How to Optimize FAQs for AI (With FAQ Schema Examples)

Optimizing an FAQ for AI means writing and structuring your questions and answers so that AI search tools pick them up and use them when they answer someone.

That’s a different goal from the one most FAQs were written for. The classic FAQ tried to reduce support emails, or to win a few extra lines in Google’s results. An AI-ready FAQ tries to be the source an answer engine leans on when a person asks a question you know the answer to.

Let’s make it concrete with an example we’ll use through this guide. Picture a hypothetical dental clinic in Bangkok that treats a lot of patients from abroad. One of them opens ChatGPT and asks: “How much does a dental implant cost in Bangkok, and how many trips do I need?”

ChatGPT doesn’t show ten links. It searches, reads a handful of pages, and writes back an answer with a few sources. If the clinic’s FAQ has a question that matches and an answer that states a price range, the number of visits and the healing time in two sentences, that answer is easy to use. If the FAQ says “Prices vary. Please contact us for a quote,” there’s nothing to use, and the clinic isn’t in the answer.

There are four parts to it, and we’ll take them one at a time. The questions have to match how people actually ask. The answers have to be quotable on their own. The page has to be readable by the crawlers that feed AI search. And the markup should describe all of that accurately, without pretending to be more than it is.

FAQs are one application of a bigger idea: answer-first writing. If you want the full method behind it, read our guide on how to write answer-first content, which is the pillar for this whole topic.

Google has switched FAQ rich results off completely. It happened in two steps, three years apart.

Step one: August 2023

On 8 August 2023, Google announced that it was “reducing the visibility of FAQ rich results”. From then on, FAQ rich results from FAQPage structured data would “only be shown for well-known, authoritative government and health websites”, and for all other sites the rich result would “no longer be shown regularly” (Google Search Central Blog). The same post cut HowTo rich results back to desktop, and Google dropped those entirely a few weeks later.

The post also gave advice that still holds: “While you can drop this structured data from your site, there’s no need to proactively remove it. Structured data that’s not being used does not cause problems for Search, but also has no visible effects in Google Search.”

Step two: May 2026

On 8 May 2026, Google added a deprecation notice to its FAQ documentation. The reason it gave: “This feature will no longer appear in Google Search starting May 7, 2026.” On 15 June 2026, Google removed the FAQ rich result documentation altogether, noting that the feature “is no longer shown in Google Search results” (Google Search Central documentation updates).

Search Engine Journal reported the wind-down schedule: the FAQ search appearance filter, the FAQ rich result report in Search Console and FAQ support in the Rich Results Test were set to go in June 2026, with Search Console API support ending in August 2026 (Search Engine Journal). Google gave no reason and published no blog post about it.

So what does that mean for our Bangkok clinic? Nothing changes about how it should write its FAQ. The questions and answers on the page are still read, indexed and used by Google’s systems like any other text. The only thing that went away is the special expandable display in the results page.

And the type itself isn’t going anywhere. FAQPage remains a valid schema.org type, and schema.org’s usage data, based on Google’s index in August 2026, shows it on somewhere between 1 million and 10 million domains (schema.org).

The honest answer is that adding FAQ schema, on its own, doesn’t appear to win AI citations. Two large studies looked at this from different angles, and neither found the boost that many guides promise.

The Ahrefs before-and-after study (May 2026)

Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026 and compared them with matched control pages from other domains that never added it. Every page was already well cited, with more than 100 AI Overview citations in February 2025. The team measured citations for 30 days before and 30 days after each page added schema (Ahrefs).

The result: “Adding schema produced no major uplift in citations on any platform.” In Google AI Mode the treated pages did 2.4% better than the controls and in ChatGPT 2.2% better, but neither difference could be told apart from zero. In AI Overviews, the treated pages did 4.6% worse, a small but statistically significant gap that the authors couldn’t explain and say may not be caused by schema at all.

