How to write FAQ schema that AI engines quoteUser-phrased questions. 60–150 word answers. Stable anchor IDs.
FAQPage JSON-LD is one of the highest-leverage structured-data additions you can make to an authority page. AI engines extract Q&A pairs from FAQPage schema and surface them in generated answers. This page covers the patterns that make FAQs quoteable — phrasing, length, anchor IDs, and the HTML-must-match-schema rule that protects you from rich-result penalties.
By Martin Yarnold · UpdatedQuoteable vs not
Quoteable
Avoid
Frequently asked questions
How should I phrase FAQ questions?
As actual user questions. "How does X work?" beats "X works like this:". AI engines retrieve based on intent match; user-phrased questions match user queries more often than statement-phrased ones. Avoid marketing-driven phrasing ("Why choose X?") in favour of operational phrasing ("How does X handle Y?"). The question is the retrieval key; write it the way the user would type it.
How long should each FAQ answer be?
60–150 words is the practical range. Too short and the answer is fragmentary — engines lift it but it does not stand alone. Too long and engines often truncate mid-answer, producing an incomplete citation. The sweet spot is one substantive paragraph that answers the question on its own merits and reads cleanly if quoted in full.
Why do FAQ items need stable anchor IDs?
Two reasons. First: deep linking. When an AI engine cites a specific FAQ, having a stable id="faq-X" lets the citation deep-link to the answer rather than the page top — better user experience and higher attribution clarity. Second: the @id in JSON-LD FAQPage schema can reference the same anchor, which keeps the structured-data view and the rendered view aligned. Once an anchor is published, never rename it; external citations will break.
Which engines actually use FAQPage schema?
Google publicly documents FAQPage rich-result eligibility, though display has changed over time as Google has tightened rich-result criteria. Other engines parse the same schema; observed behaviour suggests AI engines (ChatGPT, Claude, Perplexity, Gemini) all extract Q&A pairs from FAQPage JSON-LD into their retrieval indexes. Specific extraction rules per engine are not publicly documented; treat FAQPage as broadly useful structured data, not a guaranteed signal to any specific engine.
How many FAQ items per page?
6–12 is the practical range for a Tier 1/2 authority page. Fewer than 6 underfills the answer surface; more than 12 starts producing diminishing returns because each item dilutes the others in engine retrieval. The Promagen Sentinel authority pages target 6–10 per page; quality of question selection matters more than count.
Should the FAQ also render as visible HTML on the page?
Yes, always. Google's structured-data guidelines require that FAQPage schema describe content that is actually visible on the page. Rendering only JSON-LD without visible Q&A is a violation that can disqualify the page from rich results and is observably penalised by AI engines too. The pattern that works: render both the JSON-LD block (for engines) and the same Q&A pairs in HTML (for humans and engine HTML parsing).