Obility Editorial · · 4 min read
A summary roughly halves clicks on the page where it appears
The cleanest public evidence on this question comes from Pew Research Center. It took the browsing records of 900 U.S. adults on its KnowledgePanel Digital panel for March 2025, reran their 68,879 unique Google searches to see what each results page contained, and then looked at what each person did next. About 18 percent of those searches produced an AI summary.
When a summary appeared, people clicked a traditional search result on 8 percent of visits. When it did not, they clicked on 15 percent. Links inside the summary itself were followed on 1 percent of visits. People were also more likely to end their browsing session after a page with a summary, 26 percent against 16 percent without one.
That 1 percent is the number to remember. Being cited in an AI Overview is worth having, but on Pew’s evidence a citation does not give back the click the summary took. Pew recorded only the first three sources in each summary, so the true rate for summary links may be somewhat higher.
Google says total clicks held steady, and both claims can be true
Google disputes the gloomy reading. In an August 2025 post, Liz Reid, who leads Google Search, wrote that total organic click volume from Search to websites had been relatively stable year over year, that Google was sending slightly more quality clicks than a year earlier, and that third-party reports of steep declines often relied on flawed methods.
Read the same post closely and it concedes the mechanism. Reid writes that for quick-answer questions people may be satisfied with the response and not click, and that traffic is shifting between sites, down for some and up for others. A stable total across the whole web says nothing about your share of it. Pew measured what happens on one results page; Google reported a sum across billions of them. Your site sits in the first frame, not the second.
Buying questions are the searches most likely to get a summary
Pew also found which searches produce a summary. Only 8 percent of one or two word searches did, against 53 percent of searches ten words or longer. Searches starting with a question word such as who, what, when or why produced a summary 60 percent of the time, and searches written as full sentences 36 percent of the time.
That pattern matters more to a software or services company than the headline average. Your brand name and short product terms rarely trigger a summary. The research questions a buyer types before they know your name, such as which tool suits a team of a given size, or how two approaches compare, are long and question shaped. Those are exactly the searches where a summary sits above your link and answers first.
The summary is the visit, so read what it says about you
Pew found the typical summary was 67 words long and that 88 percent cited three or more sources. For a buying question, 67 words is room to name a handful of options and a reason for each. A buyer who stops there has formed an impression of your category without visiting any site, and whether your company was named, and how it was described, was decided inside the summary.
So the useful unit of measurement shifts from the click to the answer. Take an illustrative list of 30 buying questions. For each one, record whether an AI Overview appears, whether it names you, which competitors it names, how it frames each, and which pages it cites. That is the work Obility’s Answer Engine Insights is built around: it keeps the surrounding answer, the recommended competitors and the cited sources with every observation, grouped by audience and buying stage, because a mention is evidence, not a conversion. Engine and market coverage is something to confirm at a demo.
The output tells you where the work is. A question where the summary names three competitors and cites a comparison page you do not appear on is a content and coverage gap. A question where you are named but described with an old pricing model is an accuracy problem. A question with no summary at all is still a classic search result, and ordinary ranking work still applies.
What this evidence cannot tell you about your own site
Do not apply 8 and 15 percent to your own traffic. Pew sampled the everyday searches of U.S. adults in one month of 2025, across every topic, on Google alone, and it did not measure AI Mode. Business software research is a small and unusual slice of that, and Google can change when summaries appear at any time.
Your own numbers will not settle it cleanly either. Search Console counts AI Overview clicks inside ordinary web search results, so a drop in clicks with flat impressions is consistent with a summary and with several other causes. Split your queries by length and question form, compare click-through on each group before and after, and expect the long, question-shaped group to move first. If it does, chasing the lost click is the wrong target. Make sure the summary that replaced it describes you accurately and names you at all.