FAQ and FAQPage Schema: Answers AI Engines Cite

Two GEO score criteria are among the easiest to improve: answer structure (40 out of 100 points) and evidence and data (15 points). Together, these two criteria make up 55% of the GEO score. And the fastest way to boost both is often the same: a well-crafted FAQ page paired with FAQPage Schema markup.

The problem: most website FAQs are graveyards of corporate questions nobody actually asks. "Why choose us?" is not something people type into ChatGPT. And a poorly implemented FAQPage Schema with 8-word answers or invisible content adds zero value.

This guide shows you how to build FAQs that AI engines actually extract and cite, and what FAQPage Schema does (and does not) add to that content.

Why the FAQ + FAQPage Schema combo is so powerful for GEO

AI engines work by extracting answers. When someone asks ChatGPT or Perplexity a question, the AI searches its sources for a passage that answers directly, as factually as possible, in a self-contained format. That is exactly what a well-written FAQ provides: a clear question followed by a standalone answer.

The question-answer format is natively extractable. The AI does not need to guess where the answer starts or ends. Each Q&A pair is an independent block that can be cited as-is. It is the easiest format for an LLM to consume. It is also the native format of voice assistants: see voice search and AI assistants.

In terms of the Detekia GEO score, the combo directly impacts two criteria:

• Answer structure (40 points): the FAQ provides question-style headings, each followed by a short, standalone answer. This is the ideal format for AI citation.
• Evidence & data (15 points): factual answers (timelines, numbers, dates, linked sources) are exactly the evidence this criterion looks for.

That is up to 55 potential points out of 100. FAQPage Schema is not a criterion in itself: it describes to machines "this is a question and here is its answer." It is a useful complement, not a decisive lever: Google says no special markup is needed to appear in its AI features, and what separates cited pages in our 317-page study are visible elements: question headings, short answers right below them, tables.

→For a full overview of JSON-LD schemas to implement: Schema.org and AI — the practical guide

What an AI-optimized FAQ actually looks like (not the ones you see everywhere)

Quick test. Open the FAQ page of any corporate website. You will probably find:

• "Why choose us?" → a 3-line marketing answer with zero data
• "How does it work?" → a vague answer that links to another page
• "What are your advantages?" → a list of superlatives ("the best", "the most reliable")

These are not FAQs. They are marketing brochures disguised as questions. AI engines ignore them because they contain nothing citable: no verifiable fact, no number, no standalone answer.

5 characteristics of a good AI question

1. Natural user language: phrased the way a real person would ask ChatGPT. Not "Our return policy" but "What is the return deadline and is it free?"
2. Subject + action structure: the question contains a clear subject and a specific action. "How," "How much," "What is," "Does."
3. Specific: "How much does a yoga class cost in New York in 2026?" rather than "What are your prices?"
4. Contextualized: include geography ("in New York," "in the US") or time ("in 2026") when relevant.
5. Single and direct: one question, one answer. Avoid double questions ("How and why...").

5 characteristics of a good AI answer

1. Standalone: understandable without reading the rest of the page. If you pasted the answer into a message, it would make sense on its own.
2. Short first: one to three sentences (15 to 70 words) that answer, like the answers placed under headings on the cited pages in our study. Details can follow in a second paragraph. 8-word answers are useless; 400-word answers will never be extracted as a block.
3. Lead with the essential: the first sentence gives the answer. Details come after.
4. Factual: at least one number, duration, price, or verifiable fact. "Our classes last 60 minutes with a maximum of 15 participants" rather than "Our classes are adapted to all levels."
5. No marketing fluff: ban "we are the best," "our recognized expertise," "trust our passionate team."

How to find the right questions (and avoid the wrong ones)

The first mistake is making up questions in a conference room. Real questions come from your users, not your marketing team. Here are 5 concrete sources:

1. Google Search Console: Performance tab → filter queries containing "how," "what," "does," "how much." These are the real questions people already ask when they find your site.
2. People Also Ask (PAA): type your main keyword in Google and look at the "People also ask" questions. Google shows them because people ask them. AI engines ask them too.
3. AnswerThePublic: enter your topic and get the most common questions, organized by type (how, why, when, how much).
4. Actual customer questions: support emails, live chat messages, sales call questions. These are the most authentic questions — and often the most specific.
5. Ask ChatGPT directly: "What are the 10 most common questions about [your topic]?" The result shows you what the AI considers the key questions in your domain.

Watch out for traps: avoid questions that are too vague ("How do I grow my business?"), questions you cannot actually answer factually, and purely promotional questions ("Why are we the best?"). If the answer does not contain at least one verifiable fact, the question is not worth including.

FAQPage Schema: technical implementation

FAQPage Schema is JSON-LD markup that explicitly tells AI engines: "this page contains structured question-answer pairs." It does not replace the visible structure: question headings followed by a short answer are already readable without markup, and Google says no particular schema is required for its AI Overviews. The markup makes that structure explicit; it is a plus, not a prerequisite.

