ChatGPT Is Replacing 17% of Google Searches After 20 Weeks. Here Is What to Measure Before Q3.

A Bocconi University study published July 8 just gave us the first causal proof that ChatGPT Search is pulling queries away from Google at a measurable, accelerating rate. After 20 weeks of access, traditional search queries drop 17% per household. Only 5.2% of ChatGPT sessions produce an outbound click to any website. Your Q3 traffic forecast is based on a search volume that no longer exists for a growing share of your buyers.

The Study That Changes the Traffic Math

Researchers Qiaoni Shi, Kai Zhu, and Kai Gu analyzed desktop clickstream data from over 45,000 U.S. households tracked by Comscore over ten months. They used the staggered rollout of ChatGPT Search — paid subscribers in October 2024, free users in December 2024, logged-out users in February 2025 — as a natural experiment.

The results are not subtle. Households that gained ChatGPT Search access immediately reduced Google queries by 9.4%. By week 20, that decline deepened to 17%. This is not a survey. It is a behavioral measurement of what people actually did when ChatGPT became available.

The deepening matters. That 9.4% is not where it stops. The habit builds. Every week a buyer uses ChatGPT for a research question they used to Google, the likelihood they return to Google for the next one drops.

The 5.2% Referral Problem

Here is the number that should reshape how you think about this channel. ChatGPT produces an outbound referral in only 5.2% of conversation sessions. Google sends users to an external website 31.1% of the time. That is a six-to-one gap.

Among the 56,578 households that used ChatGPT during the study window, 74.4% never generated a single outbound click in ten months. Not low clicks. Zero clicks. For Google, that zero-click household rate was 9.6%.

Google's AI Overviews have their own click problem — Pew Research found click-through drops from 15% to 8% when an AI summary appears. But Google still structurally routes people to websites. ChatGPT does not. The answer is the product. The citation is a footnote most users never follow.

Where the Remaining Clicks Actually Land

Even the 5.2% that do click through have a landing page problem. A Previsible study of 6.77 million LLM-driven sessions across 166 websites found that ChatGPT sends 28.8% of its referrals to internal site search pages — not the page that answers the query.

ChatGPT trusts your domain enough to send someone, but it often cannot identify the right page. So it defaults to your site search. If your internal search is a box that returns ten vaguely related results, you just wasted the highest-converting referral channel available.

This is not hypothetical. Seer Interactive found that ChatGPT referral traffic converts at roughly 16%, compared to 1.8% for Google organic. That is a 9x conversion advantage. But it only works if the visitor lands on a page that continues the conversation the AI started.

What to Measure This Week

If you run demand gen, growth, or content for a B2B company, here is the diagnostic before Q3 planning:

1. Find your ChatGPT referral volume in GA4. Filter by source/medium for chatgpt.com / referral. The number will be small. That is the point. Track the trendline, not the total. Monthly LLM sessions grew 9.9x to 644,478 in May 2026 across the Previsible panel. Your growth curve probably looks similar.

2. Audit where ChatGPT sends your traffic. Pull the landing page report filtered to ChatGPT referrals. If a material share hits /search, /results, or your site search page, your internal search just became an acquisition surface. Treat it like one: featured results, clear above-the-fold answers, structured content paths.

3. Subtract 10-17% from your organic search growth forecast. The Bocconi study measured this on general households, not just tech-early adopters. Half of B2B buyers now start research in AI chatbots rather than Google, per G2's 2026 report. If your pipeline model assumes last year's search volume, the model is wrong.

4. Measure the influence you cannot see in referral data. A buyer asks ChatGPT which platforms handle AI visibility measurement. ChatGPT names your brand. The buyer Googles your brand name, visits your site, and converts. That shows up as "branded organic" or "direct" in your analytics. It was an AI-assisted discovery. Track branded search volume and direct traffic segments alongside ChatGPT referrals to see the full picture.

5. Check your citation presence, not just your ranking. Google AI Overviews now appear in over 70% of personalized U.S. search results as of July 2026 (Advanced Web Ranking data). Ranking first gives you a position. Being cited inside the AI answer gives you a conversion. Seer Interactive found brands cited in AI Overviews earn 35% higher click-through than those ranking on the same query without citation.

The Reframe for Q3 Planning

The old search model: you rank, people click, traffic comes, pipeline follows. Every step of that chain is compressing.

68% of Google searches now end without a click, up from 60% in 2024 (SparkToro and Similarweb, 2026). The remaining 32% is split between paid ads, Google properties, and organic links. A randomized field experiment published on SSRN found AI Overviews reduce engaged organic clicks by 39.8%.

On the ChatGPT side, the Bocconi study proves the migration is causal and accelerating. Your buyers are not searching less. They are asking differently, on a platform that resolves the question in place and rarely sends anyone downstream.

The operators who will win Q3 are the ones measuring both channels now — not the ones who wait for the traffic to finish declining and then try to explain it in the board deck.

FAQ

How much does ChatGPT Search reduce Google queries?

A Bocconi University study of 45,000+ U.S. households found ChatGPT Search access reduces traditional search queries by 9.4% immediately and 17% after 20 weeks. The decline is causal and habituating, measured from URL-level Comscore clickstream data across a ten-month panel (Shi, Zhu, and Gu, 2026).

What percentage of ChatGPT sessions produce a click to a website?

Only 5.2% of ChatGPT conversation sessions result in an outbound referral to an external website, compared to 31.1% for Google searches. Among ChatGPT-using households in the Bocconi study, 74.4% never generated a single outbound click across ten months of usage (Shi, Zhu, and Gu, 2026).

Why does ChatGPT referral traffic convert higher than Google organic?

Seer Interactive analysis found ChatGPT referrals convert at approximately 16% versus 1.8% for Google organic — roughly 9x higher. Visitors arriving from ChatGPT have already received context about the topic and arrive with formed intent, rather than browsing search results for orientation (Seer Interactive, 2026).