An FAQ gets quoted by AI search when each answer is a self-contained passage of roughly 40 to 60 words that opens with the answer itself and makes sense with the rest of the page removed. That is the whole rule. The question is worded the way a customer would type it, the first sentence answers it outright, and the sentences after it carry the condition, the number or the caveat. Schema helps an engine find the block. It does not make a weak answer worth quoting. This post covers the format, the evidence behind it, and how to check whether yours is working.
We ran the audit on ourselves before writing any of this. On 14 August 2026 we parsed every FAQ block on this blog, 193 answers across 71 published posts. The median answer runs 54 words, 122 of the 193 run past 50 words, and only 33 sit inside the 25 to 45 word band that extracts cleanly. By our own rule, our own FAQ blocks are too long. The rewrite queue starts this week and the same figures get re-run in the September citation report.
What does an AI engine actually take from your page?
A passage, not a page. ChatGPT, Perplexity and Google AI Overviews answer by breaking candidate pages into chunks, ranking those chunks against the question, and handing the strongest ones to the model with the citation already attached. The unit of competition is the paragraph. A page can be genuinely good and still lose, because no single block inside it answers the question without the paragraphs around it.
This is why FAQ blocks perform above their weight. A question heading followed by a short answer is already shaped like a retrieval chunk. Almost nothing else on a normal service page is. We first wrote about this in May 2026 in our guide to structuring FAQ pages so AI engines quote you, and this post revises the schema part of that advice. A hero line, a benefits grid and a founder quote are all fragments that need the page to mean anything, which is exactly what an extracted passage does not have.
Does FAQ schema still do anything in 2026?
Less than most agencies claim, and it is worth saying so plainly. Ahrefs tracked 1,885 pages that added JSON LD schema between August 2025 and March 2026 against a matched control group of roughly 4,000 pages and found no meaningful citation uplift on any platform, with Google AI Overviews showing a small decline rather than a gain. Source: Ahrefs schema markup and AI citations study, 2026. Google had already narrowed FAQ rich results in August 2023 to well known government and health sites, so the visible search reward disappeared three years ago.
We still put FAQPage schema on every post. Not because it buys citations, but because it makes the boundary between a question and its answer unambiguous to any parser, and it costs a few minutes. Treat it as labelling, not as a growth tactic. Anyone selling schema as the reason you will get cited is selling the wrapper and skipping what goes inside it.
How do you write an answer that survives extraction?
Write each answer as though it will be read alone, because it will be. Five constraints do most of the work.
- Use the client's wording for the question. Take it from your inbox or your WhatsApp, not from a keyword tool. Real questions carry the qualifiers that make an answer specific.
- Answer in the first sentence. No throat clearing, no restating the question, no promise that it depends.
- Hold the answer to 40 to 60 words. Long enough to qualify the claim, short enough to be lifted whole.
- Name the specifics inside the answer. The city, the service, the timeframe, the number. An answer that says our clients typically see results is unusable to a machine and to a buyer.
- Never refer to the rest of the page. As mentioned above and see the section below both break the moment the passage travels.
Here is the same question answered badly and then well. Before: there are many factors that affect how long SEO takes, and as we discussed above it really depends on your industry, your competition and your budget, so it is difficult to give a single answer, but most businesses do see movement eventually. That is 48 words that commit to nothing.
After: for a clinic in Sainikpuri with a complete Google Business Profile and a working website, movement on local searches usually shows in six to twelve weeks. City wide terms take closer to six months. The timeline stretches when the profile is incomplete or the site has no page naming the locality. That is 52 words, and every clause is checkable.
How do you check whether your FAQ is being quoted?
Ask the engines directly, on a fixed schedule, and write down what you see. Pick ten questions clients actually send you, run them through ChatGPT, Perplexity and Google AI Mode on the same day each week, and log three things: whether your business appears, which source the engine cited instead, and whether any sentence of yours came back close to verbatim. That last one tells you the extraction worked.
We have run a fixed query set weekly since June 2026 and published the results, including the nine consecutive weeks where our category queries returned nothing. It is uncomfortable and it is the only honest way to know. ZeroAdo, the closest competitor to us on this category, holds three separate pages on the AI optimisation query, which is a useful reminder that this is a portfolio game rather than one perfect page.
Key takeaway: AI engines quote passages, not pages. An FAQ answer that opens with the answer, runs 40 to 60 words, names the specifics and stands alone is the cheapest extractable block you can add to any page. Schema labels it. It does not rescue it.
Writing for extraction is one chapter of the framework we run for every client, the Marketing Operating System. Our pillar guide to AI search optimisation for Hyderabad businesses covers the rest of the stack, and our SEO and AIO services page shows what running it with us looks like. If you would rather have a team handle this, Vridhii Digital builds growth systems for founders who care about results. Message us on WhatsApp and send us the ten questions your clients keep asking.
Frequently asked questions
How many questions should our FAQ have?
Three to five per page is enough. Each one should be a question a client has actually sent you, and each should target a different intent. Twenty shallow questions on one page compete with each other for the same retrieval slot and dilute every answer. Depth per answer beats volume of answers.
Should the FAQ go on every page or on one FAQ page?
Put the questions on the page that answers them. A single FAQ page collects questions about unrelated topics, so no page becomes the strong match for any query. Three questions at the bottom of your clinic page will outperform thirty on a general FAQ page, because relevance is judged page by page.
Our answers are already written and they are long. Do we rewrite all of them?
Not all at once. Start with the pages that already get traffic or enquiries, since those have retrieval history to build on. Tighten the first sentence of each answer so it carries the full answer alone, then cut the rest to the essential qualifier. We are doing exactly this to our own 193 answers.
Will adding an FAQ get us into ChatGPT answers?
It improves the odds, it does not guarantee anything. Extraction also depends on whether the engine trusts your site, whether other sources corroborate you, and how contested the query is. The FAQ is the cheapest of those factors to fix, which is why it is the sensible first move rather than the only one.