Complete JSON-LD code

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is the shipping time in the US?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Standard shipping within the continental US takes 3 to 5 business days via USPS. Express shipping (1-2 days) is available for $12.99. Orders placed before 2 PM EST ship the same day. Free shipping on orders over $75."
      }
    },
    {
      "@type": "Question",
      "name": "What is your return policy?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "You have 30 days from delivery to return any item. Returns are free via prepaid USPS label. Refunds are processed within 5 business days of receiving the return. Items must be in original condition, unworn, with tags attached."
      }
    },
    {
      "@type": "Question",
      "name": "What payment methods do you accept?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "We accept Visa, Mastercard, American Express, PayPal and Apple Pay. Buy now, pay later is available via Klarna for orders over $50. All transactions are secured with 3D Secure protocol."
      }
    }
  ]
}
</script>

Where to place the markup

The JSON-LD goes in the <head> of the page or just before </body>. Both work. What matters is that the Schema content exactly matches the visible content on the page: Google requires it for structured data.

Technical pitfalls to avoid

• Invisible content: NEVER put answers in the Schema that are not visible on the page. Google may penalize you, and an answer readers can't see can't be checked.
• Answers too short: a 5-word Schema answer is useless. Start each answer with one to three self-contained sentences (15 to 70 words).
• Too many questions: limit yourself to 10 questions per page. Beyond that, the signal gets diluted and the page becomes hard to parse.
• HTML in answers: the Answer "text" field supports basic HTML (bold, links), but avoid complex structures. Keep the text readable.

Validation

After implementation, test your page with the Schema Markup Validator (validator.schema.org). It checks that the markup is valid and that questions and answers are correctly parsed. Fix any errors before publishing. Note: Google's Rich Results Test no longer supports the FAQ type since Google dropped FAQ rich results (see the dedicated question below).

→For implementing Organization, Article, HowTo and other schemas: complete Schema.org guide for AI

3 concrete before/after examples

Fashion e-commerce: "What are your shipping times?"

B2B SaaS: "How does your software work?"

Local service: "Why choose our company?"

The pattern is the same in all three cases: replace vague, promotional phrasing with concrete facts, numbers and specific details. AI engines cite information, not slogans.

Implementation checklist in 7 steps

1. Identify 5 to 10 real questions from your audience via Search Console, PAA, customer support and ChatGPT. Eliminate corporate questions nobody asks.
2. Write the answers following the 5 criteria: standalone, short first (15 to 70 words), essential first, factual, no marketing.
3. Create or improve your visible FAQ page. Content must be clearly displayed for humans, not hidden in a JavaScript accordion that crawlers cannot access.
4. Add FAQPage Schema as JSON-LD in the page head. Verify that the Schema text matches the visible content word for word.
5. Test via the Schema Markup Validator (validator.schema.org). Fix any markup errors. Verify all questions are detected.
6. Test your page with Detekia's free GEO score to measure the impact on answer structure (40 points) and evidence & data (15 points). Compare your score before and after. To check whether these answers are actually picked up by ChatGPT, Gemini, Claude or Perplexity, an AI visibility audit of your brand measures your citations query by query.
7. Update quarterly. Your users' questions evolve. Prices, timelines and numbers change too. An outdated FAQ loses credibility with AI engines.

→Understanding the 6 criteria and their weighting in the Detekia score

Frequently asked questions

How many questions should a FAQ have?

Between 5 and 10 questions per page. Below 5, the FAQ covers too few of your visitors' real questions. Above 10, the content gets diluted and the page becomes harder for AI to parse. If you have more than 10 relevant questions, spread them across multiple themed pages.

Can I put the FAQ only in Schema without displaying it on the page?

No. Google has explicitly stated that FAQPage Schema content must be visible on the page. A Schema with invisible content can trigger a manual penalty. And it is the visible questions and answers, not the markup, that get read and reused.

Do FAQs still show as rich results on Google in 2026?

No. Google first restricted FAQ rich results to government and health sites in 2023, then dropped them for all sites on May 7, 2026; the Rich Results Test no longer supports the FAQ type. FAQPage Schema nevertheless remains a valid Schema.org type, which Google says can stay in place without causing problems, and it remains useful for Bing and other tools that read markup, but it is not what gets a page cited: Google says no particular schema is needed for its AI features, and the visible questions and answers are what count. In 2026, the visible FAQ is what matters; FAQPage is a complement.

Dedicated FAQ page or FAQ integrated into each product/service page?

Both. A central FAQ page groups cross-cutting questions (shipping, payment, returns). Specific FAQs on product or service pages answer questions related to that particular offering. Each page gets its own FAQPage Schema. AI engines process each page independently — the more contextualized your FAQs are, the more likely they are to be cited for the right query.

→GEO audit: how to analyze your AI visibility step by step

Key takeaways

The FAQ + FAQPage Schema combo is the most accessible and fastest GEO lever to implement. It directly affects up to 55 out of 100 points on the Detekia score (answer structure + evidence & data). But the key is not technical — it is editorial, and the Schema is only a complement. Real questions, factual standalone answers, and clean markup. The Schema is the container; the quality of your answers is the content.

Start with your 5 most common questions. Rewrite them following the criteria in this article. Add the Schema. Measure the impact. You will see the difference within days.