FAQ sections for AI

For years the standard advice was to add an FAQ section because it could win the expandable question-and-answer box under a Google listing. That box is gone. The question-and-answer shape, though, still earns its place, for a reason that has nothing to do with rich results: it is close to the shape in which AI answer engines retrieve content. A system answering a question looks for a passage that matches the question, and a question followed immediately by its answer is exactly that kind of passage. The rich result was never the real value. The clean, question-matched passage was, and still is.

The FAQ rich result is gone

Start with the part of the old advice that is dead. Google has withdrawn its FAQ structured data documentation, and the Search Central changelog records the change: "The FAQ rich result feature is no longer shown in Google Search results, as announced in the changelog entry in May 2026" (Google Search Central changelog). That page had put the date on it: "As of May 7, 2026, FAQ rich results are no longer appearing in Google Search." The expandable Q&A dropdown under a search listing, the thing FAQ schema was supposed to buy you, no longer exists as a search feature.

It had been narrowing for a long time before that. The same page, while it still stood, limited the feature to a small class of sites: "FAQ rich results are only available for well-known, authoritative websites that are government-focused or health-focused." For most sites, the dropdown was already out of reach. Adding FAQPage markup in 2025 to win a rich result was chasing a feature that had already closed its doors to you.

None of this makes the markup harmful. FAQPage remains valid schema.org vocabulary, and if your pages already carry it, there is no reason to rip it out. But it also means the rich result cannot be the reason to build a Q&A section. If a question-and-answer format earns its place on a page today, it earns it somewhere else entirely: in how retrieval systems select passages.

One question-shaped feature outlived it. People Also Ask, the expandable related questions Google shows on most result pages, is still there, and Google has never documented FAQ markup as a route into it. Those answers are drawn from page content, which is the same lesson the rest of this article arrives at from a different direction.

How AI answer engines use a Q&A section

Modern answer engines retrieve before they generate. When someone asks ChatGPT with search enabled, Perplexity, Claude with web access, or Google's AI Overviews and AI Mode a question, the system first pulls candidate pages or passages from an index, then writes an answer grounded in what it pulled. The retrieval step is where a Q&A section matters.

The foundational technique is dense passage retrieval, described in a 2020 paper by Karpukhin and colleagues at Facebook AI. The paper showed that "retrieval can be practically implemented using dense representations alone, where embeddings are learned from a small number of questions and passages by a simple dual-encoder framework" (Karpukhin et al., 2020). In practice: one encoder turns the user's question into a vector, another turns each candidate passage into a vector in the same space, and the passages whose vectors sit closest to the question's vector win. The system is literally matching a question against passages.

Follow that to its consequence. A passage that opens by stating a question and then immediately answers it is about as close a match to a user asking that question as prose can get. Not because engines have a rule that rewards FAQ formatting, but because the passage genuinely is the thing being searched for: a direct answer to that question, labeled with the question itself. That is the entire case for the Q&A shape.

Google confirms that retrieval comes first in its own AI features, though it does not say what kind of retrieval it runs. Its guide to AI features in Search defines the pipeline: "Retrieval-augmented generation (RAG): A technique (also known as grounding) used to improve the quality, accuracy, and freshness of AI responses by relying on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index" (Google AI features guide). AI Overviews and AI Mode sit on top of ordinary retrieval, not beside it.

Passage-level selection is older than the generative features, and Google has documented it for years in the form of featured snippets. Its featured snippets page notes that "Clicking a featured snippet takes the user directly to the section of the page that appeared in the featured snippet." And on whether a site can nominate its own passage: "You can't. Google systems determine whether a page would make a good featured snippet for a user's search request, and if so, elevates it" (featured snippets documentation). Google still decides, and what it shows the reader is a section rather than the whole page. Writing sections that survive being lifted is the part you control.

On the pipeline side, the same shape appears in how retrieval systems are built. The LangChain documentation for its header-based splitter opens with: "Many chat or Q&A applications involve chunking input documents prior to embedding and vector storage" (LangChain splitter docs). Splitters commonly cut documents along headings. A heading that states a question, with the answer directly beneath it, becomes a chunk that carries its own label and its own payload. A paragraph that starts "As mentioned above, this also applies here" becomes a chunk that lost its referent the moment it was cut.

But this reasoning has a clear limit. Nothing above is a measured citation lift. No vendor publishes data showing that FAQ pages get cited more often, and this article makes no such claim. Google documents the retrieval mechanism for its own features. OpenAI documents nothing about FAQ or answer formatting for ChatGPT. Anthropic documents nothing of the kind for Claude. Perplexity documents nothing either. So a sentence like "Q&A sections get you cited in ChatGPT" is an inference from how retrieval systems are commonly built, not a documented behavior of any named engine, and it should always be labeled as one. What you can say with confidence is narrower: a question-plus-answer passage is a clean, question-matched, self-contained unit of the kind these systems retrieve. Whether it gets selected over competing passages depends on ranking, competition, and quality signals no vendor has published.

