether Google used publisher and YouTube content for AI Overviews and AI Mode without appropriate compensation, and whether publishers can refuse without losing search visibility. That case remains open.

Google had already moved on money. It announced an AI contribution pilot in June, and by September participating sites were seeing a payout panel in Search Console. The Information reported about 100 publishers enrolled, with one earning over $1 million a year and small sites earning under $1,000 over several months. Google has not disclosed how payments are calculated, Medianama noted.

The map now reads three ways. In the US, compensation runs on Google's voluntary terms unless Congress acts. In the UK, a statutory opt-out gives publishers leverage. In the EU, an abuse-of-dominance case is still live.

III.The honest read: a pleading loss, not a verdict on AI search

Scope matters. This was a motion to dismiss. The court tested whether the complaints stated antitrust claims, not whether AI Overviews are lawful in every respect. Search Engine Journal reports the dismissal was without prejudice and the order final and appealable, so the plaintiffs can go to the DC Circuit. TheWrap reports Penske could refile. Copyright theories were not before the court.

The evidence of harm is real but contested. Pew Research Center tracked 68,879 searches by 900 US adults in March 2025. Users clicked a traditional result in 8% of visits that showed an AI summary, against 15% without one. They clicked a link inside the summary in 1% of visits, and ended their session 26% of the time against 16%. Penske's complaint cited an Ahrefs estimate of a 34.5% drop in click-through and said about 20% of searches returning its links carried an AI Overview, per PPC Land.

The skeptic case comes from Google, and it is not empty. In August 2025 search head Liz Reid wrote that total organic clicks to websites were relatively stable year on year and that third-party studies relied on flawed methods. She published no figures to back the claim.

Chegg's decline also has more than one cause. When it sued, the company had lost about 90% of its market value since its 2013 listing, Tech Startups reported. Its May 2025 layoffs were attributed to students switching to ChatGPT and Gemini as well as to Google's search changes.

Both readings can hold at once. Aggregate clicks can stay flat while specific categories, such as reference, homework help and entertainment explainers, lose most of their search value. The court did not need to decide which, and it did not.

IV.The buyer now sets the price of the content it summarises

Market structure. With the US antitrust route closed for now, pricing power over AI-answer content sits with the dominant buyer. Google's pilot has one buyer, many sellers and no published formula. For smaller sites, payments amount to less than one-tenth of 1% of advertising revenue, AI Weekly reported, and declining payment does not stop Google using the content. That is a licensing market in name, with terms set unilaterally.

Capital. Google's search business is not shrinking while publishers lose referrals. Search and other revenue reached $63.3 billion in Q2 2026, up 17% year on year. Advertiser spend stays on the results page; the discount shows up in publisher valuations instead.

Labour. The adjustment lands at publishers, not at Google. Chegg cut 22% of staff in May 2025, after reporting 3.6 million subscribers in Q4 2024, down 21%, per Tech Startups. Writers and editors in reference and explainer content are most exposed.

Sizing, as an estimate. Take Penske's figure that 20% of searches returning its links carry an AI Overview, and Pew's drop in click rate from 15% to 8% on those searches. Search referrals then fall roughly 9%, all else equal (0.20 × 7/15). If AI Overview coverage of a site's queries reached 50%, the loss would be about 23%. These are illustrative numbers built on two assumptions. They ignore query mix and Google's stable-clicks claim.

Productivity and prices. Users get answers in fewer clicks, a real gain the court treated as product improvement. That surplus is not priced, so it flows to Google through ad inventory rather than to the sources the answers draw on.

V.Second-order: leverage moves to opt-outs, copyright and Congress

The opt-out becomes the bargaining chip. The UK rules let publishers withhold content from AI features without leaving search, which is the choice the US plaintiffs said they lacked. If uptake in the UK is meaningful, US publishers gain a working model to lobby for, and Google faces pressure to offer the same control globally rather than run two systems.

Copyright becomes the main US legal front. This ruling tested competition law only. Publishers' remaining claims over AI answers in US courts will turn on copyright and licensing, where the question is use of content rather than agreements over traffic.

Content strategy shifts toward what a summary cannot replace. Reid's own post said users increasingly choose forums, video, podcasts and original reviews. Reference pages lose most. Communities, events and first-hand reporting gain relative value.

VI.Precedent: the same court, the same answer

Mehta is the judge who found Google a search monopolist in 2024. In September 2025 he declined to order a Chrome divestiture in the remedies phase, writing that developments in AI had changed the course of the case.

In March 2026 he dismissed monopoly claims from local publishers Helena World Chronicle and Emmerich Newspapers, finding they lacked antitrust standing and had not shown monopoly power in online news. He also rejected their tying theory as a sale conditioned on reciprocal supply, not a purchase conditioned on another purchase.

Three rulings, one pattern. Google's monopoly in search is established law. Turning that monopoly into a claim for publishers' lost traffic has failed three times in the same courtroom. Any change in that pattern now has to come from the appeals court or from legislation.

VII.How it plays out: three scenarios

Probabilities are our analytical judgment, not forecasts from a model, and they sum to 100%.