Do FAQ Pages Still Work for AI Search?

By the alphaa team — we run AI-visibility scans across thousands of businesses and read a great many FAQ pages. Last updated 28 August 2026.

Short answer: Yes — but for a completely different reason than they used to. FAQ pages no longer earn you anything in Google's search results; Google removed FAQ rich results for almost all sites in 2023 and the visual reward is gone. What they do earn is extraction. A question-shaped heading followed by a short, self-contained answer is structurally identical to what a retrieval system is trying to build, so it can be lifted into an AI answer with minimal rewriting. The catch is that this only holds for real questions with real answers. The marketing-fluff FAQ — "Why choose us?" "Are you the best in the business?" — is worse than nothing, because it teaches an engine that your page contains no facts.

What actually changed, and when

Two separate things get conflated, so it is worth being precise.

In August 2023 Google announced it was reducing FAQ rich results to a small number of authoritative government and health sites, and rolling them back everywhere else. Google's own FAQPage structured data documentation still describes the markup, but the expanding-accordion snippet most sites were chasing is no longer available to them. That is a change to a search-result display feature.

It is not a statement about whether question-and-answer content is useful. The content format and the rich result were always separate things, and only one of them went away. AI assistants are not reading Google's SERP features — they run their own retrieval, fetch pages, and extract passages. What a deprecated snippet type does or does not render has no bearing on whether a passage on your page is a good candidate for that extraction.

Why the Q&A format suits retrieval so well

When an assistant answers a question, it needs a passage that (a) is topically matched to the query, (b) makes sense removed from its surroundings, and (c) is short enough to quote without summarising. A well-formed FAQ entry satisfies all three by construction:

• The heading is already the query. Retrieval matches semantically, and a heading phrased as the literal question a person would ask is the closest possible match to that person's query.
• The answer is self-contained. A good FAQ answer does not depend on the paragraph above it. That is exactly the property that makes a passage safe to quote — the extraction does not lose meaning.
• Boundaries are unambiguous. Heading, answer, next heading. A chunker splitting your page into passages will get the split right, which is much less certain inside a long flowing narrative.

This is the same mechanism behind writing content AI engines will quote — the FAQ is simply the most mechanical way to produce it.

The FAQ that fails, and why it fails

Roughly the pattern we see on most small-business sites:

• "Why should I choose [company]?"
• "Are you licensed and insured?" — answered "Yes, absolutely!"
• "Do you offer free estimates?" — answered "Yes! Contact us today."
• "What areas do you serve?" — answered "We proudly serve the greater metro area."

Four questions, zero extractable facts. Nobody has ever asked an assistant "why should I choose Acme," so question one can never be matched. Questions two through four match real queries and then supply an answer that cannot be quoted usefully — "yes, absolutely" is not information, and "the greater metro area" is not a place. An engine that fetches this page comes away with nothing, and pages that repeatedly yield nothing stop being selected.

There is a second failure mode worth naming: the FAQ as a dumping ground. Twenty-five questions on one page, each answered in a sentence, covering everything from payment methods to company history. This dilutes the page's topical focus so much that it competes with itself — nothing on it is about anything in particular, so it is a weak match for every query.

How to write an FAQ entry that gets quoted

The rules are simple and almost nobody follows them.

1. Use the question people actually ask, verbatim. Not "Pricing information" but "How much does a 200 amp panel upgrade cost?" Source these from your own inbox, your phone notes, and your sales calls — the real phrasing is always messier and more specific than the phrasing you would invent.
2. Answer in the first sentence. Lead with the answer, then qualify. "Most panel upgrades run $2,200–$4,500. The range widens when the mast and meter base need replacing…" If your first sentence is throat-clearing, the quotable part is buried where extraction will not reach it.
3. Keep each answer 40–120 words. Long enough to contain a real fact and its caveat, short enough to lift whole. Below that you have a slogan; above it you have an article that needs summarising.
4. Make it standalone. No "as mentioned above," no "this," no "we." Write each answer as if it will be read with no surrounding context, because it will be.
5. Include a number, a name or a condition in every answer. If an answer contains no specific, delete the question — it is not a question, it is an advert.
6. Answer the awkward ones. The questions you would rather not answer in writing — cost, how long it really takes, what you do not do, what goes wrong — are the highest-value entries, because they are the ones people ask assistants precisely because nobody publishes them.