The content is the lever; the markup is a label

A lot of surviving FAQ advice treats FAQPage schema as non-negotiable for AI visibility. Google's own guidance says the opposite: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." That sentence comes from the same AI features guide quoted above, and it is unambiguous. The retrieval systems read the page; the markup is metadata about the page.

The old FAQ rich result guidelines made the same point from the other direction. Among the content requirements was this: "All FAQ content must be visible to the user on the source page." Even when the rich result existed, the markup was only ever a label on visible question-and-answer content. Markup describing answers that were not on the page was a violation, not a shortcut.

So the priority order is settled: write the visible question-and-answer content first, and treat FAQPage JSON-LD as an optional annotation you may add afterwards. The mechanics of writing that markup, the JSON-LD structure, validation, and its remaining uses, are a separate topic and not covered here.

The per-engine answer rules you can ignore

A specific genre of GEO advice circulates around FAQ content: precise answer-length rules, stated per engine. A maximum word count for one engine, an ideal sentence count for another, a percentage lift attributed to a particular format. These figures share two properties. No vendor documents any of them; OpenAI, Anthropic, Perplexity, and Google publish no answer-length specification of any kind. And the guides contradict each other; two sources will state different figures for the same engine, so at least one of them is wrong, and neither traces to the vendor.

Discard the numbers and the underlying pattern remains, supported by the retrieval mechanism rather than by any figure. Put the answer first, because a passage can be lifted or cut and the opening is what travels. Make the answer complete, because an incomplete answer is not a match for the question above it. Length falls out of answering the question fully and stopping. If your honest answer takes one sentence, one sentence is the right length. If it takes a paragraph, trimming it to hit somebody's circulated word count only makes it a worse answer.

Writing Q&A sections that get retrieved

A retrieval system matches a question against passages, and passages are often cut along headings. The habits follow directly.

Start with the questions readers actually ask, phrased in their words, because the question is the match surface. "How much does shipping cost?" matches how people ask; "Shipping cost considerations" matches nobody's query, and support tickets, sales calls, and search suggestions are better sources than brainstorming. Open each answer with the answer itself and add the context, caveats, and detail afterwards, so that if the passage is lifted or quoted alone the point still survives. Keep every pair self-contained: no "as mentioned above," no pronoun reaching back to a previous answer, because a retrieval system encounters the pair with none of its surroundings. And hold each pair to a single question. If an answer drifts into a second question, give that question its own heading. Otherwise, the chunk is labeled with the wrong question and is harder to match.

Do not pad. Preamble in front of the answer pushes the substance away from the question it is meant to match, and filler only dilutes the passage. Let a clear heading carry the question, so splitters and anchor links get a clean boundary and the chunk arrives with its own name. The same logic scales to the whole page: a two-or-three-sentence statement of what the page establishes, placed before the first heading, gives the page an answer-first opening for the query it answers as a whole.

One warning against overdoing it. The chunk-friendly logic above does not mean shredding your content into one-line stubs. Google addresses this directly: "There's no requirement to break your content into tiny pieces for AI to better understand it. Google systems are able to understand the nuance of multiple topics on a page and show the relevant piece to users." A Q&A section earns its place when the material genuinely is a set of discrete questions. Forcing flowing explanations into fragments makes the content worse for readers without making it more retrievable.

And the honest limit on formatting altogether, from the same guide: "Creating content that people find unique, compelling, and useful will likely influence your website's presence in generative AI search in the long run more than any of the other suggestions in this guide." The Q&A shape makes a good answer easier to lift whole. It will not rescue a thin one. If the answer itself is something a system can generate without you, no format decides in your favor.

FAQ

Do FAQ pages get cited more by AI?

Nobody has published data showing that. The grounded claim is narrower: a question-plus-answer passage is a clean retrieval match that an engine can lift whole, which is a real structural advantage. "More citations" is a promise no vendor has made, and no measurement supports it.

Is schema (FAQPage) still worth adding?

It is optional. Google states that structured data is not required for generative AI search, and FAQ rich results no longer appear. The markup is a valid, harmless label; the value lives in the visible question-and-answer content it describes.

Did Google remove FAQ rich results?

Yes. As of May 7, 2026 they no longer appear in Google Search, and before removal they were limited to well-known government and health sites. Any advice that says "add FAQ schema to get the dropdown" is obsolete.

How long should an FAQ answer be for AI?

There is no documented length, and the third-party numbers in circulation disagree with each other. Answer the question completely, with the direct answer in the first sentence so the passage stands alone, and stop. Do not pad or trim to hit a figure.

Do I need a labeled "FAQ" section, or just Q&A content?

The shape matters, not the label. Any section that poses a real question as its heading and answers it directly underneath is the same shape, with or without the word "FAQ" above it. No engine documents treating a labeled FAQ block differently.