FAQ pages that AI assistants quote: what the evidence shows

Write each FAQ heading as the literal question a buyer types, then answer it in one or two sentences that stand alone. FAQ markup alone has not shown a measurable citation gain: an Ahrefs test across 1,885 pages found no measurable effect on citations in ChatGPT or Google AI Mode. The visible, quotable text is the part worth testing.

• Write each heading as the question a buyer types, and open the answer with one or two sentences that work on their own.
• FAQ markup alone showed no measurable citation effect in the controlled Ahrefs test.
• Google deprecated FAQ rich results on May 7, 2026, so FAQ schema no longer produces dropdowns in Google Search.
• Change one thing at a time and re-check the same prompts across ChatGPT, Gemini, Perplexity, Claude, Grok and Google AI.

Does FAQ schema still matter for AI search?

Not as a search display. According to Passionfruit's summary, Google deprecated FAQ rich results on May 7, 2026, so FAQ dropdowns no longer appear in Google Search.

FAQPage is still a valid Schema.org type. Digital Applied quotes Google's position: structured data that isn't used causes no problems for Search, but has no visible effect there. Elementera adds that Google says no special schema is needed for AI Overviews or AI Mode, and that any markup should match visible page content. Squin's guide makes the same point: markup must describe content users can see, so FAQ schema can only repeat questions and answers already on the page.

The citation data is mixed. An SE Ranking study of 129,000 domains found that pages with FAQ schema averaged 3.6 ChatGPT citations, against 4.2 for pages without it. The study is observational, so it doesn't show that schema caused the gap. The controlled test is more direct:

"The changes for Google AI Mode and ChatGPT were statistically indistinguishable from zero." — PleaseCiteMe, reporting the Ahrefs test (the PleaseCiteMe write-up)

GreadMe's blog claims FAQ schema strongly boosts AI assistant and voice search citations. That conflicts with the controlled result, so treat it as a claim, not a finding.

Should the heading be the question a buyer types?

Yes. Loudface advises writing the heading as the literal question a buyer types, not a clever rewrite, because the heading is what gets matched to the prompt. "Why choose us?" is not something a buyer types.

Matching wording may matter. Loudface cites an Ahrefs analysis of 1.4 million ChatGPT prompts in which cited pages had a title-to-prompt similarity of 0.602, against 0.484 for pages that were retrieved and ignored. That is a correlation, not proof that wording caused the citation.

To find the wording, use phrasings people already type. Contently recommends pulling exact phrasing from autocomplete, People Also Ask boxes and assistant follow-up prompts.

What does a good FAQ entry look like?

Take a weak entry. The heading reads "Is FAQ schema worth it?" and the answer opens "Great question, and it depends on many factors." The heading is a rewrite, and the answer needs context from elsewhere on the page.

A stronger version uses the buyer's wording as the heading: "Does FAQ schema still help AI assistants cite a page?" The answer opens with a sentence that stands alone: "FAQ schema no longer produces dropdowns in Google Search since Google deprecated FAQ rich results on May 7, 2026." The second sentence carries the test result: "An Ahrefs test found no measurable citation effect in ChatGPT or Google AI Mode."

The reason is that AI engines extract answers in isolation. Loudface's advice is one or two sentences that need nothing else on the page for context. Put the detail after them.

How long should each answer be?

Guides disagree. Growth Rocket recommends 40 to 60 words, another guide says 40 to 80, and a third says under 100. None of these ranges is backed by a published controlled test.

Treat them as starting points and test them on your own pages. The part with support is the structure: a standalone opening, then the detail. A longer answer is fine if its first sentences answer the question on their own.

How many entries, and where on the page?

Counts conflict too. MQL Magnet recommends eight to twelve question and answer pairs. Another guide recommends three to six and says more than six dilutes signal quality, without citing evidence. Keep the entries to questions your buyers actually ask, and don't add more only to reach a number.

On placement, MQL Magnet advises putting the FAQ near the bottom of an article, just before the closing call to action. That is a best-practice recommendation from a blog, not a tested finding.

How often should FAQ answers be updated?

Regularly. According to SE Ranking's analysis, as reported by NovaStacks, 95% of ChatGPT citations go to content published in the last 10 months. Review prices, dates, policies and figures in your answers on a schedule, and change the visible text when they move.

How do you test whether an FAQ gets quoted?

Build a list of prompts from autocomplete, People Also Ask and follow-up questions. Ask each assistant the same prompts: ChatGPT, Gemini, Perplexity, Claude, Grok and Google AI. Record whether your FAQ text is quoted, linked or ignored.

Measure the traffic separately. Citable's guide to tracking AI assistant traffic in GA4 covers the setup. For how one assistant picks pages, see how Perplexity chooses its sources.

Change one thing at a time, such as a heading or an opening sentence, and run the same prompts again. If you change several things at once, you can't credit a shift in citations to any one edit.

What does the evidence still not show?

Nothing in the material above isolates the FAQ format itself as a cause of AI citations. The controlled Ahrefs test found no measurable schema effect on ChatGPT or Google AI Mode. For AI Overviews, PleaseCiteMe reports a 4.6% decline relative to controls that Ahrefs could not attribute to schema alone.

The length and count ranges are hypotheses, and vendor claims about schema remain claims. Your own prompt tests are the best evidence you can get for your own pages.

Next steps

• Collect 20 real buyer questions from autocomplete, People Also Ask and assistant follow-ups, and use their wording as your headings.
• Rewrite the first one or two sentences of every answer so they make sense with nothing else on the page.
• Keep any FAQ markup identical to the visible text, and don't expect it to lift citations on its own.
• Run your question list across the assistants, log the results, change one element at a time, and review answers at least every few months.
• If you want that check run weekly, Citable measures whether ChatGPT, Gemini, Perplexity, Claude, Grok and Google AI cite or recommend a site. Its guide on getting cited by Claude shows what one assistant looks for.

Frequently asked questions

Does FAQ schema still matter for AI search after Google removed FAQ rich results?

Google deprecated FAQ rich results on May 7, 2026, so schema no longer produces dropdowns in Search. In an Ahrefs test across 1,885 pages, schema had no measurable effect on citations in ChatGPT or Google AI Mode. If you keep the markup, it must match the visible text.

How many questions should an FAQ page have?

Guides disagree: one recommends eight to twelve pairs, another three to six. Neither cites a controlled test. Include the questions your buyers actually ask and no filler.

How long should each FAQ answer be for AI assistants?

Open with one or two sentences that stand alone, then add detail.

Should FAQs go at the top or bottom of a page?

One blog advises placing them near the bottom, before the closing call to action, but that is a best-practice opinion, not a tested result.

Sources

• Does FAQ Schema Increase AI Citations? What Tests Show About Effective FAQs
• FAQ Rich Results Deprecated: Google's May 2026 Change
• Structured Data After I/O 2026: Schema Cheat Sheet
• Schema Markup SEO: The 2026 Guide to Structured Data
• How to Optimize for ChatGPT: Skip LLMs.txt, Earn Trust on Quora & Reddit
• FAQPage Schema: Still Worth It? JSON-LD Example & Rules
• How to write FAQs that AI search engines actually extract
• How to Write FAQs That LLMs Actually Cite (2026 Best Practices)
• How to Write FAQ Content That AI Engines Pull From
• FAQ Section: How to Write FAQ Sections That Rank and Get Cited by AI
• What Drives AI Citations? Data from 129,000 Domains (2026)