How to Write an FAQ Page That Google and ChatGPT Can Use

Most FAQ pages are written once, at launch, by someone guessing what customers might ask. They answer "Why choose us?" and skip "Can I change my delivery address after I order?". The result is a page nobody reads, an inbox full of the questions it should have answered, and nothing a search engine or an AI assistant can usefully quote.

This guide is for whoever owns the FAQ page and wants it to work for three readers at once: a customer arriving from Google, an AI answer engine such as ChatGPT search or Google's AI Overviews, and your own AI support agent answering in chat. All three want the same thing.

An FAQ page is a single web page that lists the questions customers ask most often, each followed by a short, complete answer. It is not a help center, which holds full articles for tasks and troubleshooting. An FAQ page holds the short answers people want before they buy or before they write to you.

What makes a good FAQ page?

A good FAQ page answers real questions, uses the customer's words, puts the answer first and stays current. Everything else in this guide follows from those four rules.

Nielsen Norman Group has made the case for years. In FAQs Still Deliver Great Value, Susan Farrell notes that NN/g has warned against made-up questions since 2002, and that questions "should echo the real concerns people contact you about", in your visitors' own vocabulary. The same article puts the reason in one line: "People don't search for your solution, they search for their problem."

The same rules make a page easy to retrieve, because search engines, AI assistants and support agents all look for a passage that matches the question and states the answer.

Where do you find the questions for an FAQ page?

Find FAQ questions where customers already ask them: your support inbox, chat transcripts, Search Console queries and sales calls. Collect a few weeks from each source, write every question in the customer's words, and rank the list by how often each one appears.

Your support inbox

Export the last four to six weeks of email and contact-form messages, and write the question each customer was really asking on one line, in their words. "Hi, I ordered on Monday and haven't heard anything" becomes "When will my order ship?". Merge duplicates and count them. The guide to reducing support tickets walks through the same exercise, including how to sort contact reasons into types.

Chat transcripts and AI agent topics

Chat is the richest source, because customers type the question the way they think it. If you run an AI support agent, read the conversations it escalated: each is a question your content did not answer. In PepoChat, the Analytics topics card labels every conversation from the visitor's first message and ranks the most asked topics over the last six weeks, so a growing topic shows up without any tagging work.

Search Console queries

Google Search Console's Performance report groups your search traffic by the query people typed. Add a filter on Queries, choose Custom (regex), and enter ^(how|what|why|when|where|which|can|do|does|is|are)\s to see only question-shaped searches that already reach your site. Questions with many impressions and few clicks are often ones your pages only half answer.

Sales and onboarding calls

Questions asked before a purchase come up in demos and sales calls, not in support. Ask whoever runs those calls for the objections they hear every week: "Do you integrate with X?", "Can I cancel any time?", "Is there a setup fee?". As the NN/g article puts it, a good salesperson can probably dictate the FAQs you need.

A first version with 15 to 30 questions is plenty for most small businesses. Delete any question you cannot trace back to a real customer.

How should you phrase FAQ questions?

Phrase each question the way a customer would type it into a search box or a chat window: a full question, in plain words, naming the specific thing they are asking about. If a customer would not say it, do not publish it.

• Use the customer's nouns. "Can I return a sale item?" rather than "Returns policy for discounted merchandise". Customers say "cancel", not "subscription termination".
• One question per entry. "How much does shipping cost and how long does it take?" is two entries, because some people only want one answer.
• Write from the customer's side. Real questions start with "Can I", "How do I", "Do you" and "What happens if". Questions that start with "Why should" are usually marketing.
• Keep it short enough to be a heading. Most good FAQ questions are under 12 words.

Do not write five versions of one question to catch every phrasing. Search engines and AI systems match meaning, not exact strings.

How do you write an FAQ answer?

Answer the question in the first sentence, then add what the customer needs to act: the condition, the number, the date and the next step. A reader who stops after one sentence should still have the answer.

An answer-first entry has four parts, in this order:

1. The answer. Yes, no, a number, or a one-line instruction.
2. The conditions. Who it applies to, the exceptions and the time limits.
3. The detail or steps. Only as much as the customer needs to act.
4. The link. To the full policy or a help article, if one exists.

