An FAQ section is one of the few places on your site where you control the question and the answer. That makes it useful for AI answers in a way a product page is not. A product page has to describe everything. An FAQ can answer one thing at a time, in the words people actually type.
Most FAQ sections fail at this because they were written for a support inbox, not for a prompt. The questions are internal ("What are your SLA tiers?") and the answers are long, hedged, and full of links. Rewriting them for AI search is mostly an editing job, not a new content project.
Write the question the way someone would ask it
People do not type "pricing methodology" into ChatGPT. They type "how much does X cost per seat" or "is X cheaper than Y for a 20-person team."
Pull real question wording from three places:
- Your search console queries that start with how, what, is, can, does, or why.
- Your sales inbox and support tickets, where the same objection shows up in the customer's own words.
- The "People also ask" boxes and autocomplete suggestions for your head terms.
Then write the heading as that question, in that phrasing. "How much does X cost per seat?" beats "Pricing." A heading that matches the prompt is easier for a retrieval system to match to the query, and easier for a person skimming the page to recognize.
One caution: do not write twenty variations of the same question to cover every phrasing. That reads as spam to humans and adds nothing. Pick the phrasing people actually use, and let the answer cover the synonyms naturally.
Keep answers between 40 and 80 words
This range is a working guideline, not a rule handed down by any AI company. It holds up because of how answers get assembled.
An AI answer is built from passages. A passage that is too short often lacks the context to stand alone — it says "yes, on all plans" without saying what "yes" refers to. A passage that is too long gets truncated or split, and the sentence carrying your actual answer may not survive the cut.
Forty to eighty words is roughly two to four sentences: enough to state the answer, add the one qualifier that matters, and stop.
A workable pattern:
- Answer in the first sentence. No preamble.
- Add the condition or exception in the second.
- Add a concrete detail — a number, a limit, a time frame — in the third, if you have one.
Avoid "It depends" as an opener. If the answer genuinely depends, say what it depends on and then give the answer for the common case.
One fact per answer
This is where most FAQ rewrites go wrong. A single answer tries to cover pricing, onboarding time, contract length, and support hours. The result is a paragraph that answers nothing cleanly.
Split it. Each question gets one fact. If a reader would need to remember two separate things, that is two questions.
This also makes your FAQ reusable. A clean, single-fact answer is the kind of passage that gets quoted directly — the same property that makes comparison pages that AI assistants actually quote work. Specific, bounded claims travel; paragraphs that hedge do not.
A quick test: cover the answer and read only the question. Can you predict the shape of the answer? If not, the question is doing too much work.
A note on hedging
Legal and marketing review often adds qualifiers until the answer is technically true and practically useless. "Pricing may vary depending on configuration and is subject to change" is not an answer. Give the real number or the real range, and put the caveat in a footnote or a separate line. If you cannot state a fact publicly, do not write a question that requires it.
Match your visible FAQ to your markup, exactly
FAQPage structured data is worth adding when the questions and answers are visible on the page. Google's documented guidance is that structured data should describe content the user can see. Marking up questions that exist only in the JSON-LD is a mismatch, and it is the fastest way to have the markup ignored.
Practical rules:
- The marked-up question text should match the visible heading. Minor punctuation differences are fine; different wording is not.
- The marked-up answer should match the visible answer, not a longer version you wish you had written.
- Do not mark up every question on a page that has thirty of them if only five are visible without interaction. Mark up what is on the page.
- Do not expect markup alone to produce AI mentions. FAQ schema is a formatting and eligibility signal, not a ranking mechanism, and no vendor documents it as a factor in whether an AI assistant cites you.
If you are building the JSON-LD by hand and keep making small syntax errors, our free schema markup generator builds FAQPage (and Organization, LocalBusiness or Product) markup from a simple form so you can paste valid JSON-LD into the page head.
What to check before you publish
| Check | Why it matters |
|---|---|
| Question matches real prompt wording | Retrieval works on wording overlap |
| Answer is 40–80 words | Survives passage splitting |
| One fact per answer | Quotable without context loss |
| No unverifiable claims | You cannot back up what you cannot cite |
| Markup matches visible text | Mismatched markup gets ignored |
One thing to do this week
Pick your five highest-intent pages and rewrite the first three FAQ questions on each using the pattern above: real prompt wording, a direct answer in the first sentence, one fact, 40 to 80 words. Then check whether the visible text and the FAQPage markup say the same thing, and fix the ones that do not.