How to Write FAQ Pages That Get Cited by AI (with examples)

FAQ pages get cited by AI when each answer leads with a clear, self-contained response in the first sentence, sits under a real question heading, and is backed by FAQPage schema that matches the visible text. The rest of this post is how to do exactly that.

Here's the thing about FAQ sections. Most of them were written to fill space on a page, not to answer a question. They sound like a lawyer wrote them on a deadline. And now that buyers are asking ChatGPT, Perplexity, and Google's AI answers instead of scrolling your site, that lazy FAQ is the single highest-leverage thing you can fix this week.

A good FAQ block is the most extractable content format on the internet. One clear question, one clear answer, sitting in a clean block of HTML. That is exactly the shape an answer engine wants to lift and quote. Get this right and you stop hoping AI finds you. You start handing it the sentence you want it to repeat.

This is the most tactical of our four AEO pillars (the primer lives at what answer engine optimization is), part of the same shift covered in SEO in the age of AI search. No theory you can't use by Friday. And yes, the FAQ at the bottom of this post is itself a worked example of every rule below. Read it twice.

Step one: source the questions buyers actually ask AI

You cannot answer a question you made up. Most FAQ pages fail here, at the very first step. Someone in marketing brainstormed "questions we think people ask" and the real questions never made the list.

Real questions come from real places. Go get them.

Pull from your own inbox and DMs first. The questions people email you and ask in sales calls are gold, because they are phrased the way humans actually phrase things. Then mine the autocomplete. Type your topic into Google and ChatGPT and watch what they suggest. Check the "People also ask" box on a Google results page. Read the reviews and threads where your buyers complain, on Reddit, in Facebook groups, in your competitors' one-star reviews.

The pattern you are looking for: conversational, specific, often a full sentence. "How much does a voice receptionist cost for a small real estate office" beats "pricing." People talk to AI the way they talk to a person. Your questions should sound like that too.

If you sell to real estate brokers, the question is "can a voice receptionist book showings while I'm with a client." If you coach speakers, it's "how do I get booked for paid keynotes without an agent." Those are sentences a human typed into a chat box. Write those down. Those are your questions.

Step two: phrase questions and answers for extractability

This is where the work lives. An answer engine reads your page, decides which chunk best answers the user's prompt, and either quotes it or summarizes it with a citation. Your job is to make your chunk the easiest one to lift cleanly.

Three rules.

Answer first. Put the actual answer in the first sentence. Do not warm up. Do not say "great question." The machine reads top down and weights the opening. Lead with the answer, then add the nuance underneath.

Self-contained. Each answer has to make sense on its own, with zero context from the rest of the page. Assume the reader, human or machine, landed on that one block and read nothing else. No "as mentioned above." No "it depends" without immediately saying what it depends on.

Concise, then complete. Two to four sentences is the sweet spot for the core answer. Long enough to be a real answer, short enough to quote whole. You can add a second short paragraph for detail, but the first 40 to 50 words should stand alone as a complete answer.

Here is the good versus bad pair. Same question, two answers.

Bad. "How long does AEO take to work?" Great question! As you probably know, every business is different and there are a lot of factors at play. SEO has always been a long game, and answer engine optimization is no exception. There are many variables we'd need to discuss before we could give you an accurate timeline for your specific situation.

That answer says nothing. No machine will quote it because there is nothing to quote. No human is reassured by it either.

Good. "How long does AEO take to work?" There is no fixed timeline, and anyone who promises you one is guessing. What drives the speed is how much clean, quotable content you already have and how competitive your topic is. A site that publishes structured, answer-first pages and keeps them current gives answer engines something to cite sooner than a thin or buried page does. So the honest answer is that it depends on your starting point, and the work is what moves it.

The good version leads with a clear, honest answer, stands on its own, and gives the reader something true to take away. That is the bar.

