FAQ Schema for AI Answers: Still Worth It?
FAQPage markup in the AI era — what the evidence supports, and the Q&A formatting that helps retrieval with or without the schema.
FAQ schema is the most requested AEO tactic and the most oversold. The honest version: the markup is cheap and correct implementations rarely hurt, the Q&A content is valuable to retrieval on its own merits, and the two claims should not be conflated. This article separates them. The content-side argument is in AEO vs SEO: What Changes and What Stays, and the audit that tells you whether you need it at all is in The AI Answer Audit Checklist.
What FAQPage markup does now
FAQPage markup once earned pages a collapsible set of questions and answers directly in Google's search results. That era is over for most sites: Google has since restricted the display of FAQ rich results to a narrow set of well-known, authoritative government and health sites. For the rest of the web, the classic search-side payoff is gone, and any pitch that still leads with "FAQ rich results" is describing 2022.
What the markup still does is unambiguous: it labels pairs of questions and answers so any machine — search crawler, answer engine, scraper — can parse them without guessing from formatting. That is real, but it is parsing help, not a ranking promise.
The evidence on schema and AI answers
No major answer engine has publicly documented that FAQPage markup is a direct input to its citations or recommendations, and you should treat any specific percentage claim to the contrary as marketing. The plausible reading of what we can observe: answer engines mostly consume rendered text, well-structured Q&A content helps because of what it is, not because of the tags around it — while structured data adds precision for pipelines that do use it.
The correct posture is cheap experimentation. FAQPage markup costs an hour per page and breaks nothing when done properly, so implement it on a few key pages, hold the rest as a control, and let your own measurements decide — the routine for that comparison is in How to Track Your Brand Mentions in ChatGPT, Gemini and Perplexity. What you should not do is spend budget on schema at the expense of the content itself.
Schema elsewhere in the stack is a different question. Organization, Product and Article markup keep the facts about your entity unambiguous, and that work feeds the same consistency layer recommendation answers depend on — it stays worthwhile regardless of what you decide about FAQPage. The distinction to hold: entity markup supports what models believe; FAQPage merely packages question-and-answer text more tidily.
Q&A formatting that helps regardless
The formatting does the heavy lifting, with or without markup:
- Real questions as headings — phrased the way buyers ask, not the way marketing writes; retrieval matches on that phrasing.
- Answer in the first sentence — then nuance, then example; the passage has to survive being lifted alone, as explained in How AI Answers Pick Their Sources.
- One question per block — no blended headings, so each chunk carries a single intent.
- Visible text that matches the markup — the long-standing schema requirement still applies: never mark up answers the page does not show.
- Keep it current — a stale FAQ block is worse than none, because it feeds engines yesterday's prices with today's confidence.
Sourced from where it should be: mine support tickets, sales calls and the related-questions that engines suggest, then write the answers properly. Five genuine questions beat thirty fabricated ones — thin Q&A sections read as filler to humans and to models alike.
A 30-minute implementation, one page
- Pick one page that targets a real buyer question (the audit's fix list names it).
- Write three to five genuine Q&A pairs from support and sales language.
- Add the block visibly on the page, question as heading, answer-first.
- Add FAQPage JSON-LD that mirrors the visible text exactly — same questions, same answers.
- Validate the markup with a structured-data testing tool, then re-run your frozen prompt set and compare against the control pages.
Expectations for that comparison: movement is usually slow, and two measurement cycles is a fair trial. If the marked-up pages show no difference after that, drop the tactic there and put the hours into content. The experiment was cheap, but it was not free.
Keep reading
- What Is Answer Engine Optimization (AEO)? — where the "make pages liftable" step and FAQ blocks fit the plan.
- How AI Answers Pick Their Sources — why self-contained chunks are the real currency.
- The AI Answer Audit Checklist — the twenty items that decide whether you need any of this.