In September 2025, a US federal judge rejected a bid to let publishers exclude their content from Google's AI products, citing insufficient evidence of harm. "The court heard evidence about Google's opt-out offerings, but no testimony from a single publisher," wrote US District Judge Amit Mehta. "The court does not doubt that publishers face new challenges because of GenAI technologies, but there can be no cure without evidence to support it."
That evidence has now surfaced. A new study by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon provides the first randomized behavioral test of how AI Overviews affect downstream publisher websites. The findings validate publisher complaints and undermine Google's central defense: the generated summaries drain traffic while delivering "no measurable improvements in perceived search quality or ease of finding information."
Though the US case has closed, the research arrives at a critical moment for regulators elsewhere. The European Commission, the UK's Competition and Markets Authority, Brazil's CADE and India's competition authority are all examining Google's AI search and its publisher impact. The study provides precisely the evidence these investigations require.
What the experiment found
The researchers recruited 1,065 US desktop Chrome users in early 2026 and asked them to install a browser extension for two weeks without knowing the study's purpose. One group had AI Overviews removed (with results shifted up to maintain normal page appearance), another was redirected to Google's Gemini-powered AI Mode for every search, and a control group saw Google unchanged. Only a single participant suspected the study involved AI Overviews, strengthening the findings' credibility.
- AI Overviews appeared organically on roughly 41% of searches.
- On those searches, outbound clicks to websites fell by about 40%, from 0.62 to 0.37 clicks per search. Zero-click searches rose from 54% to 73%.
- Traffic losses concentrated where the overview occupies top-of-page real estate and the top three organic results—the premium positions publishers compete hardest for.
- Citations within the overview generated only about 7% of outbound clicks on those searches.
- Ad clicks and clicks on Google's own pages remained unchanged, as did the number of searches users performed.
- Visitors who reached publisher sites spent the same amount of time and bounced at the same rate regardless of AI Overviews, indicating no better user experience.
- Users reported no superior search experience with AI Overviews than without them. The full AI Mode group clicked even less, with many participants dropping out or installing workarounds.
The study's design—no corporate funding, independent academics, rigorous methodology—matters as much as its findings. When Pew Research found in 2025 that people clicked traditional links 8% of the time with AI summaries versus 15% without, Google challenged the methodology to undermine it. This pre-registered study validates those earlier results through a method difficult to dismiss: randomizing what real people see in their own browsers, confirming no effect when overviews are absent, then reversing the groups to watch the effect reappear.
Google has countered falling publisher traffic by claiming AI sends fewer but higher-quality clicks. Last year, the search division's leader stated click volume was stable and click quality had risen, offering no evidence or independent audit. The new study tested this assertion and found nothing to support it—yet another reason why regulators must mandate data access and transparency to address the information imbalance between platform and publishers.
Same gatekeeper, more of the value
Following the 2024 US court ruling that Google illegally monopolizes general search, the pattern becomes clear: Google retains the attention it monetizes through ads while publishers lose visits they would have monetized through ads and subscriptions. Users gain nothing measurable in return. This mirrors the double bind documented in a separate report: the companies eroding publisher traffic also dictate the terms under which their content is used. Google alone, among AI companies, can impose this on publishers who cannot afford to leave its index. Separating its AI training from its search crawler addresses nothing about this structural problem.
The 7% citation figure exposes a deeper flaw in how policymakers and platforms treat attribution as a solution, as if a link in the AI summary restores the old bargain. It does not. Content performs its function on Google's page whether or not anyone clicks through. A behavioral study by Fehr Advice in Switzerland, replicated across several jurisdictions, found that users experienced a measurable decline in search value without news results. That study, commissioned by a Swiss publishers' association, reached the opposite conclusion from the new research but points to the same truth: news content makes search substantially more valuable to users.
