AI Search Traffic: Google Referrals to Publishers Fall 40%

Google is becoming an answer engine

For 25 years, publishers and Google had a simple deal. Publishers let Google index their work, and Google sent readers back to them. By late 2026, the evidence shows that deal breaking down. Chartbeat's annual publisher playbook found that Google Search referrals to hundreds of publishers fell 40% between July 2025 and July 2026, which is twice the decline of the year before.23 That figure has given new weight to an industry worry called "Google Zero": the idea that Google will stop working as a doorway to the web and become a destination that answers questions itself.23

Three things happened over the past month. Digital Content Next published a review of the research and concluded that AI-generated answers reduce clicks.22 The Reuters Institute described newsrooms rebuilding their strategies for what comes after search engine optimization (SEO).23 And a federal judge said that publishers' losses, while real, are not something antitrust law can fix.21 Taken together, the coverage points to a conclusion that is hard to avoid: the traffic is not coming back in its old form, and the industry is changing course.

Different studies, same answer

The research agrees to an unusual degree, even though the studies use very different methods.

• A randomized experiment. Researchers at the Indian School of Business and Carnegie Mellon followed more than 1,000 Google users. When an AI Overview appeared, clicks out to regular search results fell 39.8% and searches ending with no click rose 34.5%.22
• Real browsing behavior. Pew Research Center looked at nearly 69,000 searches. People clicked a traditional result 8% of the time when an AI summary appeared, compared with 15% when none did. Links inside the summaries got clicked only 1% of the time.22
• Keyword data. Ahrefs studied 300,000 keywords and estimated that an AI Overview cuts clicks to the top-ranked result by 58%. Clicks fell 51% for the second result and 46% for the third.22
• Google's newer AI Mode. A preregistered experiment from the University of Pennsylvania and Northeastern found that AI Mode, Google's chatbot-style search, cut outside clicks by 18.8 percentage points. It also made users more likely to switch to rival search engines.10

The Ahrefs number has grown over time. The same method showed a 34.5% drop in April 2025, almost half the current figure, which suggests the effect is getting stronger as the feature spreads.1 It is spreading fast. Similarweb found AI Overviews on 43% of Google searches worldwide in May 2026, up from 16% in January 2025.27

The numbers are not all the same kind of measurement, and they get mixed up often. Ahrefs measures how often people click the top-ranked page on queries that trigger an AI Overview. That is not a site's total traffic loss, and some coverage has wrongly described it as an average traffic drop.8 Researchers who cite the Chartbeat referral data also warn that algorithm changes, publishers' own choices and shifting audience habits explain part of the decline, not AI Overviews alone.22 Those caveats matter, but they change how big the effect is, not whether it exists.

Where the data disagrees

The most interesting disagreement is about whether the damage is levelling off. Seer Interactive tracked click rates on AI Overview queries falling from 1.76% to a low of 0.61% in September 2025. By February 2026 the rate had partly recovered to 2.4%, though it was still well below the rate on queries without an AI Overview.3 Another dataset showed a similar rebound from a December 2025 low, which analysts took as a sign that the decline may be stabilizing rather than continuing indefinitely.7

The second split is about who gets hurt. Amsive found that branded searches, where people look for a specific outlet or company, saw an 18% rise in click rate when an AI Overview appeared.2 Pages cited inside an AI Overview get far more clicks than pages that are not cited. Even so, uncited brands get less than a third of the clicks the same query would have sent before AI Overviews existed.3 Small publishers are hit hardest. Chartbeat data from earlier this year showed sites with 1,000 to 10,000 daily page views had lost 60% of their search referrals over two years. Large publishers lost 22%.24

The likely outcome is a two-tier web. Well-known brands with loyal readers keep a share of a shrinking pool, while generic and niche sites lose most of theirs. One analysis warned that this split threatens the variety of voices and competition online.2

Coverage also disagrees on a basic fact about Google's product. Some outlets reported that AI Mode became the global default after Google I/O in May.35 Google told Press Gazette directly that AI Mode is optional, and that a standard search still shows the familiar results page below any AI summary.34 Google's direct statement carries more weight. Still, AI Mode passing one billion monthly users means the difference between optional and default matters less each quarter.36

