Data Studies
Research on how ChatGPT, Gemini, Perplexity, Google AI Mode and Google AI Overviews answer the questions your customers ask: which brands they name, which pages they cite and link to, and where they send shoppers. Most studies are built on thousands of answers we collect from those engines, including the web searches ChatGPT runs behind the scenes. Many compare those answers with Google search and Google Maps results, Google Business Profiles, Google News, or robots.txt files from the public HTTP Archive crawl. Use them to see where AI answers pull from in your market.
122 studies · page 3 of 6 · Need help acting on them? See our AI SEO services.
When people ask AI engines for the best car insurance, 99.8% of the answers that name an insurer include USAA, GEICO, State Farm, Progressive or Erie, 474 of 475. Across car, life and pet insurance, the same five companies appear in 77.1% of AI insurance answers.
Perplexity links to Booking.com or an Expedia Group site in 77.1% of its answers to hotel questions, while Google AI Mode does so in 1.6%. Hotel brands' own websites are cited slightly more often than the booking giants, and Tripadvisor, cited in 20.8% of hotel answers, almost disappears from AI answers about flights and airports.
1.5 million websites, 9.9% of the 15.5 million in the public HTTP Archive crawl, block at least one AI crawler. They block the bots that collect training material about four times as often as the bots that fetch pages for live AI answers.
14.1% of the 1,000 most visited websites block Perplexity's AI crawler, 12.3 times the 1.1% of websites ranked outside the top 10 million. The same divide holds for OpenAI's GPTBot, blocked by 23.1% of the most visited websites against 8.3% of small sites.
67% of frequently cited websites that block GPTBot, OpenAI's training crawler, still appear as sources in ChatGPT answers, close to the 73% of all frequently cited websites. Blocking GPTBot trims a site's ChatGPT citations; it does not remove the site from ChatGPT.
29.3% of home pages built with a JavaScript framework deliver fewer than 200 characters of readable text to AI crawlers that read the page as the server sends it, twice the 14.4% of other home pages. On 17.5% of them, the main heading is missing until JavaScript runs.
Questions about investing draw an estimated 5,666 AI searches for every 1,000 Google searches, 5.7 times their Google volume. Money as a whole goes the other way: it makes up 16.2% of estimated AI search demand for 34,571 US keywords, against 26.3% of Google demand for the same keywords, because Americans still take money decisions to Google.
Searches from people researching or ready to make a purchase make up 27.3% of AI search demand for 6,999 US industry keywords, against 51.2% of Google search demand for the exact same keywords. People go to AI to learn and to Google to buy.
US Google search interest in "claude ai" grew roughly 347% over the past 52 weeks and "google gemini" roughly 278%, against about 24.7% for "chatgpt". ChatGPT is still searched about 19 times as often as Claude, but its search interest peaked in September 2025 and has leveled off since.
Americans ask AI tools "what is the best car" an estimated 280,393 times a month, a modeled figure that counts every similar phrasing of a short, broad question, against 2,900 Google searches for the exact question. Questions that ask for a judgment, such as the best car, job or stock, have moved to AI, while how-to questions like "how to lose weight" still get more searches on Google.
61.2% of Gemini's answers to everyday questions cite no source at all, against 2.2% of ChatGPT's. Across five AI engines, 21% of answers give the reader nothing to click, and on recipe questions Gemini leaves out sources 93.3% of the time.
A ChatGPT answer cites 3.7 sources on average, the fewest of five AI engines, while a Perplexity answer cites 20, more than five times as many. On ChatGPT, a website has to be one of a handful of sources to get cited at all.
53.7% of AI answers to "best product" questions that name a brand do not link to any of the brands they name. The average answer names 4.9 brands and puts a clickable link on 1.2, and Gemini names 5.6 brands per answer while linking almost none.
Only 0.9% of the sources cited by ChatGPT, Gemini, Perplexity and Google's AI features are a website's homepage. The other 99.1% point to a specific inner page, and the average cited page sits 2.6 folders deep in its website.
When ChatGPT and Google AI Overviews answer the same question, only 8.4% of the websites they cite are the same. Across five AI engines, any two share 15.4% of their sources on average, so being cited by one engine says little about the others.
In 70% of AI answers that cite two or more sources, one website is cited with two or more different pages. On Gemini and Perplexity the rate passes 91%, and YouTube is the website most often cited several times in one answer.
5.7% of the sources ChatGPT cites are PDF documents, against 0.32% for Google AI Overviews and 0.42% for Google AI Mode. Across five AI engines, 1.5% of citations are documents rather than web pages, and 57% of the PDFs cited come from government, university or military websites.
Only 7.8% of the sources Google AI Mode cites are linked inside its answer text; the rest sit in a separate source list. ChatGPT links 100% of its sources in the text, Gemini 94.6% and Perplexity 45.7%.
99.9% of the links ChatGPT cites carry a tracking tag, utm_source=chatgpt.com, that names ChatGPT as the source of the visit in your analytics. Google AI Overviews and Perplexity tag none of their links, and together with Google AI Mode they account for 82% of all AI citations.
ChatGPT retrieves 12.8 search results per answer and cites 2.9 sources, but only 0.3% of the retrieved results are the exact pages it cites. ChatGPT finds a website through search, then cites a different page on it: 68% of the websites it cites appear among the websites it retrieved.
Asked for the best product or service, Gemini answers "it depends" without naming any pick in 15.3% of its answers, more than three times the 4.2% of Google AI Overviews. In finance, AI engines name an explicit pick in just 21.1% of answers.
Not one of the 7,048 shopping product cards ChatGPT showed lists Amazon as the store. Walmart is the store on 15.8% of them, and ChatGPT shows product cards in 45% of its answers to "best product or service" questions.
Perplexity cites at least one government, military or university website in 58.1% of its answers to everyday questions, ahead of ChatGPT at 49.8% and far ahead of Google AI Mode at 26.7%. On tax questions, 71.7% of AI answers cite an official source.
Ask for the best product and ChatGPT links to a retailer or shows shopping cards in 45.9% of its answers, sixteen times Gemini's 2.8%. On everyday questions, Perplexity is the engine most likely to send you to a store, and on product questions Amazon is the store it links to most.
Make Search Your Best Channel
Get a tailored growth plan for your site, backed by a senior team with 20+ years shipping results.
Get Your Free Proposal