Measuring these impacts requires enormous resources. The AI Overviews study cost around $30,000 for two weeks of testing, excluding months of browser extension development—all to measure something Google observes for every search it processes. The Fehr study incurred similar costs. Google almost certainly conducts such experiments internally without sharing results. Publishers cannot even isolate AI Overview traffic in Google's own tools, let alone run A/B testing.
Independent researchers should not have to reconstruct a dominant firm's internal data. Regulators must compel disclosure. The UK's CMA, which designated Google with "Strategic Market Status," requires the company to provide publishers with detailed user engagement metrics—including how often AI Overviews send users to publisher sites—beginning December 3. However, this individual publisher data remains unavailable to external independent researchers or industry bodies, limiting its utility for research and policymaking. The EU opened a formal antitrust inquiry into AI Overviews as a potential abuse of dominance, but the case continues. The DMA requires Google to share anonymized data only with rival search engines and AI chatbots starting in the new year, not with downstream sites.
The opt-out trap
One of the study's most significant policy implications emerged unintentionally: when users did not receive AI Overviews, downstream sites recovered roughly 40% of their lost clicks.
An opt-out mechanism works fundamentally differently. Under the UK Competition Authority's new rules and Google's announced controls, a publisher can exclude its content from AI Overviews while remaining in ordinary search results. But the overview itself does not vanish, nor can a publisher compete for top-of-page placement.
This creates the opt-out trap. Each publisher acting alone arguably gains more by staying in, while publishers acting collectively could recover far more by withholding at sufficient scale to degrade the overview. This collective-action problem explains why individual opt-outs leave Google's position unchanged and fail to rebalance the market. The exception is content with no substitute—original local reporting above all—where local news outlets hold leverage they have rarely deployed, particularly since many lack the technical expertise to block via robots.txt, especially given fear of retaliation in traditional search results.
US courts are closing the door but regulators need not
The judicial path in the US has stalled. Judge Mehta dismissed a local news class action by Helena World Chronicle and Emmerich Newspapers in March 2026, and in September dismissed antitrust suits by Penske Media and Chegg over AI Overviews, finding no formal agreement guaranteeing publisher traffic and treating search and AI Overviews as a single product. Each ruling acknowledged the harm while rejecting the legal theory. The judge overseeing the Google Search monopoly case expressed sympathy for publishers' predicament, arguing that Congress should act—a call the judge reiterated in his rulings.
The US represents only one jurisdiction, and the study arrives precisely when other regulators are investigating. The European Commission opened an inquiry in December 2025 into whether Google imposes unfair terms on publishers by using their content for AI Overviews without fair compensation or a genuine refusal option. The UK's CMA will eventually assess whether its opt-out actually works once its mandate takes effect, signaling further measures to ensure fairer value exchange. Brazil's CADE opened a formal proceeding in April into Google's use of news content, including in AI Overviews. Each body needs evidence of harm, of the value Google extracts relative to what it returns, and of whether opt-outs offer meaningful choice. This study provides all three, though US desktop data will carry less weight in Brussels or Brasília than local replications.
Policy recommendations
- Make AI use opt-in. Separate consent to indexing from consent to AI use, so the default no longer feeds the overview and the burden shifts to Google.
- Let publishers act together. Collective bargaining with an antitrust safe harbor, or a statutory licence with collective distribution, gives the long tail of local, ethnic and non-English outlets the leverage no single publisher possesses.
- Pay for use. Licensing rates should rest on how much content AI products consume, with clicks as a floor.
- Mandate disclosure. Domain-level data on AI Overview impressions, citations and clicks, reported to publishers and regulators.
- Fund independent evidence beyond the US and across devices. Replications in Brazil, Canada, the EU, India and the UK would provide the evidence competition authorities seek. Budget remains an obstacle. A publicly available version of the extension would help, but does not address the need to pay study participants.
The evidence is now clear: Google's AI Overviews harm downstream traffic while delivering no measurable user benefit. Google's Search dominance has already been litigated. The question now is whether lawmakers and regulators will act before publisher challenges become insurmountable and market imbalance deepens further.