The study has limits, and Ahrefs lists them. It pooled every schema type (Article, FAQ, Product, HowTo, Organization) rather than reporting FAQPage on its own. It only covered pages that were already heavily cited, so it can’t say whether schema helps an uncited page get noticed. The 30-day window might miss slower effects. And many sites added schema at the same time as other changes. Search Engine Roundtable’s coverage reached the same reading of the numbers (Search Engine Roundtable).

The SE Ranking correlation study (November 2025)

SE Ranking took a different route. It analyzed 216,524 pages across 129,000 domains in 20 niches and compared how often pages with and without certain features were cited by ChatGPT. Pages with FAQ schema averaged 3.6 citations against 4.2 for pages without it. Pages with an FAQ section averaged 3.8 against 4.1 without. Pages that used question-style headings averaged 3.4 against 4.3 for plain headings (Search Engine Journal).

Before you rip out your FAQ, read the researchers’ own explanation. FAQ sections often sit on simpler support pages, which earn fewer citations anyway. Their predictive model actually treated a missing FAQ section as a negative signal. This is a correlation study: it shows what cited pages tend to look like, not what happens when you change one thing.

What the platforms themselves say

Google is clear about its own products. Its guide to generative AI features lists “overfocusing on structured data” as a myth: structured data “isn’t required for generative AI search, and there’s no special schema.org markup you need to add”, though it’s still worth using as part of your wider SEO (Google Search Central).

Microsoft sounds warmer. Its October 2025 guide says schema can label content “as a product, review, FAQ, or event” and turns plain text into data machines can interpret with confidence (Microsoft). In March 2025, attendees at SMX Munich reported that Bing’s Fabrice Canel said schema markup helps Microsoft’s LLMs understand content. That came from a LinkedIn summary of his talk rather than a published Microsoft statement (Search Engine Roundtable).

Put together, the picture is consistent. Schema can help machines understand what’s on a page. It doesn’t make a weak answer worth quoting. Our view, after checking many pages in ChatGPT and AI Mode, is to treat FAQPage markup as cheap, accurate labeling, and to spend most of your effort on the questions and answers themselves. For the wider schema picture, see our guide to which schema types actually help AI search.

The best FAQ questions come from your customers, not from a keyword tool. Keyword tools help, but they show you how people search, not what they’re worried about.

Here’s the order we work in when we build an FAQ for a page.

Your inbox and chat logs. Email, contact forms, WhatsApp and, for businesses in Thailand, LINE Official Account chats. These are gold because they’re in the customer’s exact words, including the awkward ones. Our Bangkok clinic would find the same five questions asked every week: price, number of visits, pain, how long they need to stay in Bangkok, and whether the price includes the crown.

Your sales and front-desk team. Ask them what people ask right before they book, and right before they walk away. Those two moments produce the questions that decide a purchase.

Google’s People Also Ask and Search Console. People Also Ask boxes show related questions for your topic. In Search Console, filter the Performance report for queries that contain words like “how”, “what”, “can”, “does” and “cost” to see the questions your pages already get impressions for.

The AI tools themselves. Ask ChatGPT, Gemini, AI Mode and Perplexity the main question your page answers, then look at the follow-up questions they suggest and the sub-points they cover. Those follow-ups are a decent window into fan-out.

Forums and communities. Reddit threads, Quora and, in Thailand, Pantip show the questions people ask each other when no brand is listening. They’re also where you find the doubts nobody puts in a contact form.

Competitors that already get cited. Run your questions in Perplexity, which shows numbered sources, and look at which pages it cites. What do those pages answer that yours doesn’t?

Once you have a long list, cut hard. A question earns a place only if a real customer would ask it about this page and the answer helps them decide or act. “What is your mission?” fails that test. “Is the implant price per tooth or for the whole treatment?” passes.

For the full method of finding the prompts people type into AI tools, see our guide to prompt research and AI search keywords.

Write each FAQ question the way a real person would ask it, with one clear intent, and with the subject named in full.

Questions do two jobs. For readers, they’re signposts. For retrieval systems, they’re labels that help match a passage to a search. Microsoft makes the point with headings: a vague heading like “Learn More” tells a machine little, while a specific one like “What Makes This Dishwasher Quieter Than Most Models?” defines a clear slice of content (Microsoft). The same logic applies to every FAQ question.