Write numbers instead of adjectives. "Fast shipping" invites a follow-up question; "Orders placed before 2 pm on a business day ship the same day" does not. Add dates where they matter, and repeat the subject inside the answer: "Yes, you can return sale items within 30 days" beats "Yes, within 30 days", because the answer will often be read without its question, in a search snippet, an AI summary or a chat reply.

Most answers fit in 40 to 120 words. When one needs more than five or six steps, or screenshots, move it into a help article and keep a two-sentence summary with a link on the FAQ page. NN/g recommends the same for long FAQs: a list of questions that lead to separate answer pages.

How should you structure an FAQ page?

Group the questions by topic, order them by how often they are asked, give each question its own heading, and put a list of jump links at the top. Keep every answer in the page HTML, so readers, search engines and AI crawlers all see the same text.

Group by customer task and order by demand

Name the groups after what the customer is doing: "Ordering", "Shipping and delivery", "Returns", "Billing", "Your account". Inside each group, put the most asked questions first, using the counts you collected. A group with more than about ten questions is usually two groups.

One question per heading, with an anchor link

Mark each question as a heading, for example an H2 per group and an H3 per question, and give every heading a stable id so your team can link straight to one answer, such as /faq#cancel-subscription, from emails and chat replies. A list of questions at the top, each linking to its answer, lets readers rule out whole topics at a glance; NN/g calls the question list followed by the answers "still the best underlying design" for short and medium-length FAQs.

Accordions are fine when the answer is in the HTML

An accordion, a question you click to reveal its answer, keeps a long page short on a phone. In a 2023 article on accordions, Nielsen Norman Group noted that collapsed content is easier to miss, and that when users need several answers at once it is better to show everything. Google's FAQ guidelines, while they existed, accepted answers inside expandable sections; a July 2021 entry in the Search Central documentation updates says "clicking an expandable section to view the answer is a valid use case".

The real risk is an answer that JavaScript fetches or inserts only on click. In a December 2024 analysis of crawler traffic, Vercel found that none of the major AI crawlers it measured rendered JavaScript, including OpenAI's and Anthropic's; Google's Gemini and AppleBot did. The native <details> and <summary> elements, or a panel collapsed with CSS, keep the text in the HTML. To check, open View Page Source and search for a sentence from one answer. If it is not there, crawlers that do not run scripts cannot see it.

Do you still need FAQ schema in 2026?

Not for Google. FAQPage structured data is still valid schema.org markup, but Google limited FAQ rich results to well-known government and health websites in August 2023 and stopped showing them altogether on May 7, 2026. Keep the markup if your CMS already adds it, and do not expect a visible result from it.

FAQPage structured data is a block of JSON-LD, machine-readable data inside a script tag, that labels each question on the page and its accepted answer using the schema.org vocabulary. A minimal version looks like this:

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "Can I return a sale item?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes. Sale items can be returned within 30 days of delivery for store credit. Items marked Final Sale cannot be returned."
      }
    },
    {
      "@type": "Question",
      "name": "How long does shipping take?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Orders placed before 2 pm Eastern on a business day ship the same day. Standard delivery takes 3 to 5 business days in the US."
      }
    }
  ]
}

The block goes inside a <script type="application/ld+json"> tag on the FAQ page, and every question and answer in it should match the visible text word for word.

For years this markup could turn a search result into an expandable list of questions. Google announced the first restriction in Changes to HowTo and FAQ rich results on the Search Central blog, and its documentation changelog records the rest:

The 2023 post also said there is "no need to proactively remove it", because unused structured data "does not cause problems for Search, but also has no visible effects". Google's guide to its AI features adds that structured data is not required for generative AI search.

So if your CMS already outputs FAQPage markup that matches the page, leave it; the posts on this blog still output it, generated from the same text readers see. Do not spend engineering time adding it, and never mark up questions that are not visible on the page. Rewriting weak answers will do more.

Why do AI answer engines quote clear Q&A pages?

Because they retrieve passages, not whole websites. An AI answer engine is a search tool, such as ChatGPT search or Google's AI Mode, that writes a direct reply from the pages it retrieves and links to them. A question heading followed by a sentence that states the answer is easy to find and quote, though nobody can promise a citation.

