Almost every enquiry a small business receives begins with somebody typing a question into a box. For about twenty years the job of a website was to be one of the links that came back, and an entire discipline grew up around climbing that list. That arrangement is quietly ending. Search engines and chat assistants now write the answer themselves, and a growing number of people read it and stop there. In this post we'll look at what the measurements actually show, what it changes about how a business earns attention, and what's worth doing about it now.
The Pew Research Center followed 68,879 Google searches made by 900 U.S. adults during March 2025, and found that people clicked one of the ordinary result links on 8% of the visits where an AI summary appeared, against 15% of the visits where no summary was shown [Pew Research Center, 2025]. About 18% of those searches produced a summary at all, so this is not yet the whole of search. On the searches where a summary does appear, though, close to half the onward clicks go somewhere else. That gap is the story.
A summary at the top of the page changes what people do next
The same study found that the links inside a summary were clicked on 1% of visits, which is another way of saying almost never. People were also likelier to stop browsing altogether afterwards, ending their session 26% of the time following a search with a summary against 16% without one. Read one way, that is a straightforward loss of traffic. Read another way, it describes somebody who got what they came for and had no further use for a list.
The more useful finding sits underneath. Pew reported that 88% of the summaries it captured cited three or more sources, and that a typical summary ran about 67 words. Something has to fill those citations, and what fills them is ordinary web pages written by ordinary businesses. Being one of them is the new version of ranking. The competition moved from the list to the answer, and the answer has room for about three names.
Fewer visitors has not meant fewer customers
The worry underneath all of this is easy to state. If the assistant answers the question, nobody arrives, and the website stops earning its keep. The largest published dataset on the point suggests the opposite is happening. Adobe, working from more than a trillion visits to U.S. retail sites, found that people arriving from AI assistants converted 42% better than traffic from other sources in March 2026 [Adobe, 2026]. The rest of the engagement picture points the same way.
| Measure | Visitors arriving from AI assistants, compared with other traffic |
|---|---|
| Conversion rate | 42% better |
| Time on site | 48% longer |
| Pages per visit | 13% more |
| Engagement rate | 12% higher |
Two caveats belong here, because they're the sort of thing that gets dropped in the retelling. This is retail data, and a dental practice, a law firm, or an HVAC contractor may not behave like an online store. It is also Adobe measuring sites that run Adobe's own analytics, which makes it a very large sample rather than a random one. What holds up is the direction and the volume, with AI-referred traffic to those sites up 393% year over year in the first quarter of 2026. Fewer visitors, and more of them already convinced. That is a trade most owners would take.
Most of the preparation is work you already recognise
A small industry is forming around the idea that AI search demands a new technical toolkit. Google's own documentation says otherwise. Its guidance for site owners states that there are no additional requirements to appear in AI Overviews or AI Mode, that no new machine-readable files are needed, and that no special structured data has to be added [Google Search Central]. A page has to be indexed and eligible to show a snippet. That is the entry fee, and it hasn't changed.
What does matter is whether a machine can read what you wrote. In the same Adobe analysis, retail product pages scored 66% on a machine-readability check, meaning roughly a third of what those pages say is invisible to a language model reading them. Homepages managed 75%, and the strongest sites in the set reached 82.5% against 54.2% for the weakest. Text that only exists inside an image, a slider, or a script that runs after the page loads is text nobody is going to quote. Plain writing in plain HTML is having a good year.
What to measure once the click stops being the measurement
If citation is what you're competing for, the reporting has to follow it there. That means tracking how often an assistant names your business when somebody asks the question you'd want to be asked, how you get described when it does, and which competitors turn up alongside you. It is closer to media monitoring than to a rank report. Somebody has to read it, not average it. None of this is as sudden as it looks, either, since answer engines were already reshaping the results page well before AI Overviews had a name.
The older work does not go away. Somebody still has to keep the site quick, keep the pages indexed, and make sure the thing a customer actually asks about is written down somewhere in a sentence a machine can lift. At Barreras IT we run that visibility tracking alongside conventional search work in our SEO, SERP and GEO packages, mostly because splitting them apart produces two reports that disagree with each other. The businesses getting a clear read on this are the ones treating it as one job rather than two.
A reasonable first step, if you only do one thing
Write down the five questions somebody would ask before choosing a business like yours, put each of them to ChatGPT, Gemini, and Perplexity, and record what comes back. In about twenty minutes you'll know whether you get mentioned, how you get described, and who is being recommended instead of you. That is a baseline, and a baseline is the thing most missing from these conversations. Everything else is easier to argue about once you have one.
The wider point holds whichever assistant ends up in front. The web is now read by machines before it is read by people, and the pages that get quoted are the ones that said something clearly enough to be quoted. That was good advice before any of this, which is the most reassuring thing about it. If you'd like a hand putting that baseline together, we're happy to run it with you.
Sources
Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, July 2025. Adobe, U.S. retailers see surge in AI traffic, but many websites are not entirely readable by machines, April 2026. Google Search Central, AI features and your website.