Where the FAQ should live

This is where most sites go wrong structurally. A single sitewide /faq page collecting every question you have ever been asked is the weakest arrangement. Better:

• Put four to eight questions at the bottom of each service page, specific to that service. This concentrates topical signal — the page is about panel upgrades, and its FAQ is about panel upgrades — instead of diluting it.
• Give genuinely big questions their own page. If "how much does X cost" needs 800 words, it is an article, not an FAQ entry. Write the article, and keep a short summary answer in the FAQ that links to it.
• Keep a sitewide FAQ only for genuinely cross-cutting things — payment, insurance, scheduling, service area — and keep it short.

One practical note: if your FAQ answers are hidden inside an accordion that only injects text into the DOM after a click, verify that the answers are present in the raw HTML. Most accordion components ship the text and merely hide it with CSS, which is fine. Some fetch it on demand, which means several crawlers see the questions and none of the answers. Check with curl -s https://yoursite.com/your-page | grep -c "a phrase from an answer", and see JavaScript rendering and AI crawlers if it comes back zero.

Should you still add FAQPage schema?

It is cheap and low-risk, so yes — with realistic expectations. It will not produce a rich result for a typical business site any more. What it does is make the question-and-answer relationship explicit and machine-readable rather than inferred from your HTML structure, which is a small, real help to any system parsing your page. Follow Google's documented rules regardless: the marked-up content must be visible on the page, and the format is for genuine FAQ content, not for promotional copy dressed as questions.

Do not mark up an FAQ that does not exist visibly, and do not stuff keywords into question headings. Both are content-quality problems that outlast any single search feature. Our schema markup guide covers which types earn their keep.

How to tell whether yours is working

You cannot measure this in Search Console, because most of the value shows up in answers that never generate a click. What you can do:

1. Ask the question yourself. Take an FAQ entry, paste the question into ChatGPT, Perplexity, Gemini and Claude with search enabled, and read the citation list. If your page is cited, the entry is doing its job. If a competitor or a directory is cited, read their answer and ask what it contains that yours does not — it is almost always a number.
2. Repeat it several times. A single run is not a measurement; answers vary between identical prompts for reasons unrelated to you, as we explain in why AI answers change every time.
3. Watch referral traffic from assistant domains in your analytics. It will be small, but the trend and the landing pages tell you which answers are being surfaced — tracking AI traffic has the setup.

Questions we get asked

Didn't Google say FAQ pages are spam?

No. Google reduced the FAQ rich result to a narrow set of sites, which is a display decision. It has never said question-and-answer content is bad, and its guidance continues to describe the markup for sites that qualify. What Google does discourage is FAQ content written purely to trigger a search feature — which is a fair complaint, and describes most of the FAQs written between 2019 and 2023.

How many questions should a page have?

Four to eight per service page. Enough to cover the real decision points, few enough that each one is substantive. If you find yourself padding to reach a number, stop — the padding questions are the ones that signal an empty page.

Can I generate FAQs with AI?

You can draft the structure with AI, but the specifics have to come from you, because the whole value of the format is the facts — your prices, your process, your constraints — and a model does not know those. A generated FAQ full of plausible generalities is precisely the low-information page that fails. See does AI-generated content hurt your AI visibility for where the line actually sits.

Will an FAQ get me recommended by ChatGPT?

On its own, no. An FAQ makes your existing facts extractable; it does not create authority, fix a site an engine cannot fetch, or override the third-party sources an engine trusts for your category. It is one of the highest-return formatting changes available, which is not the same as a strategy. And nobody can guarantee placement in an AI answer — AEO shapes the public signals engines read, it does not control their output.

The bottom line

The FAQ page lost its search-result reward and quietly gained a better one. For AI search, question-shaped headings with short, standalone, fact-carrying answers are the most directly extractable content you can write — a pre-chunked, pre-matched passage handed to a retrieval system that was going to have to build one anyway. The format is not the point, though. The facts are. An FAQ full of specifics is the cheapest AEO win on most sites; an FAQ full of "yes, absolutely!" is a page that teaches engines to skip you.

Run a free AI visibility scan →