Why AI cites pages with statistics: what the research shows, and how to publish yours
AI systems look for numbers when they write, and a clear, sourced statistic travels far. What the latest data on AI search says, why it works, and a practical guide for any business.
Short answer: when AI writes an article, a report or a slide, it searches for numbers to back up its sentences, and it tends to pick pages where a specific figure sits next to a source and a date. Publishing real, well-presented statistics about your field is one of the most reliable ways to be quoted and linked by AI-written content. It works for a skincare brand, a dental clinic or a roofing company alike, as long as the numbers are true.
Why this matters now
AI is where a growing share of research starts and ends. About 24% of ChatGPT messages are people looking for information (OpenAI, 2025). When Google shows an AI summary, only 8% of users click a classic result, against 15% without one, and just 1% click a link inside the summary (Pew Research, 2025). And 60% of searches now end without any click at all (Bain & Company, 2025).
When people do arrive from AI, they are worth more: AI-referred visits to US retail sites grew 393% in a year and convert 42% better than other traffic (Adobe, 2026). Yet only 6.8% of US websites are cited by AI at all (Similarweb, 2026). Being the source AI quotes is a real, and still uncrowded, opportunity.
Why a number travels further than a sentence
- More content is drafted with AI. Blog posts, reports, pitch decks and newsletters increasingly start as an AI draft.
- The draft needs evidence. When the model lacks a fact, it looks it up, with searches such as '[topic] statistics 2026'.
- A specific number is the most useful find. 'The market is growing' adds little; 'X% of patients do Y (Source, 2026)' is quotable, specific and looks like evidence.
- Figures are rarely cross-checked. An answer is written in seconds, so a well-presented number often goes straight into it. Honest sourcing matters for exactly that reason.
- Recent data wins. Those searches often include the current year, so a clearly dated page is the one that matches.
How AI uses what it finds
ChatGPT handles about 2.5 billion messages a day and has around 900 million weekly users (OpenAI). Google shows AI Overviews to 2 billion people a month and AI Mode to 100 million (Alphabet). Behind those answers, AI reads the web differently from people:
- AI quotes, people don't click. Sessions end right after an AI summary 26% of the time, against 16% after a normal results page (Pew Research, 2025). If your number is in the summary, it reaches the reader even without a visit.
- AI summaries cite several sources. 88% of Google's AI summaries cite three or more sources (Pew Research, 2025), so there is room for a specialist next to the big names.
- More and more crawling is for answers. About 18% of AI crawling is for search and 2% for live user requests (Cloudflare, 2025): pages fetched at the moment someone asks.
- Specific beats general. A passage with a clear figure, a source and a year is the easiest thing for a model to lift into an answer or a draft.
According to Adobe (2026), AI-referred visits to US retail sites rose 393% year over year, and they converted 42% better than other traffic.
What controlled research says
A controlled study of 'generative engine optimisation' (Aggarwal et al., KDD 2024) rewrote pages in nine ways and measured how much each version was used in AI answers. Adding quotations raised visibility by 41% and adding statistics by 31%, with citing sources close behind; for sites ranked fifth, citing sources lifted visibility by 115%. Stuffing keywords did not help. Our summary of the study.
Publishing statistics AI can use
- Make each figure stand on its own. AI picks up a sentence or two, not the whole page, so write every statistic so it still makes sense when quoted alone.
- Name the source and year in the same sentence. 'According to [Source] (2026), X%…' helps the model, the reader and your credibility.
- Use headings your customers would search for. '[Topic] statistics 2026', '[Topic] cost in [city]', '[Treatment] success rate'.
- Date the page and keep it current. An outdated figure loses to a newer one.
- Keep the text readable for crawlers. Put figures in plain text, not in images, tabs or scripts, and make sure robots.txt and your firewall don't block AI crawlers.
- Favour a clear layout. A simple list of figures works better than a heavy design.
- Add your own data where you can. Collected figures earn some citations; figures only you have make you the original source.
What original data can look like in your category
| Business | Data you already have | Example of a citable statistic |
|---|---|---|
| Dental clinic | Treatment records, booking data | 'Our average Invisalign treatment took [N] months across [N] patients in 2025.' |
| Home services | Call logs, job records | 'Our median emergency response time in Denver was [N] minutes in 2025.' |
| Ecommerce brand | Orders, reviews, customer surveys | '[N]% of our customers with sensitive skin reported [result] in our 2026 survey of [N] buyers.' |
| Agency | Client results, audits | 'Across [N] client sites we audited in 2026, [N]% blocked at least one AI crawler.' |
Brackets mark where your real figures go. Publish only numbers you can back with data, and say how they were measured.
Checking whether it works
- AI bots in your server or CDN logs: about 18% of AI crawling is for search and 2% for live user requests (Cloudflare, 2025); look for retrieval agents such as ChatGPT-User, OAI-SearchBot and PerplexityBot, not only training crawlers.
- AI referrals in analytics: sessions from chatgpt.com, perplexity.ai and similar sources.
- Repeated AI questions: ask the questions your page answers several times on each engine and count citations. One answer proves nothing (why).
- New sites linking to the page, visible in your SEO tool.
- Your numbers elsewhere: search your exact figures in quotes to see who uses them, with or without a link.
Ethics, and what this tactic won't do
Never publish numbers you can't support. Invented figures can spread far once AI starts repeating them, and in health, finance or safety they can harm people. Accurate, sourced statistics do the opposite: they make the information people get from AI better.
It is also one tactic, not a strategy. Whether AI recommends your business depends on more than statistics: being in the sources AI trusts for your category, saying what you offer in your customers' words, and being readable by AI crawlers. And because AI systems change, what works today needs measuring again tomorrow.
Sources
- Aggarwal et al., 'GEO: Generative Engine Optimization', KDD 2024.
- OpenAI (2025–2026): ChatGPT usage and weekly users; Alphabet (2025): AI Overviews and AI Mode users.
- Adobe (2026): AI-referred retail traffic and conversion. Bain & Company (2025): zero-click searches.
- Pew Research Center (2025): clicks and sessions with Google AI summaries. Similarweb (2026): AI citation presence. Cloudflare (2025): AI crawling.