How to Write FAQ Answers for AI Search That Get Quoted

FAQ answers for AI search get quoted when each answer stands completely on its own: it names its subject, settles one question in a few sentences, and carries something checkable such as a date, a limit or a named feature with a link to where that came from. ChatGPT, Perplexity, Bing Copilot and Google's AI Overviews and AI Mode all assemble replies from passages, so each of them lifts a single answer out of a page rather than the page itself, and anything that leans on the heading above it or the sentence before it breaks the moment it travels. Google switched its own FAQ rich result off on 7 May 2026, which leaves the answer text itself as the whole payoff.

Six checks decide whether an FAQ answer earns its slot and survives being lifted out of the page:

• Standalone. The answer makes sense read cold, with no reference back to the question or to an earlier answer.
• Named subject. The opening sentence repeats the thing being asked about instead of starting with a pronoun.
• One checkable specific. A date, a limit, a price or a named feature, with a link to the document it came from.
• Single question. The answer resolves one question and stops rather than drifting into the next.
• A decision. The answer closes with what the reader should do, not with a restatement of itself.
• Owned question. No other page on the site already declares the answer to the same question.

What makes an FAQ answer quotable by an AI engine?

An FAQ answer is quotable when it is a complete unit: one question, one subject named in full, a few sentences that settle the question, and nothing pointing outside itself. ChatGPT, Perplexity, Bing Copilot and Google's AI Mode assemble replies from passages, so the passage competes for the citation, not the article around it. Which crawler each of those engines uses to decide what it may quote is a separate matter, covered in my post on how AI assistants select the sources they cite.

In the client FAQ blocks I write for agencies, the test is mechanical. Copy a single answer into a blank document and read it cold. If a stranger cannot tell which product, feature or process the answer is about, the answer has failed and the first sentence gets rewritten with its subject restored. Google's own guidance for AI features in Search says "The best practices for SEO remain relevant for AI features in Google Search", which is a useful reminder that this work is editorial rather than technical: see Google's documentation on AI features and your website. If an answer needs its heading in order to make sense, rewrite the opening sentence.

What does a quotable FAQ answer look like, before and after?

A quotable FAQ answer differs from a weak one in three places: the opening words, the specific it carries, and the way it ends. The clearest way to show the standard is one answer written twice, so both versions below answer the same question, "Will our FAQ section still show up as a panel in Google?".

Before. "Unfortunately it's gone now. Google changed how this works a while back, so there's not much point maintaining it any more."

After. "Google's FAQ rich result no longer appears in Search. Google's Search Central documentation changelog, in an entry dated 8 May 2026, states that the feature will no longer appear in Google Search starting 7 May 2026. An FAQ section still earns its place as the shortest route to a reader's question and as the passage an answer engine can quote. If your FAQ block was built to win that panel, rewrite the answers for readers instead of deleting the block."

The rewrite names its subject in the opening words, carries one dated specific with the document it came from, settles a single question and closes with an instruction. What the weak version shows is the usual failure pattern, which is not bad information but an answer written as a reply to somebody standing in the room.

Did Google switching off FAQ rich results change what an FAQ block is for?

Google's FAQ rich result is gone, and the change hits the payoff rather than the practice. The Search Central documentation changelog entry dated 8 May 2026 is headed "Deprecating the FAQ rich result feature" and states "This feature will no longer appear in Google Search starting May 7, 2026", followed by an entry dated 15 June 2026 headed "Removing documentation for the FAQ rich result feature", whose stated reason is that "The FAQ rich result feature is no longer shown in Google Search results".

What I see agencies get wrong here is reading that deprecation as the moment FAQ content stopped paying off, and then deleting the block. On the client sites I look after, that panel had not appeared for years before it was switched off, so May 2026 changed nothing a client could see. The changelog does not say whether FAQPage markup still does anything else, so treat the answer text, not the markup, as the asset. For what structured data still earns on a Webflow site, see my notes on Webflow schema markup. If an FAQ block exists only to win a rich result, repurpose it; if it exists to answer questions, keep it.

