Guide · AI search

Do people click fewer websites when they use AI answer engines?

Yes. In a 2024 lab study of 21 highly educated participants, people using AI answer engines tended to click one source, against an average of four on Google. Larger studies of Google’s AI features and of AI assistants point the same way, so each question now sends visits to fewer websites.

The short version

  1. In a 2024 lab study, people clicked about one source per question with answer engines and an average of 4 on Google (Narayanan Venkit and colleagues).
  2. They also looked at far fewer sources, hovering over an average of 2 with answer engines against 12 on Google.
  3. In a 2026 US experiment with 1,100 people, sending every search to Google’s AI Mode cut clicks to outside websites by 18.8 percentage points.
  4. In a 2026 panel, page visits followed 38.9% of AI assistant sessions, against 60.2% of the same people’s search sessions.
  5. The one source people click may not be a Google winner: only 8.3% of ChatGPT’s citations in our study ranked in Google’s top 10.

Do answer engine users click fewer sources than Google users?

Yes, in the one lab study that compared them directly: about one source against four.

Narayanan Venkit and colleagues (opens in a new tab) asked 21 participants to research questions in their own field and debate topics. Each used one answer engine, an AI tool that writes an answer and cites sources, and Google for the same questions. The engines were YouChat, Bing Copilot and Perplexity, as they were in 2024.

EngineSources clicked per question (average)
Google3.80
Perplexity1.29
YouChat1.07
Bing Copilot0.76

The authors sum it up: “With answer engines, users tend to click on a single source and trust the model’s selection.” On Google, participants clicked and read an average of 4 sources.

The gap was wider for sources people merely looked at. Participants hovered over an average of 2 sources with answer engines and 12 on Google.

How much weight should you put on that study?

Some, but not much on its own: it was small, expert-heavy and used 2024 engines. The 21 participants were mostly people with or working toward a PhD, chosen so they could judge answers in their field. Each session lasted 90 minutes and followed a think-aloud format, which can change how people behave.

Several of the authors work at Salesforce AI Research, an industry lab. The answer engines tested have changed a lot since 2024, and none was ChatGPT. Treat the one-versus-four figure as a clear direction, not a precise rate for your buyers. For ChatGPT itself, see how people use ChatGPT alongside Google.

Do larger studies show the same pattern?

Yes: field data on Google’s AI features and on AI assistants also shows fewer clicks out. The studies differ in method, but they point the same way.

StudyWhoFinding
Chapekis and colleagues (opens in a new tab), Pew Research Center900 US adults, March 2025A search result was clicked on 8% of Google pages with an AI Overview, against 15% without
Wang and colleagues (opens in a new tab)1,100 US users, March 2026Forcing every search into AI Mode cut outside clicks by 18.8 points
Iannelli and Ai (opens in a new tab)US and British panel, February 2026Page visits followed 38.9% of assistant sessions, against 60.2% of search sessions

An AI Overview is the AI summary at the top of Google’s results; AI Mode is Google’s conversational search. In the Pew panel, people clicked a source cited inside the AI Overview on only 1% of visits. In the AI Mode experiment, fewer people clicked through to news sites, Reddit and Wikipedia alike.

The assistant panel study comes from authors at Scrunch AI, a commercial company, and was not yet peer reviewed. In that panel, 34.1% of assistant sessions had no observed web visit at all, against 19.5% of search sessions.

Why do people click fewer sources with AI answers?

Mostly because they trust the engine’s pick and are shown fewer sources to choose from. In the lab study, the answer engines displayed an average of 4.31 sources per answer and cited 3.0 of them.

Agreement also cut clicking. When a question matched the participant’s own view, they clicked an average of 0.48 sources and often accepted the answer with little checking. They checked more when the question challenged their view.

People notice the narrower choice. When Wang and colleagues asked AI Mode users for their impressions after a week, 13.4% mentioned limited links or a narrow range of sources. Another 15.3% described difficulty getting to specific websites.

Which websites lose the clicks?

Every source except the one the AI puts in front of the reader. On Google, people in the lab study opened about four results, so several sites could win a visit from one question. With an answer engine, the visit goes to the single source the person trusts, if any.

That source is often not the one that ranks. In our rankings study, only 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question. Our comparison of Google’s AI surfaces found AI Mode cited 16.3% of top-10 pages, against 29.7% for AI Overviews.

And Grossman and colleagues (opens in a new tab) found that on average only 18% of sources were shared between an AI Overview and Google’s ordinary results for the same search.

What should you do about it?

Aim to be the one source the AI names, and expect fewer visits from everyone else. Practical steps:

  1. Check who gets cited. Run your buyers’ key questions through ChatGPT, Perplexity, Google AI Mode and others, and note which sources appear.
  2. Be the obvious primary source. Original data, clear definitions and plainly stated facts give an AI engine a reason to cite you rather than a summary of you.
  3. Earn mentions on cited sites. If the engines favor review sites, publications or forums in your category, your presence there matters as much as your own pages.
  4. Value mentions, not only clicks. A citation that wins no click can still shape a buyer’s shortlist.
  5. Check what is said about you. In the lab study, around 30% of statements from You.com and Perplexity were not supported by their listed sources, so errors about your brand are possible.

If you want help becoming the source AI answers cite, see our generative engine optimization service.

What does the research not tell us yet?

The direct comparison rests on one small 2024 study, so much is still open. The main gaps:

  • Scale. No large study has counted per-question clicks for ChatGPT, Perplexity or Gemini users against Google users.
  • Buyers. No study covers business buyers or high-value purchases, where people may check more sources.
  • Current tools. The engines in the lab study have changed, and ChatGPT was not among them.
  • Click value. Nobody has tested whether the one click an answer engine sends is worth more than a typical search click.

Frequently asked questions

Do people click the links in AI answers?

Less often than on Google. In a 2026 panel, page visits followed 38.9% of AI assistant sessions against 60.2% of search sessions by the same people.

How many sources do people check when using AI search?

About one per question, in a 2024 lab study, against four on Google. On Google’s own AI Overviews, only 1% of visits led to a click on a cited source.

Does ranking on Google still matter for AI answers?

Yes, but it does not guarantee a citation. AI Overviews in our study cited 29.7% of top-10 pages, and ChatGPT’s citations ranked in Google’s top 10 only 8.3% of the time.

Why do people trust AI answers without clicking?

They tend to accept the engine’s choice of source. In one lab study, people asking questions that matched their own view clicked an average of 0.48 sources.

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.