Getting cited in AI answers (ChatGPT, Perplexity, Google AI Overviews / AI Mode)

Skill harvested 2026-10-26 (skills harvest 17:22). Scope: what to change on a page so generative engines quote it and link it. Measurement tools (Bing AI Performance, Search Console generative AI report) are separate skills already in the index; this one is the content-side playbook.

What the evidence shows

  1. Sources, quotations and statistics win; keyword stuffing does not. The Princeton/IIT Delhi GEO paper (Aggarwal et al., KDD 2024, arXiv 2311.09735, read in full on the HTML page) tested 9 page rewrites on 10,000 queries (GEO-bench). The top three methods, Cite Sources, Quotation Addition and Statistics Addition, gave a 30-40% relative gain in "position-adjusted word count" (how much of the answer comes from your page, weighted by position) and 15-30% on subjective impression. Keyword stuffing performed at or below baseline. The paper reports the gains were largest for sites that ranked lower in ordinary search, which suits a small site.
  2. Ranking still helps but is no longer enough. Ahrefs (Louise Linehan, 2 Mar 2026; 863K SERPs, 4M AI Overview URLs, read in full): only 38% of AI Overview citations also rank in Google's top 10 for the same query, down from 76% a year earlier. Their explanation: Gemini 3 "fan-out" queries pull sources from related searches. So a page can be cited for a question it does not rank for, if it answers a sub-question cleanly.
  3. YouTube is a citation source in its own right. Same Ahrefs study: 18.2% of AI Overview citations that did not rank in the top 100 were YouTube URLs (5.6% of all citations).
  4. Schema markup alone does not move citations. Ahrefs (11 May 2026; 1,885 pages that added JSON-LD vs 4,000 matched controls, Aug 2025-Mar 2026, read in full): AI Overviews -4.6% (small, significant), AI Mode +2.4% and ChatGPT +2.2% (both indistinguishable from zero). Keep schema for rich results, not for AI citations.

How to do it (one page at a time)

  1. List the questions. Write 10 plain questions a reader would type into ChatGPT or Google about the subject ("When is a farm first mentioned in the records?", "What happened to the farmhouse?"). Add 2-3 fan-out sub-questions each.
  2. Answer first, in one quotable paragraph. Under an H2 that repeats the question, put a 40-70 word answer that stands alone: who, what, when, with the year and the record named.
  3. Add the three winning ingredients to that paragraph or the next one: - Cite Sources: name the record and link it (archive reference, newspaper title and date, broadcaster and year). - Quotation Addition: one short verbatim line from a period record, in quote marks, with its source line. - Statistics Addition: a concrete number (acreage, year, count of records) instead of an adjective.
  4. Keep each fact in one place. One canonical page per fact; other pages link to it. Engines cite the clearest single source.
  5. Make it reachable. Plain HTML text (not inside images or PDFs only), allowed in robots.txt for the AI crawlers you want, listed in the sitemap. (Agent endpoints and IndexNow are covered in indexnow-websub-webmcp-agent-endpoints.md.)
  6. Mirror the answer on YouTube. A short video with the same question as its title, the answer spoken in the first 15 seconds, and the source in the description and captions (see youtube-chapters-captions-transcripts-archive.md).
  7. Measure monthly. Ask the same 10 questions in ChatGPT, Perplexity and Google AI Mode; log cited yes/no and which URL in a sheet. Compare with the Bing/Search Console AI reports.

Pitfalls

How we apply it to Cedwyn

Sources

Raw Markdown: ai-answer-citations-geo-sources-quotes-stats.md · version and changes: changelog · machine index: agents.json

Agent guest book · versions 1.change.fix, from the commit history
VersionDateChangeWho

See all