What the deprecation changes, in practice:

• Gone. Any expectation that FAQ markup will produce a visible panel under a Google Search result.
• Gone. The argument for adding FAQ markup purely to occupy more vertical space in a classic search result.
• Unchanged. The value of short, single-question answers to a reader deciding whether to contact you.
• Unchanged. The shape an answer engine can lift cleanly, which is one question with one self-contained answer.
• Unknown. Whether FAQPage markup carries weight anywhere else: the changelog entries do not address it.

Keeping the block while dropping any expectation of a visible Search feature is the position I take on the client sites I look after.

How long should FAQ answers for AI search be?

An FAQ answer for AI search works best at roughly 60 to 110 words. A paragraph that size has room to name its subject, settle the question and carry one specific, and it is short enough for ChatGPT or Bing Copilot to quote whole rather than trimming it mid-sentence.

The rule I apply on client projects is one paragraph per answer, written as plain text rather than inside a styled rich-text block, with a hard stop the moment the question is settled. Padding costs more than brevity, because a padded answer buries the one sentence worth quoting. Counted in tokens rather than words, the band I aim for is about 80 to 150 tokens, and that word-to-token conversion is my own arithmetic rather than a published figure from any vendor. If an answer genuinely needs a second paragraph, that second paragraph is a separate question and belongs in its own slot.

• Too short. Under about 40 words, an answer usually drops either its subject or its evidence.
• About right. One paragraph, subject named, one sourced specific, one closing instruction.
• Too long. A second paragraph, a second question, and an engine forced to choose which half to quote.

Why does an answer that opens with a pronoun fail once it is quoted?

An answer that opens with a pronoun loses its meaning the moment it is separated from its question, and separation is the normal case rather than the exception. A reply assembled from passages carries the passage, so an answer beginning "It depends on your plan" arrives as a sentence about nothing in particular.

When I audit a site's FAQ block, pronoun openers are the single most common defect, and the repair takes seconds: swap the pronoun for the full name of the thing. "It costs more on larger sites" becomes "A Webflow Site plan costs more on larger sites". The same rule covers "this", "that" and "as mentioned above", all of which point at text that will not travel with the quote. If the first three words of an answer could sit at the top of any other answer on the site, rewrite them.

Should the same FAQ text appear twice on one page?

FAQ text belongs in one place per page. A rendered FAQ section plus a duplicate FAQ heading inside the article body hands an answer engine two competing copies of the same answer, and which copy gets quoted stops being your decision.

On the projects I lead, each answer is written once and lives in one place, and the article body never repeats it. Each page carries six questions, which covers the real follow-ups without turning a page into a row of thin answers. Where FAQ content should be stored in a Webflow build is a separate architectural decision, covered in my notes on Webflow CMS structure. If an answer exists in two places on one page, delete one of them.

• Keep. One FAQ block per page, in its own fields, with each answer written once.
• Cut. A second FAQ heading in the body that restates the same questions.
• Watch. Answers that quietly duplicate a paragraph already in the article.

Which questions are actually worth putting in an FAQ block?

Questions worth answering are the ones a buyer types and a practitioner can settle from experience. Google describes the query fan-out behind its AI features as issuing "multiple related searches across subtopics and data sources", so a page is weighed sub-question by sub-question rather than as a single block of text, and query fan-out is the reason a discrete question and answer pair is worth more than the same sentences buried in a long section.

For an FAQ block that means every question is a separate entry in the draw, which is why question selection beats formatting. Alphabet's June 2026 investor presentation reports AI Overviews at more than 2.5 billion users a month and AI Mode at more than 1 billion monthly users, so those sub-questions are being asked at a scale worth writing for: see the Alphabet investor presentation and Google's AI features documentation. In the work I do with agencies, the questions that earn citations are narrow operational ones that only somebody who has run the job can answer, never head terms that a vendor's own documentation already settles. Which pages tend to get cited in the first place is a different question, covered in my post on which pages on a service business site get cited by AI, and the wider discipline is laid out in answer engine optimization. If a vendor's documentation already answers a question plainly, skip it and write the one that documentation cannot answer.