A few rules make the difference.

Use the customer’s words. If patients ask “Does it hurt?”, don’t write “What is the patient comfort level during the procedure?” Keep the plain version. AI tools understand synonyms, and Google says you don’t need to chase every long-tail variant, but plain wording still matches more real questions and reads better.

One intent per question. “How much does an implant cost and how long does it take?” is two questions. Split it. Each answer then has a single job, and each one can be retrieved for its own search.

Name the subject. “How long does it take?” means nothing out of context. “How long does a dental implant take from start to finish?” works on its own.

Add the qualifier people use. Location, audience, budget, situation. “Can I get a dental implant in Bangkok during a two-week trip?” is a real question with real search demand from medical travelers, and a generic page won’t answer it.

Don’t stuff keywords. A question like “Best dental implant Bangkok price cheap affordable?” is written for an old idea of SEO. Readers skip it, and the GEO researchers found that keyword stuffing tended to perform worse than leaving the text alone in their tests.

Ask in the language your customers use. If you serve customers in two languages, as most Thai businesses do, write the Thai FAQ from Thai questions, not as a translation of the English one. Thai customers ask about different things in different words. The same goes for any bilingual market.

Should the question be an H2, an H3 or an accordion summary? Any of these works, as long as each question is clearly marked and sits right above its answer.

FAQs work best close to the topic they answer, so most questions belong on the page they’re about, with a central FAQ page for the questions that apply everywhere.

Think about how retrieval works. When someone asks about implant healing time, the best passage sits on the implant page, surrounded by everything else about implants. That context helps search systems understand the page and helps the reader who clicks through. A generic FAQ page that mixes implants, whitening, parking and refunds is a weaker home for that answer.

So we split FAQs into two kinds.

Page FAQs. Five to ten questions at the end of a service, product or location page, answering what buyers ask about that specific thing. Our clinic’s implant page gets implant questions. Its veneers page gets veneer questions.

Site FAQs. A central FAQ or help center for questions that apply across the business: payment methods, booking and cancellation, opening hours, language support, insurance. Each question there can link to the page with the full answer.

What you should avoid is copying one FAQ block onto every page. The same ten questions repeated on fifty URLs adds no new information, competes with itself, and dilutes the signal about what each page is for. Google’s guide also warns against producing lots of near-duplicate content mainly to match fan-out queries, which it says can fall under its scaled content abuse policy (Google Search Central).

How many questions should each FAQ have? As many as real customers ask about that page, and no more. For most service and product pages that lands at about 5 to 10. There’s no official number from Google, OpenAI or Perplexity; the limit is usefulness. For the clinic’s implant page, a strong set is cost and what’s included, number of visits, total time, pain, who isn’t a candidate, what happens if an implant fails, aftercare, and payment options for overseas patients. Order them by how often they’re asked, with price and timing first.

A valid FAQ schema example is a small JSON-LD block that lists each question and its answer, using the FAQPage, Question and Answer types from schema.org.

Schema.org defines FAQPage as “a WebPage presenting one or more Frequently asked questions”, and Question as a specific question, for example one “collected in a Frequently Asked Questions (FAQ) document” (schema.org, schema.org). Each Question points to its answer through the acceptedAnswer property.

Here’s a complete FAQ schema example for the implant page of our hypothetical clinic. The text in each name and text field is copied exactly from the questions and answers shown on the page.

FAQ schema example: FAQPage JSON-LD

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "How much does a single dental implant cost in Bangkok, including the crown?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "A single dental implant at our clinic costs 45,000 to 65,000 baht including the abutment and crown, depending on the implant brand."
      }
    },
    {
      "@type": "Question",
      "name": "How many visits does a dental implant need?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Most patients need two visits about three to six months apart. The first visit takes about a week, including the consultation and surgery."
      }
    }
  ]
}
</script>

Hypothetical example with illustrative prices. Types and properties from schema.org. Validate before publishing.

@context. Tells parsers the vocabulary is schema.org. Always “https://schema.org”.

