AI Search and Its Impact on UX: Fewer Clicks, More Answers

What changed in behavior?

Three measurements point in the same direction, with different methods and limits worth knowing.

• Pew Research Center (July 2025). It tracked the browsing of 900 US adults in March 2025. On visits with a Google AI summary, 8% included a click on a traditional result; without a summary, 15%. Only 1% of visits with a summary included a click on a link inside the summary itself. The summary also appeared on 8% of one- or two-word searches and on 53% of searches with ten or more words.
• SparkToro (June 2026). Using Similarweb panel data, 68.01% of US Google searches between January and April 2026 ended without a click, against 60.45% in 2024. The author warns that each year’s panel is not fully comparable, and the study does not isolate how much of the rise comes from AI summaries, which Ahrefs says appear on more than 20% of searches.
• Agarwal and Sen (SSRN, April 2026). A randomized experiment with 1,065 US participants over two weeks in January and February 2026: a Chrome extension hid the summaries for one group. On queries where a summary appeared, outbound organic clicks fell 38% and zero-click searches rose from 54% to 72%. It is a draft without peer review, and the authors mention that participant attrition may have affected the results.

The sensible takeaway is the direction (fewer clicks, especially on long, informational questions), not a single number. These are Google data from the US; they do not describe other search engines or other countries. To see how answer products work, read conversational search.

What does it mean for the landing page?

Someone who arrives after seeing a summary already has a basic answer. They come to verify, compare, act, or find what the summary did not give them. That changes the page hierarchy:

• Start where the summary ends. If the general question is answered, go to the next step: a concrete case, a tool or a comparison.
• Show the source and the date. Someone arriving to verify needs to know how recent the information is.
• Offer what the answer cannot give. Original data, a form, a calculator, a real example.
• Make authorship visible. When answers blend together, it matters more to know who signs.

This is design judgment, not a measured result: none of the studies cited proves that these changes win visits back. Treat them as hypotheses and measure them.

How do you adapt a site, step by step?

1. List the 20 most frequent questions about your topic and check how much of each Google’s AI summary already answers.
2. For each page, define what the visitor needs to do or know after that answer.
3. Make sources, review date and authorship visible.
4. Record a baseline (visits, queries, time, conversion) before touching anything.
5. Change one page at a time and compare with the previous period.

For the side of being cited inside the answer, see what AEO is and how to write content that LLMs cite.

How do you measure the quality of a visit?

A visit with a specific question behaves differently from one with a general term. If your analytics tool allows it, separate three things: the share of visits that arrive with long queries (for example, five or more words), the time spent in the section that answers their question, and the next action, whether that is reading another case, downloading a template or getting in touch. If that share and that action rise while sessions fall, you may be losing drive-by visits and keeping the ones that matter. Search Console shows queries but does not separate visits that saw an AI summary: there is no official metric for that.

Without a baseline, any improvement gets credited to what you did, when it may be due to the season or to a campaign that happened to coincide.

What about internal search?

If your site has a search box, it also receives more natural-language questions, because people get used to asking in sentences. Review the queries that returned no results over the last thirty days: they show which questions arrive and go unanswered. Add synonyms and full sentences to the documents that should appear, and write a short answer for queries that repeat. To decide whether you need more than keywords, compare with semantic vs. keyword search.

When does it change less?

When your business depends on an action inside the site (a purchase or booking that only happens on your platform) and your traffic comes from branded searches. There, the experience on the page matters more than the change in search.

Example

Illustrative example: a page about “what is a design token” receives visits that leave within a few seconds, and the team suspects the search summary already gives the definition. They redesign the page: at the top, one line with the definition; then a token in code and in Figma, a table of common mistakes and a downloadable template. Before the change they note time on page and downloads for the last month; four weeks later they compare. If the improvement shows up only in visits with long queries, the hypothesis holds. If nothing changes, the page was not the problem.

Common mistakes

• Believing lost traffic returns if you publish more of the same content.
• Repeating the general definition the user has already read.
• Measuring sessions and not the quality of visits.
• Hiding the date and the authorship of the content.
• Quoting a “traffic drop” figure without saying which study it comes from, in which country and with what method.
• Designing internal search as if queries were still two words long.

Sources

• Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results (July 22, 2025)
• SparkToro, In 2026, Less than One Third of Google Searches Still Send a Click (June 9, 2026)
• Search Engine Journal, Study Confirms Google AI Overviews Cut Organic Clicks 38% (April 27, 2026), on the Agarwal and Sen draft on SSRN