www.mandalytics.com/guides/how-ai-tools-choose-sites/Guides
How do AI tools decide which sites to name?
The short answer
AI tools tend to name sites that other sites mention, that answer the question clearly in text they can read, and that look fresh. Each tool also has favourite sources, so there is no single ranking to climb. Most numbers below come from vendor studies, so read them as direction rather than law.
Who do AI tools listen to first?
For questions about a brand or a business, most citations come from other people talking about you: reviews, "best of" lists, press and directories. In one summary of the research, about 57% of citations for brand questions came from reviews, lists and press, 17% from directories, and only about 4.5% from the brand's own About or FAQ pages. A study of 216,000 pages by SE Ranking found that the number of sites linking to a page was the strongest single predictor of being cited by ChatGPT.
What to do: get listed where AI already looks. Find the review sites and directories your customers' questions lead to, claim your profile, and ask happy customers for detailed reviews. In a local business audit, the business AI Mode picked had 392 reviews, and the reason it gave came from what the reviews said, not the star rating.
What kind of content gets quoted?
- A direct answer first. Put the answer in the first two sentences, then explain.
- Question headings. A heading like "How much does it cost?" matches the way people ask.
- Tables and lists. Pages with tables were cited about 2.5 times more often in one study.
- Real numbers, quotes and named sources. In Princeton's GEO tests, adding statistics, quotes and sources lifted visibility by roughly 30 to 40%, while keyword stuffing did nothing.
- Specific facts. Prices, sizes, steps, who it suits. Use the swap test: if a competitor could paste your sentence unchanged, make it more specific.
Does freshness matter?
It seems to. Pages cited by AI tools are about 25% newer on average than Google's normal results, and ChatGPT and Perplexity lean towards pages updated within roughly the last month. What to do: refresh your key pages when something changes and show a visible "updated" date, as this page does.
Can AI tools read my pages?
Not always. GPTBot, ClaudeBot and PerplexityBot read only the raw HTML a page sends and do not run JavaScript (Vercel and MERJ measured this). If your text only appears after scripts load, they see an empty page. Gemini is the exception because it uses Google's renderer. Read more in Is your site open to ChatGPT and Perplexity?
Do all AI tools use the same sources?
No, and this is why chasing one score is a trap. For the same brand questions, ChatGPT, AI Overviews and Perplexity share only 2 of their top 10 cited sites (Agenzy). ChatGPT's web answers follow Bing's results closely (see Why Bing matters for ChatGPT), Perplexity tracks Google closely, and Claude is reported to use Brave. The honest way to measure is to ask the real questions in each tool and count how often you are named, which is what the Mandalytics report does every week.
Frequently asked questions
Does ranking on Google still matter for AI answers?
Yes, but less than before. In March 2026 only 38% of AI Overview citations came from the top ten results, down from 76% in mid-2025 (Ahrefs). Being mentioned by others and answering clearly matter alongside rankings.
Does schema markup get me cited by AI?
Not on its own. Studies saw no measurable lift from schema alone. Visible structure (clear headings, lists, tables) helped more. Schema is still worth adding when it matches what the page shows.
Is your site open to ChatGPT and Perplexity?
Why Bing matters for ChatGPT