An FAQ is the most agent-friendly format on your whole site. It is literally a list of questions and answers, which is exactly what an assistant is trying to assemble. Yet most FAQs are wasted, because they’re built for a support deflection metric, not for being quoted. A few changes turn the same section into one of your strongest sources of AI citations.
1. Use the questions people actually ask
Write each question in the words a customer would type into ChatGPT, not in internal or marketing phrasing. “How much does it cost?” beats “Investment & value.” “Can I cancel anytime?” beats “Commitment flexibility.” The closer your question matches the user’s real query, the more likely an agent maps their question to your answer.
To find these, look at support tickets, sales call recordings, and search queries. The questions that keep coming up are the ones your FAQ should lead with.
2. Answer in the first sentence
The opening sentence of each answer should resolve the question completely and stand on its own, because that’s the unit an agent quotes. Add detail afterward if it helps, but never make the reader (or the machine) wait for it.
Weak: “Great question! We know returns can be stressful, and we’ve worked hard to make ours painless…”
Strong: “You can return any item within 30 days for a full refund. Start a return from your account; we email a prepaid label.”
3. One question per entry
Don’t bundle “How do I sign up, and what does it cost, and can I cancel?” into one block. An agent extracts a Q&A as a pair; three questions sharing one answer means none of them gets a clean, quotable response. Split them so each question owns its answer.
4. Mark it up with FAQPage schema
Clear Q&A copy gets you most of the way; FAQPage JSON-LD closes the gap by telling an agent this is the question, this is the answer, with no parsing of your HTML required. Every post on this blog uses it (including this one). It won’t rescue a vague answer, but it removes all ambiguity from a clear one.
{ "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [{ "@type": "Question", "name": "Can I cancel anytime?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. Cancel in one click from your account; you keep access until the end of the billing period." } }] }
Get schema generated from your own page
Agent Speak reads your content, finds the gaps, and hands you a starter JSON-LD block built from what’s actually there, paste-ready.
Common questions
Yes. FAQPage JSON-LD labels each question and answer explicitly, so an agent doesn’t infer structure from your HTML. It can’t rescue a vague answer, but it removes ambiguity from a clear one. See how to get quoted by AI for the wider picture.
Lead with a one-to-two-sentence answer that stands alone, then add detail if needed. The first sentence should resolve the question completely. That is the part an agent is most likely to quote.
Both a dedicated FAQ page and inline FAQs on relevant pages work. What matters is that the Q&A pairs are clean and marked up; an agent cares about structure and clarity, not location.
Written by Michelle Savage, creator of Agent Speak. Related: Write for humans and AI at once.