Step three: structure so the machine can parse it

Phrasing gets you halfway. Structure gets you the rest. Answer engines and crawlers read your HTML, not your design. A pretty accordion that hides everything in collapsed divs with no semantic markup is hard to parse, and Google explicitly requires that FAQ content be visible to the user on the page, whether shown directly or in an expandable section. Per Google's current structured data documentation, all FAQ content must be visible on the source page to be eligible for any FAQ treatment.

Keep the structure boring and correct:

• One question per heading. Use a real heading tag, an h2 or h3, for the question. Not bold text. Not a styled span. A heading.
• One answer per block immediately under its question, in a paragraph or short list.
• One Q&A per block. Do not stuff three questions into one answer. Each block should be liftable on its own.
• Logical order. Group related questions. Put the question the most people ask first.

This matters because the cleaner the question-answer pairing in your markup, the easier it is for a model to map "user asked X" to "this page answers X" and pull your exact words.

Step four: add FAQPage schema, and be honest about what it does

Now the part everyone gets wrong, in both directions. Some people think FAQ schema is dead. Others think it is magic. Both are wrong.

Here is the real state of things, straight from Google. In August 2023, Google narrowed FAQ rich results so they only showed for well-known, authoritative government and health sites. Then, per Google's structured data documentation, as of May 7, 2026, FAQ rich results stopped appearing in Google Search entirely. So if your goal was getting that expandable FAQ widget under your search listing, that ship has sailed for almost everyone.

But that is not why you add the schema anymore.

You add FAQPage JSON-LD because it is a clean, machine-readable map of your questions and answers. It tells any parser, Google's or an AI crawler's, exactly where each question ends and each answer begins, with no guessing. Structured FAQ content still helps machine extraction even though the Google rich-result reward is gone (the GEO research on what gets cited points the same direction). The visible, well-structured content is doing the heavy lifting. The schema is the labeled index on top of it.

Two rules with that schema. The text inside each acceptedAnswer must match what a visitor actually sees on the page. Do not stuff the schema with answers that aren't on the page. That is a violation and a fast way to get ignored. And per Google's guidance, FAQPage is for content where there's a single answer per question. If users can submit competing answers, that's a different schema (QAPage), not this one.

Step five: ship it, then watch what gets quoted

Publish the page. Validate the JSON-LD in Google's Rich Results Test or the Schema Markup Validator so you know the syntax is clean. Then do the thing almost nobody does: go ask the questions yourself. Type your own FAQ questions into ChatGPT, Perplexity, and Google's AI answers and see whether you show up. If a competitor's words come back instead of yours, you have your next rewrite.

That loop, write, publish, check who got quoted, rewrite, is the entire game. It's the system, not the secret.

Stat row

• Per Google's August 2023 announcement, FAQ rich results had already been limited to well-known, authoritative government and health sites.
• Per Google's current guidance, all FAQ content must be visible to the user on the source page to qualify for FAQ treatment.

"A site with clean FAQ pages on day one moves faster than one starting from scratch. The cleaner the question-answer pairing in your markup, the easier it is for a model to pull your exact words."

The five-step FAQ checklist

1. Source 8 to 12 real questions from your inbox, sales calls, autocomplete, and "People also ask."
2. Phrase each answer to be answer-first, self-contained, and tight: 40 to 50 words that stand alone.
3. Mark up one question per heading, one answer per block, one Q&A per block.
4. Add valid FAQPage JSON-LD where the schema text matches the visible page text exactly.
5. Validate the schema, publish, then ask AI your own questions and rewrite whatever didn't get quoted.

Want the page graded for you?

You can do all of this by hand. Or you can let the free AEO Audit Tool scan your site and tell you exactly which questions you're missing, where your schema is broken, and which answers are too vague to get quoted. It's free, it ships in June 2026, and it's the same tool we use on our own pages.

If you want the full method (the eight modules on sourcing questions, engineering quotable answers, and tracking your citation rate over time) the AEO Cohort waitlist is open now via the Academy. The cohort runs October 20 to 24, 2026. We built this in our own business first. If it works for us, the goal is for it to work for you.

← Back to Community