Core updates are no longer the main story

Google's ranking updates have been busy this year. A February update affected only Discover, Google's news feed. Core updates followed in April and May, a June update targeted spam, and a July update hit AI content farms.31 For many sites, though, these updates were a side issue compared with the steady growth of AI answers on the results page.31

The May core update launched in the same week Google moved AI Mode to its Gemini 3.5 Flash model and redesigned the search box. That made it unusually hard for sites to tell whether a traffic drop came from ranking changes or from AI answers absorbing clicks.36 As a result, a site can keep its top ranking and still lose visits, so rankings and traffic now need to be tracked separately.36 There is also a sign of the shift on Google's balance sheet. One analysis noted that ad revenue from Google's partner network of outside sites fell while revenue from Google's own search pages rose.8

Optimizing for citations instead of clicks

The marketing industry has responded with new labels: answer engine optimization (AEO) and generative engine optimization (GEO). The definitions overlap. Some guides treat them as two names for the same work: getting AI systems to mention and link to a brand.14 AEO usually means formatting content so an AI can lift it out as a direct answer, while GEO means getting a model to cite a source when it writes a longer response.11 Academic GEO research found that adding quotations, statistics and citations increased how much of an AI answer drew on a given source.11

There is a problem these guides tend to skip. One explains that the AEO model is built around zero-click results by design.14 Strategies built on that model aim to be visible inside answers that, by design, rarely send anyone to the source. Some editors accept that trade. One editor at The Athletic argued that being left out of AI sports answers would be a missed opportunity even without immediate clicks.23 That makes sense as a brand-building strategy. It does not pay for a newsroom.

Smaller AI companies are not filling the gap. In Chartbeat's dataset, ChatGPT accounted for 0.02% of publisher referrals and Perplexity for 0.002%. Google Search sent about 500 times as much traffic as ChatGPT.22 A pilot between Cloudflare and OpenAI helped ChatGPT find hundreds of millions of pages that ordinary crawlers had missed. Whether that sends any readers back to those sites is still being measured.24

Who is protecting publishers

The legal route has mostly closed in the US. On September 30, Judge Amit Mehta dismissed antitrust lawsuits from Chegg and Penske Media. He found the harm was real but said it was "an issue for Congress [or regulators]". It was the third dismissal of a publisher lawsuit over AI Overviews.21 Regulators elsewhere have done more. The UK's Competition and Markets Authority ordered Google to let publishers opt out of its AI features, and Google made that control available worldwide in Search Console by August 31.21 French newspaper publishers have also complained to their country's competition authority.30

Meanwhile, Google has offered smaller fixes:

• More visible links. It has added more links, placed them earlier in AI answers, and expanded news carousels.
• Preferred Sources. Readers can choose outlets they want to see more often. Google says readers are about twice as likely to click a site after adding it.21
• Payments. A pilot program pays about 100 publishers for content used in its AI products.21 One estimate put the payouts at roughly 0.1% of revenue, which bears no relationship to what publishers are losing.

These fixes do not address the core problem. When an AI answer has already answered the question, a more visible link gives the reader little reason to click.21

The financial damage is already showing up. Reach's half-year digital revenue fell £6.9 million, and Business Insider cut 21% of its staff.28 Wirecutter is looking for income beyond affiliate commissions.27 Its problem is clear in the numbers: the top result on shopping searches gets clicked 11.91% of the time without an AI Overview and 3.51% with one.27

What happens next

Blocking AI is not an easy way out either. A Rutgers and Wharton study found that large publishers lost about 7% of their weekly visits within six weeks of blocking AI crawlers.29 The stronger strategy, which Reuters Institute interviewees keep returning to, is to use any remaining search and AI exposure to build direct relationships with readers through newsletters, subscriptions and return visits.23

The reasonable reading is that the open web's traffic model is not about to collapse overnight, and parts of it may be stabilizing.3 The terms, however, are now set almost entirely by Google.21 Publishers that get through this period will probably be the ones that stop counting on referral traffic and treat every AI citation as a first step toward a direct relationship with a reader.