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 4 of 6 · Need help acting on them? See our AI SEO services.
44.9% of the 10,000 websites ChatGPT, Gemini and Google AI Mode cite most often are low-authority sites, and together they collect 10.1% of all AI citations. Being cited does not take a famous domain, though the biggest sites still win the most: the top 3.6% by authority take 34.7% of citations.
Websites registered more than 20 years ago collect 79.8% of the citations ChatGPT, Gemini and Google AI Mode give out, four in every five. Young sites still get in: 8.7% of cited websites are less than two years old, but together they collect 1.45% of AI citations.
66% of the sources ChatGPT, Gemini, Perplexity and Google's AI features cite in their answers are .com websites, against 14% for government (.gov) and university (.edu) websites. ChatGPT is the one AI engine that cites government and university websites more often than .com.
Ask Perplexity an everyday question from the United States and 49.3% of the time its answer cites a website from another country, against 5.5% for ChatGPT. Across five AI engines, 14% of answers to US questions cite at least one foreign site, led by Britain's NHS and BBC.
ChatGPT sends 7.9% of its citations to websites that get fewer than 100 Google visits a month, six times the 1.2% rate of Gemini and Google AI Overviews. Across all AI engines, 7.5% of the 10,000 most-cited websites have almost no Google audience, one in thirteen.
62% of AI answers to "best software" questions cite the website of a vendor they recommend, so the vendor's own pages help make the case for it. One answer in six cites the vendor's own blog or guide, and on Perplexity it is 48%.
70% of AI answers to "best product" questions cite at least one review or comparison site that earns commissions on sales. Perplexity cites one in 99.4% of its product answers, and finance questions rely on them most.
Ask Gemini whether a brand is any good and 72% of its answers do not cite the brand's own website, leaning on review sites instead. Across ChatGPT and Gemini, one answer about a brand in three leaves the brand's website out.
One company, Microsoft, supplies 51% of the help center and documentation pages AI engines cite when answering software and tech questions. Help pages make up 10% of all AI citations on software topics, and more than one in five on ChatGPT.
AI engines cite LinkedIn in 1.6% of their answers to business, careers and software questions, fewer than one answer in 60. Gemini did not cite LinkedIn once in 1,136 business answers, and LinkedIn posts were cited 15 times in 4,966 answers.
The brand an AI engine names first in a product recommendation gets 9.1% fewer words than the brands named after it, 30 words against 33 on average. On Perplexity the first brand gets 18.4 words against 26.3 for the rest.
53.7% of AI answers to "best product" questions never mention a price, so more than half of AI shopping advice leaves out what the product costs. ChatGPT quotes a price in 69.2% of its product answers and Perplexity in 14.1%.
93.3% of the brands AI engines recommend in product answers come with no downside at all: no con, no drawback, not even a "but". On Perplexity the figure is 97.4%.
Ask ChatGPT or Gemini "What is [brand]?" and 7.2% of top brands, about one in 14, get confused with something else or not recognized at least once. Brands named after surnames and places suffer most: Bryant was confused in all four of its answers.
Ask ChatGPT "What are the best alternatives to [brand]?" twice, a few hours apart, and the top three competitors it names change for 82.9% of brands. Gemini changes them for 93.6%.
The average Google AI Mode answer runs 315 words, more than twice the 141 words of the average Perplexity answer to the same everyday questions. Across five AI engines, the average answer is 229 words long.
89.3% of Google AI Overviews are mostly bullet points, against 37.5% of ChatGPT answers to the same questions. Across five AI engines, 60.5% of answers are mostly lists and 91.6% contain at least one.
80% of AI answers to everyday questions end by asking for more detail or offering another step, four answers in five. Google AI Mode does it in 90.2% of its answers.
97.9% of AI answers to everyday questions use none of the standard disclaimer phrases, such as "not medical advice" or "consult a professional". On health questions 0.8% of AI answers use one, and on legal questions 0.5%.
Google AI Mode lays out products side by side in a comparison table in 40.2% of its answers to "best product" questions, 31 times the 1.3% rate of Google AI Overviews. Across five AI engines, 21.7% of product answers include a table.
Google AI Mode illustrates 72.5% of its answers to "best car" questions with images, against 9.5% of its answers overall. Money questions almost never get a picture: 0.7% of mortgage answers include one.
Ask ChatGPT for the best plumber, dentist or restaurant in a US city and 95.1% of the time it answers with its built-in list of local business cards, holding about 12 businesses. Gemini, asked the same questions, showed no such list.
AI answers to health questions read at a grade 11.3 level, close to a high school senior's, the hardest of any topic and four and a half grades above recipe answers. On Gemini, health answers read at grade 12.6. The average AI answer reads at grade 9.3.
When a shopper spells out what they care about, ChatGPT cites 35.8% more websites in its answer: 4.21 sources per answer for a detailed product question against 3.10 for the short version. The detailed question also changes ChatGPT's top recommendation three times in four.
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