ChatGPT Search access linked to fewer searches, different referral patterns

Broader access to ChatGPT Search was associated with a 9% drop in traditional search queries, and the decline reached 17% after 20 weeks of exposure, in a July 2026 Bocconi University paper by Qiaoni Shi, Kai Zhu and Kai Gu. The study used URL-level Comscore desktop clickstream data from the United States covering October 2024 through July 2025 and more than 45,000 households.

The pattern was not just fewer searches. The paper found that ChatGPT produced a clean outbound referral in only 5.2% of conversation sessions, far below Google’s 31.1% referral rate across the same sample. Among ChatGPT-using households, 74.4% never generated a single clean referral to an outside website during the ten-month window; among Google-querying households, 9.6% had zero clean referrals. The authors describe that as a shift in the web’s referral bargain, with AI search satisfying information needs inside the intermediary instead of sending users outward.

The losses were concentrated in informational behavior. The paper says the largest search-referral declines appeared in informational categories, while transactional and recreational searches changed far less. That split matters for publishers, educators and reference sites, because it points to the exact queries most exposed when users start with ChatGPT instead of a traditional search engine: research questions, reference lookups and fact-finding tasks that used to spill traffic to outside pages.

Where the clicks go also looks different. ChatGPT’s outbound traffic is more likely to reach reference sites, academic resources, tools, SaaS products and developer content, while ad-supported sites receive far less of that mix. Google referrals still skew toward large, familiar destinations such as YouTube, Reddit and Wikipedia. For AI search visibility, that is the more important shift than raw referral volume: the system is not only changing how much traffic is sent, but which kinds of sites are being rewarded when it is sent.

Semrush adds scale to that picture. In its own analysis of more than 1 billion lines of U.S. clickstream data across a 200 million-user panel, outbound referral traffic from ChatGPT grew 206% in 2025. Even so, Semrush found ChatGPT’s search feature was enabled on only 34.5% of queries in February 2026, down from 46% in late 2024, which means most responses still relied on training data rather than live web search. Search Engine Land’s coverage of the Semrush data said Google accounted for 21.6% of all ChatGPT referral traffic, and the top 10 domains captured just over 30%.

That mix points to an early behavioral shift, not a simple traffic dip. Publishers and marketers now have to watch which intent classes are losing volume, which domains AI systems favor, and whether referral concentration keeps tightening around a small set of destinations.