Where the questions come from matters more than how they are worded:

• Sales calls. The objection a prospect raises before signing, kept in the words they used.
• Scoping threads. The clarification a client asks for twice, on two different projects.
• Post-handoff messages. The thing a content editor emails about once the site is theirs.
• Your own audits. The defect you keep finding on other people's builds and explaining from scratch each time.

Scored against that test, "How much does a Webflow Site plan cost?" fails as an FAQ entry on an agency site, because Webflow's own pricing page answers it and will be the page quoted. "What breaks when two agencies edit the same Webflow site in the same week?" passes, because no vendor documents it and only somebody who has run that situation can answer it. The second kind is worth a slot even when nothing is searching for it in volume yet.

The pattern holds on my own site. One page, on the white label Webflow workflow, earns the large majority of the AI citations I can see in Bing's AI performance report, while the head-term comparison posts that attract far more classic search traffic earn almost none. Narrow operational questions are the ones answer engines come back for.

Where do FAQ answers for AI search go wrong?

An FAQ answer goes unquoted when it cannot be read apart from the page around it, and the damage always shows up in one of six places: a missing subject, a missing specific, a second question smuggled into the answer, an ending that sells instead of deciding, a dependence on the heading, or a question another page already owns. Checking an answer means one pass per check over its phrasing, and each repair is a rewrite of one or two sentences rather than a restructure of the post. Whether the page itself is a plausible citation candidate in the first place is a different and larger question, settled in my post on which pages on a service business site get cited by AI; the checks below cover only the wording of the answer in front of you.

Check	What a failing answer looks like	What the failure costs	The repair
Standalone	Answer only parses when the question is read first	Lifted out of context, the answer reads as a non sequitur	Restate the question's subject inside the first sentence
Named subject	Opens with "it", "this", "that" or "as mentioned above"	The quote names no entity, so it cannot be attributed to your topic	Replace the pronoun with the full name of the product, feature or process
Checkable specific	Confident prose with no date, limit, price or named feature	Nothing in the answer can be verified, which leaves a sourced page the stronger candidate	Add one specific and link the document it came from
Single question	Answer settles one question then drifts into a second	Half the passage is off-topic for whichever question is being asked	Split the answer and give the second question its own slot
Decision	Closes by restating itself or by pitching a service	A promotional ending reads as marketing copy rather than as an answer	End with a conditional instruction: if X, do Y
Owned question	Another page on the site already declares the answer	Two of your own pages compete to be the quoted source for one question	Link to the page that owns the answer instead of writing a second version

On the audits I run, the answers that fail are rarely badly informed; they are written as the fourth paragraph of something longer. The sequence that fixes them is the sequence of the table: subject, specific, scope, instruction. If an answer fails more than one of the checks, rewrite it from the question rather than patching sentences.

Read the answer out of context

Reading an answer out of context is the fastest of the six checks and catches most of the damage. Paste one answer into an empty document, with no heading and no neighbours, and read it as a stranger would. Anything you have to supply from memory is something an answer engine will not supply either.

Check the question against what you already published

Checking a new question against your published pages is the sixth check and the only one that looks outside the page being written. Search your site for the question you are about to add, and if an existing page already answers it, link to that page instead of writing a second version. The site I run keeps one declared answer per question and links out for everything else, which is why several sentences in this post point at other pages rather than repeating what they already settle.

Who can review an FAQ block before it goes live?

A Webflow developer who writes these blocks for a living can read an FAQ block cold in a few minutes and tell you which answers would not survive being quoted. Bring one page and its FAQ block to a short call and we will go through the answers together, mark the weak ones, and leave you with the rewrite order and the specific each answer is missing. Book a 15 minute slot and bring the live URL; agencies are welcome to bring a client page instead of their own.