FAQ content for AI search, how to write answers engines can use

Why does FAQ content suit AI search?

AI engines answer questions. Content that is already shaped as a clear question with a direct answer is easy for them to find, understand and reuse.

A good FAQ answer is a small, self-contained passage. It names the question, gives the answer and adds a fact or two. An engine building a reply can take that passage whole. It does not need to piece together an answer from three paragraphs spread across a page.

This is not a trick. It is how a helpful page for a person looks too. Someone scanning your service page for "how long does it take" wants the same short, clear answer that an AI engine does.

What happened to FAQ rich results in Google?

Google has phased them out in two steps. In August 2023 Google announced it would limit FAQ rich results to well-known, authoritative government and health websites. For almost every business site, the expandable FAQ boxes under a search result stopped showing.

Then in May 2026 Google deprecated the feature entirely. Its documentation updates page records that FAQ rich results would no longer appear in Search from 7 May 2026. In June 2026 Google removed the documentation, noting the feature is no longer shown.

So the visible reward for FAQ markup in Google has gone. That changes the reason for writing FAQs. It is no longer about winning extra space on a results page. It is about answering real questions well.

Should you keep FAQPage schema?

Keep it if it is already there, and add it if it is quick. It is no longer something to chase.

FAQPage is still a valid type on schema.org. It labels your questions and answers clearly for any system that reads markup. It does no harm when it matches the page. Google's guide to generative AI features also says structured data is not required to appear in its AI features.

Two rules still apply. The markup must match the visible questions and answers word for word. And it must describe genuine questions, not adverts dressed up as questions. The schema markup guide has a FAQPage example you can adapt.

Where do the best FAQ questions come from?

From the people who buy from you. The best questions are the ones customers already ask, in their own words.

Sales enquiries are the most underused. A dental group's reception team hears the same ten questions every week. Those questions are almost certainly being asked of AI engines too. The People Also Ask box is a close second, as it shows the follow-on questions around a topic.

The prompt research guide covers these sources in more detail, including how to turn them into prompts you track.

How do you write an FAQ answer an AI engine can use?

Open with the answer. Then add the facts that make it useful. Keep it to one short paragraph.

Here is a weak answer to "how long does it take to fit new windows".

At Example Windows, we pride ourselves on a first class installation service. Every home is different, and our friendly team will always work around you to deliver the best result.

It never answers the question. Here is a better one.

Most homes take one to two days. A typical three bedroom house with eight to ten windows is usually done in a day. Larger homes, bay windows or listed buildings can take longer. We confirm the timing at your survey.

The second answer works for a person and for an engine. It gives the answer first, then a specific example, then the exceptions. It could be lifted into an AI answer as it stands.

A few habits help every answer.

1. Answer in the first sentence. Yes, no, a number, a time or a short statement.
2. Add one or two specific facts. Prices, times, areas, standards or steps.
3. Make it stand alone. Avoid "as mentioned above" or "see below".
4. Use the words people use. If buyers say "worktops", do not only say "work surfaces".
5. Stay factual. Leave out claims like "the best" unless you can back them up.

Where should FAQs go on your site?

Mostly on the pages they belong to. A question about implant costs belongs on the implants page. A question about your service area belongs on the location page.

This helps in two ways. The page around the FAQ gives it context, which makes the answer easier to understand. And it keeps the page focused on one topic, which is easier for both people and engines to follow.

A central FAQ page still has a place. Use it for general questions that apply across the business, such as payment options, guarantees, complaints and opening hours. Link from it to the relevant service pages for anything more specific.

Do FAQs work for B2B and professional services too?

Yes, and often better. B2B buyers ask a lot of detailed questions before they ever speak to sales. Many of those questions now go to an AI engine first.

A B2B software company might get asked whether it integrates with a certain accounting package, where data is stored, how long setup takes, and what happens at the end of a contract. Those are exactly the questions a buyer types into ChatGPT when drawing up a shortlist. If the answers are clear on your own pages, the engine has something accurate to work from.

Professional firms are similar. An accountant, a solicitor or a training provider gets asked about fees, timescales, eligibility and process. Plain, honest answers to those questions are useful to buyers, and they are hard for competitors to copy word for word because they describe how you work.

How often should you update FAQ answers?

Whenever a fact in them changes, and on a regular review otherwise. Every few months suits most businesses.

Out of date FAQs cause real problems. An old answer saying you offer free surveys, when you now charge, can be picked up and repeated by AI engines for months. That is one of the common ways AI gets your business wrong.

Give each FAQ an owner, usually the person responsible for that service. Add the review to the same routine you use for price lists and opening hours. If an answer has a date or a price in it, note when it was last checked.

What mistakes make FAQ content less useful?

Most of them come from writing FAQs for search engines rather than for customers.

• Questions nobody asks, invented to fit a keyword.
• Answers that repeat the question and then say nothing.
• Sales copy in the answer instead of facts.
• The same FAQ block copied onto every page of the site.
• Markup that does not match what is on the page.
• Answers that were right two years ago.

Each of these makes the content less likely to be useful to a person. In our experience, content that is not useful to a person is rarely what an AI engine chooses to cite.

How Axiom GEO helps

On the Premium plan and above, QA Finder pulls People Also Ask questions, Google autocomplete and questions from Reddit threads. It also extracts questions from pasted enquiries or call notes, ignoring personal details and not storing the text. For each question it builds a content pack with a quick answer, an FAQ block with FAQPage JSON-LD, and author and article markup, ready to paste. Content opportunities shows AI answer gaps, where engines cite others and not you. The schema generator builds FAQPage markup for pages that do not have it yet.