Search traffic was never a contract

Search referral traffic is a by-product of how a search engine chooses to present results, and on October 1, 2026 a US federal court said, in effect, that it is nothing more than that.

US District Judge Amit Mehta dismissed the antitrust suit Penske Media Corporation brought against Google over AI Overviews, along with a similar case from Chegg (Press Gazette). The line from the opinion that will be quoted for years is short: publishers had pleaded an expectation that Google would send them traffic, and "an expectation is not an agreement" (TheWrap via Yahoo).

This post is not legal analysis. It is about what the ruling means for anyone who reports on, plans around, or budgets against visibility in Google's AI features.

#What the case argued and why it failed

Penske filed in the US District Court for the District of Columbia on September 12, 2025, joined by 14 subsidiaries including Rolling Stone, Variety, Billboard and The Hollywood Reporter (PPC Land). The core theory was a de facto bargain: publishers let Google crawl their content, Google sends traffic back, and AI Overviews broke that exchange by answering on the results page. The complaint said Penske's organic affiliate revenue had fallen by more than a third from its peak to the end of 2024 (Press Gazette).

The court did not dispute that harm. It found that the publishers had not pleaded any agreement in which Google promised to deliver a specific amount of traffic, or any traffic at all, in return for content (Press Gazette). The other theories fell too. The tying claim failed because search and AI Overviews were not treated as separate products with distinct demand, and the monopoly maintenance claims failed on standing; with the federal claims gone, the court declined to hear the California unjust enrichment claims (Relevant Audience).

The dismissal was without prejudice, so Penske can refile (TheWrap via Yahoo). The opinion also says the court does not treat the alleged harms lightly (Relevant Audience). Sympathy is not a remedy, though, and for now the US position is clear enough to plan around.

#The terms are set by product decisions

If referral traffic is "simply how a general search engine works" (TheWrap via Yahoo), then the terms of that traffic are whatever the current product does. That has practical consequences, because the product changes often and without notice.

The size of the effect is already measured. Ahrefs compared 300,000 keywords and found that an AI Overview on the page correlated with a 34.5% lower average click-through rate for the top-ranking result, against similar informational keywords without one (Ahrefs). Nothing in the US ruling obliges Google to change that. The layout of the results page, the number of links in an AI answer, and where those links point are design choices, and design choices are revisited every quarter.

This is the right frame for a visibility programme. A citation in an AI Overview is not an entitlement earned by ranking well. It is an outcome of a system you do not control, under rules that can be rewritten at the next release.

#Where the terms are written down

The contrast with the UK is useful because it shows what binding terms actually look like. On June 3, 2026 the Competition and Markets Authority ordered Google to give publishers a way to keep their content out of AI Overviews, AI Mode and AI features in Discover without losing ordinary search visibility, to attribute publisher content with clear links in AI results, and to let publishers opt out of their content being used for AI fine-tuning, with nine months to implement (SiliconANGLE). Google also said it would build a dashboard showing which pages appear in AI results (SiliconANGLE).

Those are obligations, not expectations. Some of them have already reached everyone: the Search Console generative AI control rolled out to all properties worldwide on August 31, 2026 (Search Engine Land). The lesson for a US-based team is that the controls you get were shaped by a regulator elsewhere, and the next round of controls (page-level grounding, attribution rules, reporting) will probably arrive the same way.

#What to do with this

Treat AI referral share as a variable, not a baseline. Forecasts that assume today's click-through from AI Overviews will hold are forecasting a product decision. Model at least one scenario in which AI answers send materially fewer clicks than they do now, and know which pages and revenue lines are exposed if it happens.

Measure dependence on a single surface. Report what share of organic sessions, leads or affiliate revenue comes from Google Search, and how much of that runs through queries that now show AI features. Penske's affiliate loss (Press Gazette) was concentrated in exactly the kind of content that AI answers summarise well.

Use the controls that exist, deliberately. The generative AI control, nosnippet, and Google-Extended do different things (Search Console Help). Write down which one you use, why, and when you last reviewed it. If the ruling tells you anything, it is that silence is read as consent to the default.

Watch the jurisdictions where rules are being written. Track CMA implementation deadlines and Google's compliance announcements as roadmap inputs. Features built for UK compliance have a habit of shipping globally.

Put contracts where contracts are possible. If your content has licensing value, the place to capture it is a negotiated agreement with explicit terms, not an assumed exchange with a search engine. Where no such deal is on offer, the value you control is the value you can measure: being cited accurately, being linked, and being chosen when a user does click through.

Keep measuring what the answers actually say. A citation is the one part of this arrangement you can observe directly. Log which engines cite you, for which queries, with what text and which link, and date every measurement so a product change shows up as a step in the data rather than a mystery in next quarter's report.

The ruling does not change how AI Overviews work. It removes one assumption: that the traffic was owed. Planning without that assumption is less comfortable and more accurate.