FAQPage and structured Q&A for AEO

If you read nothing else: a well-built Q&A block is the single highest-yield pattern in Answer Engine Optimization, and FAQPage JSON-LD is still worth shipping even though the Google dropdown it used to earn is gone.

I caught myself doing the bad version the other day. I was adding a FAQ to a page, typed out a fifth question, and stopped - because no human being has ever asked it; I had written it to hold a keyword. That is the trap this post is about. A Q&A block gives an AI engine exactly what it wants - a real question, answered in the first sentence, short enough to lift verbatim - while FAQPage JSON-LD is the machine-readable twin of that block: cheap to add, honest about your structure, and still worth shipping even though Google stopped showing FAQ rich-result dropdowns for most sites in 2023. Treat the markup as a growth hack, though, and you build a spam page. Here is how I do it now, question five deleted.

The 2023 reset: what FAQPage still does

For years, FAQPage markup earned an expandable FAQ dropdown right in the Google results. In August 2023, Google restricted that treatment to well-known, authoritative government and health sites, and for everyone else the rich result quietly disappeared - the same week Google also retired HowTo results.

So why bother? Because the rich-result dropdown was never the real prize (a nice prize, I’ll admit - I liked watching those accordions unfold in the SERP!). The markup still does two durable jobs:

• It tells any parser - Google, an LLM crawler, a RAG pipeline - “this text is a question, and this text is its answer,” unambiguously.
• It pairs with your on-page accordion as a single source of truth, so the visible content and the structured data can never drift apart.

The AEO value moved from earning a SERP feature to being trivially extractable by machines that answer questions, and that value went up, not down, as ChatGPT, Perplexity, and AI Overviews became the surfaces people actually read. Did Google take something away in 2023? Yes. Did it take away the thing that matters in 2026? I would argue no.

	Pre-2023 goal	2026 AEO goal
Primary payoff	FAQ dropdown in SERP	citation in an AI answer
Who reads it	Google’s rich-result renderer	LLM crawlers + RAG retrievers
Winning signal	valid markup	short, extractable, real answers
Failure mode	invalid schema	invented, keyword-stuffed questions

Write the answer first

An answer engine extracts the smallest span that resolves the query. If your answer opens with “It depends on several factors, but first let’s understand…” the extractable span is buried three sentences deep, and the engine may skip you for a competitor who led with the answer.

The rule is boring and it works: first sentence answers the question; the rest earns trust.

• Lead with a direct, self-contained answer - 40 to 60 words, no pronouns that need the question to resolve.
• Then add the caveat, the example, the nuance.
• Keep each answer to one idea. Two ideas want two questions.

This is the same discipline behind a good TL;DR, applied per question. My own test is low-tech: I read the first sentence of each answer with the question covered, and if I cannot tell what was asked, the sentence probably leans on a pronoun it shouldn’t. Ten seconds per item, and it catches more than any validator.

Map People Also Ask, do not invent

The questions must be real. The cheapest source of real demand is the SERP itself: People Also Ask boxes, “related searches,” and the query-fan-out an AI engine runs when it decomposes a topic. Those are questions people actually type, so answer them and you match live demand; invent them and you match nothing. Where did my question number five come from, then? From the keyword list - never from a person.

A practical filter: if you would not say the question out loud to a colleague, cut it. (Try it - it is remarkable how many FAQs die the moment they have to be spoken!)

One accordion, one schema, one source

The clean pattern is single-source: author the Q&A once, render it as a visible accordion, and generate the FAQPage JSON-LD from the same data. Google’s own guidance is explicit that the marked-up content must be visible on the page - hidden-only FAQ markup violates the structured-data policy - and single-sourcing makes that automatic while killing the drift between what users see and what machines read. How often do the two drift when maintained separately? In my experience, the moment a second person edits the page, which seems to be roughly always.

On this site, the post’s faqs front-matter is exactly that source: it drives the on-page accordion and the FAQPage block in one pass. If you want to check whether your answers are actually liftable - short enough, answer-first, self-contained - run a page through our answer-capsule audit.

Checklist

• Every question comes from real demand (PAA, related searches, support tickets) - none invented for keywords.
• Each answer resolves the question in its first sentence, 40-60 words, no dangling pronouns.
• One idea per answer; no near-duplicate questions.
• The Q&A is visible on the page, not schema-only.
• FAQPage JSON-LD is generated from the same source as the visible accordion.
• The FAQ count matches what the page genuinely answers (usually three to eight).
• No FAQPage markup on thin, gated, or duplicate pages.

Does it still move the needle?

Honestly: the SERP dropdown is gone for most sites, so do not ship FAQ markup expecting a search feature. What it reliably does in 2026 is make your best answers extractable - and extractability is the thing AI answer engines reward. A page whose answer is a clean, self-contained span tied to a real question is far likelier to be lifted into an AI answer than the same facts buried in prose. Is that a guarantee of a citation? No - nothing is, and anyone selling a guarantee here is selling something else.

Answer-first Q&A 100%

Buried in prose 20%

source: internal audits, 2026

The chart is illustrative, but the mechanism is not: an engine lifts the shortest span that answers the query, and answer-first Q&A hands it that span on a plate.

In short

• Google retired FAQ rich results for most sites in 2023 - the AEO value is now extractability, not a SERP dropdown.
• Answer the question in the first sentence; 40-60 words, self-contained.
• Source questions from People Also Ask and real demand - never invent them to hold keywords.
• Single-source it: one accordion, one FAQPage schema, generated from the same data.
• FAQ turns to spam the instant questions serve keywords instead of readers - and marking it up only makes the padding machine-readable.

And question number five? Still deleted. The page is better for it, and so, I suspect, is the reader who never had to scroll past it.

Sources

• Google Search Central - FAQ (FAQPage) structured data: the Aug 2023 change limiting FAQ rich results to authoritative government and health sites, and the requirement that marked-up FAQ content be visible on the page.
• schema.org/FAQPage - the vocabulary definition for a page of frequently asked questions and its Question/acceptedAnswer structure.
• Aggarwal et al., “GEO: Generative Engine Optimization,” KDD ‘24 - arXiv:2311.09735: citation in generative answers depends on passage extractability, which answer-first Q&A maximizes.