--- title: "Which Searches Trigger AI Overviews? What Pew and Semrush Data Show" description: "Long, question-style searches trigger AI summaries far more often than short ones in Pew's data, and Semrush shows the feature spreading. Here is what to check on your site." author: "Team" category: "Search & AI Visibility" date: 2026-11-10T07:25:01.971Z canonical: "https://contentagents.dev/blog/which-searches-trigger-ai-overviews-what-pew-data-shows" --- # Which Searches Trigger AI Overviews? What Pew and Semrush Data Show ![church, pews, benches](https://cdn.pixabay.com/photo/2016/09/04/22/08/church-1645414_1280.jpg?w=1200&q=75) > Long, question-style searches trigger AI summaries far more often than short ones in Pew's data, and Semrush shows the feature spreading. Here is what to check on your site. The shape of a search predicts whether Google answers it with an AI summary. In Pew Research Center's tracked-browsing data from March 2025, summaries appeared on a small share of short searches and on most long ones. A second dataset, from Semrush, shows the feature spreading beyond the kind of query Pew described. Together they tell a content team where to look first, and where not to trust a single study. Content and SEO leads deciding which pages are most exposed are the intended readers. It reads two studies closely, says what each does not show and ends with a one-hour test on your own data. ## How the evidence was gathered Pew followed 900 US adults through March 2025 and logged 68,879 unique Google searches; [12,593, about 18%, produced an AI summary](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/). Semrush analysed more than ten million keywords from its database across 2025 and refreshed the study in December. We did not run our own study. Both sources are snapshots, and one is a vendor of AI visibility tools. ## Pew: the odds depend on how a search is built Pew's table is simple. [One- or two-word searches produced a summary 8% of the time, and searches of ten or more words 53% of the time](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/). Searches opening with a question word such as who, what, when or why reached 60%, while full sentences with a noun and a verb reached 36%. Longer queries usually describe a situation or ask for a comparison, which a summary can answer in a few lines. Short ones are often brand lookups, where a list of links still serves best. For a B2B site, that puts the long-tail "how do we" and "what is the difference" guides at greatest risk, and brand or product-name searches at least. ## Pew: the summaries themselves are brief Pew's median summary was 67 words, ranging from 7 to 369, and [88% cited three or more sources](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/). Wikipedia, YouTube and Reddit were the sites linked most often in summaries and in standard results alike. Short answers built from several pages favour concise passages that stand alone. They also make joining a group of cited sources a sensible goal, where becoming the only one is not. ## Semrush: the feature moved beyond information-seeking The Semrush study tracked the share of queries showing an AI Overview through 2025. It put that share at [6.49% in January 2025, 24.61% in July and 15.69% in November](https://www.semrush.com/blog/semrush-ai-overviews-study/), and reported that the informational slice of overview-triggering queries fell from 91.3% in January to 57.1% in October, as commercial, transactional and navigational queries gained share. The article warns that its navigational baseline is inconsistent and that it does not claim causation for the click effects. The signal for planning is direction, not level. Trigger rates moved a great deal within a single year, and the mix widened. A page that looked safe because its query was commercial in early 2025 may look different now. ## Limits you should respect **Google has changed.** At its May 2026 event Google said [AI Mode had passed one billion monthly users](https://blog.google/products-and-platforms/products/search/search-io-2026/). Pew's 18% describes March 2025 and should not be projected forward. **The samples are narrow.** Pew covers US adults in one month, not B2B researchers. Semrush covers its own keyword database and does not state the market or how keywords were chosen. **Neither study measures value.** Pew recorded what happened on the results page and did not follow people who later searched for a cited brand. Semrush notes that a higher zero-click rate on overview keywords may reflect informational intent rather than the overview. **Length is a marker, not a cause.** The data cannot rule out other ways in which long queries differ from short ones. ## Our stance: treat query length as a first filter, not a map The neat reading of the Pew data is to protect short queries, rewrite for long ones and move on. We would not stop there. The Semrush trend shows the set of queries that trigger overviews is not stable, so a plan built purely on length will age within months. Use length to decide where to look first, then check your own pages directly, because your audience and your intents are not the average of US adults. The cost is that this takes recurring work: a quarterly re-split of your queries and a fixed prompt list to see whether you are named. We would relax the habit only if Google published trigger rates by query class, which would let a team plan from a stable source. Until then, a snapshot is a hypothesis. ## Putting it to use Start in Search Console by separating long question-style queries from short ones, as [the reporting guide](/how-to-report-on-ai-overviews-in-search-console) describes. Open question pages with an answer and a cited source, and keep titles and snippets sharp on brand and short queries. Track naming and citation for the prompts that matter with the [measurement guide](/ai-visibility-metrics-complete-measurement-guide). One question can also fan out into several searches, the argument behind [topic clusters in the AI search era](/topic-clusters-and-pillar-pages-in-the-ai-search-era). ## Test your own exposure in an hour Download sixteen months of queries and sort them into three bands by word count: up to two, three to nine, and ten or more. For each band, set clicks, impressions, click-through rate and average position against the previous year. If the longest band loses most clicks while position holds, that resembles Pew's pattern. A flat result suggests other forces matter more, such as how strong your brand is or what intent your pages serve. Record the outcome as an observation. For example: ten-word-plus queries lost some share of clicks at stable position while short ones held. Finish with one action for the long band, for example new openings on the ten pages that lost the most. ## Where we would stop short of a conclusion Short queries should not be called safe for good, as product changes can widen summaries to them. A high summary share does not make a page worthless, because being cited can still build awareness. And neither dataset describes how B2B buyers research. ## If the numbers move again Expect them to. Treat each new study as a prompt to repeat the one-hour test, not as a replacement for it. Keep the three-band export in a folder with the date, so that next quarter you can see change in your own data rather than argue about someone else's. That habit is worth more than any single percentage in this article. ## FAQ ### What kinds of searches trigger AI Overviews? In Pew's March 2025 US data, long and question-style searches triggered summaries most: 53% of searches of ten or more words and 60% of question-format searches. ### Do short searches get AI Overviews? Rarely in that data: 8% of one- or two-word searches produced a summary. ### How many sources does an AI summary cite? In Pew's data, 88% of summaries cited three or more sources. ### Is this still true today? Not necessarily. Semrush's 2025 data shows trigger rates and query types shifting, and Google has since expanded AI Mode, so treat Pew's figures as a March 2025 snapshot. --- Source: https://contentagents.dev/blog/which-searches-trigger-ai-overviews-what-pew-data-shows