@type: FAQPage. Says the page presents frequently asked questions. Use one FAQPage block per page and put every question in it, rather than one block per question.

mainEntity. The list of questions. Each item is a Question.

name. The full question text, exactly as it appears on the page.

acceptedAnswer. An Answer object whose text property holds the full answer, again exactly as shown. When Google still supported FAQ rich results, its documentation allowed basic HTML inside the answer text, such as links, lists and bold. Plain text is the safest choice for any parser.

That’s the minimum. You can add more, such as a url on each Question pointing to its anchor, or an author on the Answer, but those extras aren’t needed to describe a normal FAQ correctly.

If you’d rather not write JSON by hand, our free schema generator builds FAQPage markup in your browser: pick FAQPage, paste your questions and answers, and copy the result.

Use FAQPage for FAQs your business writes, and QAPage only for pages where users answer a single question.

People mix these up all the time, because both involve questions and answers. Google’s QAPage documentation draws the line clearly. QAPage is for a page with “one question followed by its answers”, and “users must be able to submit answers to the question”. It lists “an FAQ page written by the site itself with no way for users to submit alternative answers” as an invalid use, along with a blog post or how-to guide that answers a question (Google Search Central).

Unlike FAQ rich results, Google still documents the Q&A rich result for QAPage, so forum-style sites have a live feature to aim for. For a normal business site, though, QAPage almost never fits.

FAQPage also isn’t the only markup an FAQ page might carry. The page can still have Organization, BreadcrumbList or, for a local business, LocalBusiness markup alongside it. Our guide to schema markup for AEO covers how those types fit together.

Most FAQ schema problems come from one root cause: the markup says something the page doesn’t.

Google’s general structured data guidelines apply to every type. “Don’t mark up content that is not visible to readers of the page.” Don’t mark up irrelevant or misleading content. Don’t use structured data to deceive. Breaking these rules can lead to a structured data manual action, which removes rich result eligibility but, Google says, doesn’t affect how the page ranks (Google Search Central).

Here are the mistakes we see most often when we audit sites.

Markup that drifts from the copy. The FAQ was rewritten, the schema wasn’t. Now the markup says one price and the page says another. That’s confusing for any machine reading both, and it’s exactly the kind of inconsistency that makes an AI tool less sure which figure to trust.

Questions that exist only in the markup. Some plugins and some agencies add extra questions to the JSON-LD that never appear on the page, hoping for more coverage. That breaks Google’s visible-content rule.

Advertising dressed as an FAQ. “Why is our clinic the best in Bangkok?” with three paragraphs of praise. It’s not a question anyone asked, and the answer has nothing in it to quote.

The same FAQ markup on every page. Usually a theme or footer widget that injects the same block site-wide. It tells machines nothing about each page.

Broken JSON. A missing comma, a smart quote, an unescaped double quote inside an answer. One error can make the whole block unreadable. Validate after every change.

Treating schema as the strategy. Adding FAQPage to thin answers and waiting for citations. As the Ahrefs data shows, the markup doesn’t carry the page.

An FAQ can only be cited if the crawlers behind AI search can fetch it, so check access before you rewrite a single answer.

Allow the right crawlers in robots.txt. For Google’s AI features, that’s Googlebot, and the page must be indexable with a snippet allowed. For ChatGPT search, it’s OAI-SearchBot. OpenAI treats it separately from GPTBot, its training crawler, so you can allow search while blocking training (OpenAI). For Perplexity, it’s PerplexityBot, which Perplexity says is used to surface and link websites in its results, not to crawl content for AI foundation models (Perplexity). For Copilot, it’s Bingbot.

Check the firewall and CDN. Some security and CDN settings block AI crawlers by default. robots.txt can say “allow” while the CDN still returns an error. Check your server logs for requests from these user agents and look for 200 responses.

Keep the page indexable. No noindex, no nosnippet on the FAQ section, and a canonical that points to the page itself.

Keep the text in HTML. As covered above, answers that only appear after JavaScript runs are a risk.

Our free AI Crawler Checker reads your robots.txt and shows which AI crawlers are allowed or blocked, which is a two-minute first check.

