Title and paragraphs that match the buyer's question
Within the same website, a title and paragraph that match the question people ask is the on-page property that survived robustness tests. Retrieval still explains far more citation variance than page features.
Best study is at least large-n observational or controlled, with some replication. Outcome data is still mostly correlational.
Toronto What Drives AI Citations? (arXiv 2610.15033, preprint): retrieval 32% of within-site variance vs 2.8% for all page features. Only query-matching title/paragraph survived all five robustness tests.
How to fix
Answer your buyers' questions in the words they use, not your internal terms. AI agents look for pages that match the question.
- List the questions buyers really ask (from calls, emails, searches and reviews) and add them to the free check, so it grades your questions, not our examples.
- Note the words buyers use: "unfair dismissal lawyer", not "employment dispute resolution services".
- Use those words in your title, your H1 and a heading per main question.
- Answer under each heading in a sentence or two that repeats the question's key words. The check ignores your brand name, so "Acme" alone never counts.
- Cover each question once. Repetition won't help and can trip the "Reads naturally" check.
How to check it worked: run the free check again with your own buyer questions. The "Your page uses your buyers' words" row should pass: "Your title and text match {n} of {total} buyer questions." It passes at 7 of 10 or more.
What the free check looks for
- Check title, H1 and paragraphs against the buyer questions (lexical overlap in the free check; not live fan-out embeddings).
Caveats
- — ChatGPT fan-out queries overlap only 2.7% between same-day runs (Toronto). The free check is a proxy, not the engine's retrieval set.
- — Included in the free page check.
Studies and docs
2026
Held the website fixed for 90 days and replayed ChatGPT's own searches. Retrieval of the page explains 32% of within-site citation variance; all 14 page features together explain 2.8%. Only query-matching title/paragraph survived all five robustness tests. FAQ, schema, author and summary boxes look useful across sites (Simpson) and are null within a site.
2026
Cited titles have higher cosine similarity to the prompt (0.602 vs 0.484) and to fan-out sub-queries. Vendor observational study.
See what AI agents can find on your site
Our free AI visibility check shows whether AI agents can reach, read and quote your site, with each fix graded by evidence. There's no overall score, and no one can promise citations.
Free. No password.