The FAQ page is an underrated AI visibility tool — how to do it properly

The FAQ page is the most underrated page type on Finnish company sites, and AI search turned it into one of the most valuable. The question and answer format is exactly the structure generative search engines move in: when someone asks ChatGPT the same thing your customers ask you on the phone, a well-made FAQ is a ready answer in a ready format. Here is how to make an FAQ page genuinely useful, for the customer, for the search engine and for the language model.

Why FAQs work so well in AI search

Three reasons that together make the format exceptional:

The format matches how people ask. People put whole questions to an AI, not keywords. An FAQ page is already a set of question and answer pairs, so the model does not have to mine an answer out of an essay.

The answers are self-contained. A good FAQ answer works when lifted out, which is the core requirement of quotability. The passage ”Delivery takes 3 to 5 working days, and next-day delivery is available for an extra fee” can move into an answer as it stands.

FAQPage schema makes the pairs machine-readable. When an FAQ is marked up with structured data, the question and answer relationship is not left to interpretation. The same markup also opens the possibility of appearing in Google’s rich results.

Write the questions people actually ask

The most common fault in FAQ pages is that the questions were invented in a marketing meeting rather than collected from customers. ”Why choose us?” is not a question anyone asks. It is an advert with a question mark, and both reader and machine recognise it.

Real questions come from three places: sales and customer service emails and calls, question-shaped searches in Search Console, and what AI services already answer to questions in your field. Collect 15 to 30 real questions and organise them by theme: price, schedule, process, contract terms, technical conditions. Put price and schedule questions first, because they are asked most and answering them directly separates you from competitors who hide.

The length and shape of an answer

A good FAQ answer is two to five sentences: the first answers, the rest qualify. A one-word answer gives no system anything to lift; a five-hundred-word answer belongs in an article, not an FAQ. If a question needs long treatment, give a concise answer and link to a deeper article, so the FAQ also works as a node in your internal linking.

Answer the uncomfortable questions directly too. ”What does the service cost” deserves a range, not ”get in touch and we will talk”. An evasive answer is the same as no answer in AI results, because the model quotes the competitor who gave the number.

One combined page or an FAQ on every page?

Both, for different jobs. Service and product pages should carry the questions specific to that topic, because they support the page’s own search visibility and answer the buyer in the right context. A combined FAQ page gathers the general questions — pricing model, contracts, process — into one place that customer service and AI systems find easily. Just avoid repeating the same question identically across several pages: give each question one home and link to it.

Technical implementation in WordPress: details and summary elements or a plugin accordion, with FAQPage schema generated automatically from the content. The markup can be checked with Google’s rich results test.

Why a question format suits a language model

An FAQ section is not magic in itself. It works because it happens to match how questions are answered.

When a model builds an answer, it looks for a passage that addresses the question and stands on its own. A question heading followed by a direct answer is exactly that: the heading states the question in the user’s own words, and the paragraph below can be lifted without needing anything around it for context. A page of flowing prose may contain the same information and still lose, because every useful sentence depends on the one before it.

That is the whole mechanism, and it explains the failure mode too. An FAQ where the answers are vague or promotional matches the shape but has nothing worth quoting, and gets passed over.

Writing an answer that can be quoted

Four habits separate an answer that gets used from one that does not.

• Answer in the first sentence. No wind-up. The qualifications and exceptions come after, once the reader already has the answer.
• Keep it to two or three sentences. Long enough to be complete, short enough to quote whole.
• Include a number where one exists. A price range, a timescale, a percentage. Numbers make an answer verifiable and are the part most often carried into a summary.
• Answer the real question. If the honest answer is that it depends, say what it depends on rather than avoiding the question.

The last point is where most FAQ sections go wrong. A question like what does this cost, answered with a note that every project is different, tells the reader nothing and gives a model nothing to quote. A range with the factors that move it within the range does both.

Finding the questions worth answering

Invented questions produce an invented FAQ. Four sources give you the real ones, and they cost nothing.

Your own inbox and phone notes come first, because the questions customers actually ask are already recorded there. Search Console is second: filter your queries for ones beginning with how, why, what and whether, and you have a list of phrasings in the customer’s own words. Third, the related questions box in search results shows what else is asked around your topic. Fourth, ask an AI service your own main question and note which follow-ups it raises.

Eight to twelve questions per page is a workable range. Fewer leaves obvious gaps; many more usually means the page is trying to cover two topics and would serve both better if split.

Structured data, and what it is for

FAQPage markup describes your question and answer pairs in a form a machine reads without having to interpret the layout. It is worth adding, with an accurate expectation of what it does.

Google removed FAQ rich results from most search listings, so the visible reward in search results is largely gone. The markup still matters because it removes ambiguity: a service reading the page does not have to guess which text is a question and which is its answer. For a page built around questions, that is the difference between being parsed correctly and being parsed approximately.

Two rules keep it valid. The markup must match what a visitor sees on the page, and every answer must be complete in the markup rather than truncated. Marking up content that is hidden or different from the visible version is a policy violation and gets the page ignored.

Accordions, and whether to use them

The usual worry is that answers hidden behind a click are invisible to search engines and models. In practice the answer depends on how the accordion is built.

If the text is in the page source and merely hidden with CSS until clicked, it is read normally. The native details and summary elements work this way and need no scripting. If the text is fetched by JavaScript only when the section opens, some crawlers never see it, because they do not execute scripts.

The test takes seconds: view the page source and search for a sentence from a closed answer. If it is there, the accordion is fine. If it is not, the answers need to be in the markup from the start.

Keeping an FAQ from going stale

An FAQ ages faster than the rest of a site, because it is where the prices, timescales and product details live. A section listing last year’s prices actively misinforms the models reading it.

Twice a year, read the answers and check three things: are the numbers current, do the questions still match what customers ask, and has anything been added to the product that raises a new question. Remove questions nobody asks any more rather than letting the list grow indefinitely.

Adding a visible update date helps both readers and models judge whether the information still holds. It costs one line and it is the single easiest signal of currency you can give.

Frequently asked questions

Why is an FAQ page important for AI search?

Because people put whole questions to an AI and an FAQ offers the answers ready in the same format. Marked up with FAQPage schema, the question and answer pairs are also machine-readable, which makes them easier to lift into answers.

How many questions should an FAQ page have?

Quality before quantity: 10 to 30 genuine questions suits most. More important than the number is that the questions are actually asked and the answers are direct.

Should price questions be answered publicly?

A range is worth giving. The companies that reach AI answers on pricing questions are the ones that state a number, and hiding does not stop the customer seeing a price. It only ensures they see a competitor’s price first.

Where does FAQPage schema go?

On every page that has an FAQ section, describing that page’s questions. In WordPress the markup can be produced automatically from the theme or an SEO plugin, so it stays in step with the content by itself.

Does FAQ structured data still do anything now that Google removed the rich results?

The visible reward in search listings is largely gone, but the markup still tells a machine which text is a question and which is its answer instead of leaving it to be inferred from layout. On a page built around questions that is the difference between accurate parsing and approximate parsing.

Are answers hidden inside an accordion invisible to AI services?

Only if the text is loaded by JavaScript when the section opens. If the answer is in the page source and merely hidden with CSS, it is read normally. View the source and search for a sentence from a closed answer to find out which case applies.

Building FAQ sections and automating their schema is part of our GEO work. On this site every FAQ produces its FAQPage markup by itself. We cover writing question-shaped content more broadly on the content production service page.