• AI visibility

Guide · AI search

Is ChatGPT replacing Google search?

Not on current evidence. The best behavioral data shows people using AI assistants such as ChatGPT alongside search engines rather than instead of them: in a 2026 browsing panel, search still appeared in nearly half the sessions that included an assistant. But assistant sessions more often end without visiting any website, and no study can yet say whether AI is shrinking total search demand.

By Underneath Team Guide Updated

The short version

  1. A search engine appeared in 46.0% of browsing sessions that included an AI assistant, in a February 2026 panel of US and British users (Iannelli and Ai).
  2. 34.1% of assistant sessions included no visit to any outside website, against 19.5% of search sessions by the same people.
  3. People tend to search first and then open pages, but browse first and then turn to an assistant: a 20.6-point reversal in direction between the two.
  4. ChatGPT runs its own web searches, a mean of 3.7 per answer in our study, yet only 8.3% of the pages it cited ranked in Google’s top 10 for the question.
  5. Google’s own AI changes search more directly: AI Overviews appeared on 36% of searches in a 2026 US experiment, while people chose AI Mode for only 0.6%.

What does the best evidence on behavior show?

People use AI assistants alongside search, not instead of it, in the one panel that tracks both.

Iannelli and Ai (opens in a new tab) studied an opt-in panel of US and British users in February 2026. The panel’s software recorded conversations with ChatGPT, Gemini and Perplexity on the web, together with the same people’s searches and page visits. The authors then grouped each person’s activity into sessions and looked at where searching and browsing fell around assistant use.

A search engine appeared somewhere in 46.0% of sessions that included an assistant. It came before the first assistant prompt in 33.8% of them. The authors’ conclusion is careful: “containment and cross-surface coexistence occur at the same time.” Some assistant sessions stand alone, and many sit inside ordinary searching and browsing.

Read this study with its limits in mind. The authors work at a company that sells AI-visibility software, the paper was not yet peer reviewed, and they do not disclose the panel’s size. Use of assistants through phone apps was not observed. The main results did repeat in a second month, March 2026.

How is using an assistant different from searching?

Assistant sessions more often stay self-contained, and they tend to come later in the journey.

The same people behaved differently around the two tools. The table compares their sessions that included an assistant with their sessions built around search alone.

Measure (February 2026)Sessions with an AI assistantSearch sessions, same people
No visit to any outside website34.1%19.5%
Typical session length22 minutes9 minutes

Within the same person, the gap in self-contained sessions was 13.0 percentage points. In March 2026 it was 35.3% for assistant sessions, close to February’s figure. Our article on how often assistant users skip websites looks at this gap in detail.

Direction differed too. Around search, page visits mostly came after: they followed the last search in 60.2% of sessions and preceded the first in 46.7%. Around assistants this flipped: page visits came before the first prompt in 45.5% of sessions and after the last in 38.9%. In the authors’ words, search tends to open the journey, while assistants “sit deeper inside it”.

“Self-contained” does not mean the question was answered. The authors stress that the data cannot show whether a need was met. The levels also vary widely: self-contained assistant sessions were 50.9% in the United States and 20.7% in Great Britain.

Is AI shrinking the number of Google searches?

No study we reviewed answers this cleanly; the evidence is mixed and indirect.

Iannelli and Ai deliberately make no claim about search volume. They say their design cannot show whether AI creates or destroys search demand, because people turn to assistants at moments when they are already busy online.

Other evidence points both ways. Zhang, Cui and Zhang (opens in a new tab) summarize one study finding that adopting AI assistants cut total searches by roughly 20%, and another finding that it increased the number of websites people visit. We could not review either directly.

Single-site data shows how hard this is to read. Watanabe and Nakayashiki (opens in a new tab), who work for the company that owns the site they studied, found that ChatGPT referrals to pages they had not changed grew 3.5 times between January and May 2026. Google organic clicks across the site fell about 20% from late 2025 into 2026. One site cannot show what caused either change.

Does ChatGPT still depend on search engines?

Yes: ChatGPT runs its own web searches behind many answers, though it cites different pages from Google.

In our study of hidden searches, ChatGPT ran a mean of 3.7 searches per answer, Gemini 1.9 and Claude 0.76. In other words, the assistant often does the searching on the user’s behalf.

The results do not mirror Google. In our study of AI citations and Google rankings, only 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question. 68.5% were not in Google’s top 100 for the question or for any of ChatGPT’s own searches.

Huang and colleagues (opens in a new tab) found the same gap at the level of whole websites. OpenAI’s search-enabled GPT model cited a median of 3 sources per answer. Its 100 most-cited domains overlapped only 24% with those of Google’s AI Overviews and 25% with Google’s normal results.

Is Google’s own AI changing search more than ChatGPT is?

For most businesses, probably yes, because Google’s AI answers sit inside the search people already use.

In the field experiment by Wang and colleagues (opens in a new tab), 1,100 US volunteers saw AI Overviews on 36% of their searches before the study began. They chose Google’s chat-style AI Mode for only 0.6% of searches. On a fixed set of benchmark searches, Aral, Li and Zuo (opens in a new tab) found 67% of US searches answered by AI in 2025, against 42% in 2024.

People resisted being pushed further. When every search was sent to AI Mode for a week, the share of people who tried Bing, DuckDuckGo or Yahoo rose 11.2 percentage points, and clicks to outside websites fell 18.8 points. Our article on AI Mode as Google’s default covers that result.

What should you do about it?

Plan for both channels, because the evidence shows people using them together rather than swapping one for the other.

  1. Do not move your whole search budget to AI. Search still tends to open the journey toward websites.
  2. Treat AI assistants as a later step in the customer journey, where people compare and decide. Content that helps a decision matters there.
  3. Do not assume your Google rankings carry over. Only 8.3% of ChatGPT’s citations ranked in Google’s top 10 for the question in our study.
  4. Measure AI referrals separately, and compare them with pages you did not change. One site’s untouched pages saw ChatGPT referrals grow 3.5 times in five months.
  5. Watch Google’s own AI features, which reach far more searches than people choose to send to chat tools.

If you want help planning for both, see our generative engine optimization service.

What does the research not tell us yet?

The research cannot yet say whether AI assistants reduce, add to or reshape total search demand.

  • No causal study of search volume was available to us; the behavior panel describes patterns, not causes.
  • That panel was run by an AI-visibility software company, is not yet peer reviewed, covers two months in two countries and misses phone apps.
  • Nobody knows whether self-contained assistant sessions actually solved the user’s need.
  • Differences between business buyers and consumers are not measured.
  • Single-site referral data cannot be generalized to other websites.

Frequently asked questions

Are people using ChatGPT instead of Google?

Mostly alongside it. In a February 2026 US and UK panel, a search engine appeared in 46.0% of browsing sessions that included an AI assistant.

Does ChatGPT use Google search results?

ChatGPT runs its own web searches, a mean of 3.7 per answer in our study, but only 8.3% of the pages it cited ranked in Google’s top 10 for the question.

Do AI assistants send people to websites?

Less often than search does. 34.1% of assistant sessions in the 2026 panel included no visit to any outside website, against 19.5% of search sessions by the same people.

Is Google losing users to AI chatbots?

The evidence does not show it yet. The panel study found people using both, and when a 2026 experiment forced Google’s AI Mode on people, 11.2 percentage points more of them tried another search engine.

Sources

Free strategy call

Some questions are easier to answer about your own business.

Bring the one that matters most. On a free 30-minute call we’ll take a first look at it and send you a short written read afterward.

Ask about my businessHow we work