Keep FAQ schema if it’s accurate, fix it if it isn’t, and add it to new FAQs if your setup makes it easy. Don’t make it a priority project.

Here’s our reasoning, based on the sources in this guide.

Keeping it costs nothing. Google said in 2023 that unused structured data “does not cause problems for Search”, and the 2026 deprecation didn’t change that. FAQPage is still valid schema.org, and other systems that read schema, including Microsoft’s by its own account, may use it to understand the page.

Removing it gains nothing, unless it’s wrong. Markup that no longer matches the page is the one case where we’d remove or rewrite it straight away.

Adding it is a small, low-risk job, not a growth lever. Ahrefs found no clear citation uplift when well-cited pages added schema, and Google says no special markup is needed for its AI features. If your CMS or SEO plugin adds FAQPage automatically from an FAQ block, let it. If adding it means a developer ticket, put that ticket behind rewriting your answers.

Stop expecting rich results. If your reporting still has a line for FAQ rich result impressions, retire it. That feature is gone for everyone.

For our Bangkok clinic, that means a simple plan: keep the FAQ block its WordPress plugin already marks up, make sure the markup updates when prices change, and spend the real effort on rewriting the eight implant answers so each one leads with a fact.

Use this checklist before you publish or update any FAQ. It condenses everything in this guide into the checks we actually run.

Questions. Every question comes from a real customer source, has one intent, names its subject, and uses the customer’s wording in the right language. No filler like “Why choose us?”.

Answers. The first sentence answers the question completely, restates the subject and includes a specific, current fact. Sensitive answers cite a source or a named reviewer.

Page, markup and access. Questions sit on the page they’re about, the answer text is in the HTML source even when collapsed, FAQPage JSON-LD copies the visible text word for word and passes the Schema Markup Validator, and Googlebot, Bingbot, OAI-SearchBot and PerplexityBot can reach the page. Then the questions go into your monthly AI visibility prompt set.

If you only have time for one thing this week, pick your most important page and rewrite its FAQ answers so each one opens with a fact. That single change does more than any markup.

And if you want a second pair of eyes, we’ll check how AI tools answer your customers’ questions today and show you which FAQ answers to fix first. Our AEO service covers the full process, from questions to measurement.

1. Google Search Central: Documentation updates (FAQ rich result deprecated 8 May 2026, docs removed 15 June 2026)
2. Google Search Central Blog: Changes to HowTo and FAQ rich results (8 August 2023)
3. Search Engine Journal: Google drops FAQ rich results (Matt G. Southern, 10 May 2026)
4. Google Search Central: Optimizing your website for generative AI features on Google Search (updated 10 July 2026)
5. Google Search Central: AI features and your website (updated 10 December 2025)
6. Google Search Central: General structured data guidelines
7. Google Search Central: Q&A (QAPage) structured data
8. Schema.org: FAQPage
9. Schema.org: Question
10. Ahrefs: We tracked 1,885 pages adding schema. AI citations barely moved (Linehan and Guan, 11 May 2026)
11. Search Engine Roundtable: Study says adding schema did not improve AI citations (13 May 2026)
12. Search Engine Journal: New data on the top factors influencing ChatGPT citations (SE Ranking study, 26 November 2025)
13. Microsoft Advertising Blog: Optimizing your content for inclusion in AI search answers (Krishna Madhavan, October 2025)
14. Search Engine Roundtable: Microsoft confirms schema helps its LLMs understand content (20 March 2025)
15. Aggarwal et al.: GEO: Generative Engine Optimization (KDD 2024)
16. OpenAI: Overview of OpenAI crawlers
17. Perplexity: Perplexity crawlers
18. Schema.org: Schema Markup Validator
19. Google Search Central on YouTube: Structured data for web developers

Keep reading

• Content & technicalHow to Write Answer-First Content for AI Search
• Content & technicalSchema Markup for AEO: What Actually Helps AI Search
• Content & technicalPrompt Research: How to Find AI Search Keywords
• AI platformsHow to Rank in Google AI Overviews