The evidence is modest but consistent:

• Google grounds its AI features in its index. Google's guide to optimizing for generative AI features says AI Overviews and AI Mode use retrieval-augmented generation, "also known as grounding", over the Search index, and a page must be indexed and eligible for a snippet to be used. It recommends headings "that provide a clear structure", and lists "chunking" content, llms.txt files and rewriting pages just for AI among the things you can ignore.
• ChatGPT search has its own crawler. OpenAI's crawler documentation says OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features", and that sites which want to appear there should allow it in robots.txt. GPTBot, which collects content that may be used to train OpenAI's models, is listed separately.
• Most AI crawlers read the HTML as delivered. That is the Vercel finding above, and the reason answers must not depend on JavaScript.
• Specific facts help. In the paper GEO: Generative Engine Optimization, accepted at KDD 2024, researchers found that adding statistics, quotations and source citations raised a page's visibility in generative engine answers by up to 40% on their benchmark, while keyword stuffing did little. The effect varied by topic.

Together these support the habits above: a checkable answer near the top of each entry, numbers in the text, everything in the HTML, and no robots.txt rule blocking the crawlers you want. They do not support a promise. A clear FAQ page improves your odds of being quoted without guaranteeing it.

How do you keep an FAQ page current?

Give the page an owner, update it the day a price, policy or feature changes, and review the whole page every month. Show a "last updated" date so readers can see the answers are current.

An FAQ page goes stale quietly: nobody reports a wrong shipping price, they just write to you after being charged it. Four habits prevent that:

• Name one owner. The owner does not write every answer, but makes sure each one gets written and checked.
• Update on change. List the facts the FAQ depends on, such as prices, shipping times and the returns window. When one changes, search the page for the old value and fix it the same day.
• Review monthly against new questions. Compare what the inbox, chat topics and Search Console show with what the page covers. Add new questions and delete the ones nobody asks any more.
• Keep one version of each fact. If the returns window appears on the FAQ, the returns page and a PDF, they will drift apart. Contradictory pages are a common cause of wrong AI answers, as the guide to stopping chatbot hallucinations explains.

Keep anchors stable too, since old emails and chat replies link to them. If you keep FAQPage markup, schema.org's lastReviewed property records the date the page was last reviewed for accuracy.

How does an FAQ page help your AI support agent?

An AI support agent answers from the content you give it, and a good FAQ page is one of the best sources you can add: short, factual and already written as questions and answers. In PepoChat you can import it with the rest of your site, or write the same answers as Q&A pairs or Help Center articles.

A knowledge source is one item in the agent's knowledge base, such as a crawled website, an uploaded file or a connected app. PepoChat's agent searches the knowledge base on every reply and escalates to your team inbox when it finds nothing relevant. Three documented details matter here:

• A crawl counts as one source. The website import fetches a single URL, a sitemap, or a crawl of up to 200 pages and 3 levels deep. A whole crawl or sitemap counts as one knowledge source, which matters on the Free plan's limit of five. Pages are re-fetched weekly and re-indexed only when their words change.
• Only the main content is kept. Navigation, headers, footers and cookie banners are left out, and a JavaScript-only page is read as an anonymous visitor's first request sees it.
• Q&A pairs and Help Center articles do not count as sources. A Q&A pair stores the question with its answer, so retrieval matches the way visitors phrase it; a workspace holds 100 on Free and 1,000 on a paid plan. Published Help Center articles appear in the widget's Help tab and are mirrored into the knowledge base automatically: 50 on Free, 300 on Starter and Growth. Their text still counts toward the plan's knowledge-base text cap.

What to do next

Start with the inbox. Spend an hour this week listing the last month of questions in the customers' own words, and rewrite your five most asked answers so the first sentence answers the question. Then check View Page Source for one answer, and your robots.txt for the crawlers you want.

If you want an AI agent to answer from the page, the guide to training an AI chatbot on your website and PDFs covers importing and testing. To try it on your own FAQ, create a free PepoChat workspace. The website crawl, Q&A pairs and the Help Center are all on the free plan, which includes 200 AI replies a month and 5 knowledge sources, with no credit card.