FAQ Schema for AI Search: Still Worth It After Google Pulled Rich Results

In August 2023 Google narrowed FAQ rich results to a small set of authoritative government and health sites. A lot of teams read that as "FAQ schema is dead" and ripped it out.

That was the wrong conclusion. The visual reward disappeared; the machine-readable value did not.

Key takeaways

• Google restricted FAQ rich results in 2023 — most sites no longer get the expandable snippet.
• The markup is still parsed as structured question-and-answer pairs, which is exactly the shape an answer engine wants.
• It is the single most direct way to hand an engine a pre-formatted answer.
• Only mark up content that is actually visible. Marking up hidden content is a violation, not a shortcut.

Why does FAQ schema still matter?

Because it removes the hardest step. Without markup, an engine has to infer which part of your page answers a question — parsing headings, guessing paragraph boundaries, deciding where an answer starts and stops. FAQPage schema hands it an explicit pair: this is the question, this is the answer.

Everything else in AEO is about making extraction easier. This is the one mechanism that makes extraction unnecessary.

What does correct FAQ schema look like?

\\\`json

{

"@context": "https://schema.org",

"@type": "FAQPage",

"mainEntity": [

{

"@type": "Question",

"name": "How long should a landing page be?",

"acceptedAnswer": {

"@type": "Answer",

"text": "Long enough to answer every objection a buyer has, and no longer. For a simple offer that is often one screen. For a considered purchase it may be several thousand words. Length is an output of the decision being made, not a target to hit."

}

}

]

}

\\\`

Three rules that catch most people:

1. The question and answer must appear on the page, visibly, in substantially the same words.
2. `text` accepts limited HTML — basic formatting is allowed, but keep it simple.
3. One FAQPage per page. Multiple blocks compete and may all be ignored.

What breaks it?

The failure modes are boring and common:

The trailing comma one deserves emphasis. A single syntax error voids the whole block, and nothing on the page looks broken — you simply get no benefit and no warning. Validate before you ship.

How do you choose the questions?

Use the questions your buyers actually ask, in their words. Sources worth mining, in rough order of quality:

1. Sales and support conversations. The highest-signal source you have, and the one most teams ignore.
2. Search Console queries where you get impressions but few clicks — the question was asked, your answer did not satisfy.
3. "People also ask" boxes for your core terms.
4. Competitor FAQ sections — not to copy, but to find the objections you have not addressed.

Do not invent questions to fill a quota. A FAQ with three real questions beats one with twelve invented ones, and the invented ones dilute the page for humans.

How do you write the answers?

Write 40–60 words, self-contained, answering the question directly in the first sentence. Do not build up to it. Do not reference other parts of the page. Assume this paragraph will be read entirely alone, because when it is quoted, it will be.

Weak:

> This is a great question and one we get a lot! There are several factors to consider, and the answer really depends on your specific situation and goals, which we will break down in the sections below.

Strong:

> A landing page should load its main content in under 2.5 seconds on mobile. Past that, Google classifies the experience as needing improvement and visitors begin abandoning before the page renders. Test on a throttled connection, not on office wifi, or you will measure the wrong thing.

Implementation steps

1. Pick three to eight real questions from sales, support or Search Console.
2. Write each answer at 40–60 words, self-contained.
3. Publish them visibly on the page — a genuine FAQ section, not hidden markup.
4. Generate the FAQPage JSON-LD from that same text.
5. Validate the JSON. A syntax error voids the entire block.
6. Confirm there is exactly one FAQPage block on the page.
7. Re-check after any content edit, so markup and visible text do not drift apart.

Step seven is where most implementations rot. Someone edits the visible copy six months later, the markup keeps the old wording, and the two quietly disagree.

What else should be on the page?

FAQ schema works best alongside the markup that says who you are. Organization schema with name, url, logo and a sameAs array tells an engine which company is answering. Without it you have a good answer with no resolvable author.

If you want to see which of these you are missing, our free AI citation check reads your live HTML and reports exactly which schema types are present, whether they parse, and what is absent — plus the answer-structure checks that decide whether the markup has anything worth pointing at.

Related reading: How to get cited by ChatGPT and AEO vs SEO.