Households with ChatGPT Search run fewer Google queries and click less

Matt G. Southern of Search Engine Journal published a piece on 4 September that reconciles three findings people have been citing as if they conflict. Similarweb says 95% of ChatGPT users also use Google. A Bocconi University working paper says the same households run about 9.4% fewer traditional queries after gaining ChatGPT Search access. A field experiment on arXiv says users routed to Google's AI Mode click through to external sites 18.8 percentage points less often. All three are true because they measure different things.

Overlap counts people, not queries

Similarweb put the share of ChatGPT users who also appear in Google's audience at 95% in September 2025, and reported the rate held at 95% through May 2026 while visits to generative AI platforms grew 70% year on year. The piece calls out the limit: one Google search in a month is enough to be counted as a Google user, so someone who moved most research to ChatGPT and only uses Google for maps and store hours still shows up in the overlap.

The same households run fewer queries after ChatGPT Search

The Bocconi paper uses Comscore U.S. desktop clickstream data from October 2024 to July 2025, comparing households that gained ChatGPT Search access to similar households with no prior ChatGPT or Claude activity. Households with access ran on average 3.14 fewer weekly queries on Google, 9.4% below the pre-expansion baseline of 33.51. The gap grew to 17.0% after 20 weeks. A within-adopter comparison gave a smaller 4.9% drop, rising to 8.2% after 20 weeks, which the authors read as selection into adoption rather than the effect of access.

The breakdowns by category matter more than the average. Referrals to academic sites fell 32.8%, reference sites 26.5%, developer sites 15.1%, and news sites 13.4%. Marketplaces and entertainment were flat. Google's household reach in the same panel held at about 49% while ChatGPT rose from 4.6% to 7.9%.

Forced AI Mode cuts external clicks

A field experiment on arXiv on 18 August by Wang, Gleason, Bart, Wilson and Metaxa recruited 1,100 U.S. Chrome users who use Google as their primary search engine. For one week, participants were assigned to standard Google, Google with AI Overviews hidden, or Google with every search routed to AI Mode. Baseline AI Mode use was 0.6%; the treatment group hit 94.7%, which the authors call forced adoption rather than an opt-in.

Under that condition, click-through to external sites fell 18.8 percentage points, with a 95% confidence interval of -22.2 to -15.3. News clicks dropped 12.5 points, Reddit 21.2, Wikipedia 9.9. Session length rose 0.43 minutes. Use of Bing, DuckDuckGo and Yahoo went up 11.2 points, and trust, satisfaction and usefulness ratings went down. Of 309 AI Mode users answering an open question, 47 reported difficulty reaching specific websites.

Impressions hold while the click disappears

A product site or a documentation hub can hold its Google impressions and still lose the click, and that is the pattern these three datasets describe together. A dashboard that tracks only impressions and average position will show the site is fine; one tracking clicks and referrals will show the drop; one tracking brand mentions inside AI answers will show where the query actually landed. Southern names branded search lift as the metric Similarweb says would move if AI were creating demand, and notes that most analytics tools do not attribute a later branded Google visit back to an earlier ChatGPT session. For a team whose users ask a chatbot first, that is the measurement gap to close before assuming referral loss is a ranking problem.

Source

Original reporting: People Aren't Leaving Google For AI. They're Using Both by Matt G. Southern, Search Engine Journal, 4 September 2026.