FAQ page or H2 sections? What AI models actually cite

Put the answers to buyer questions inside the page that owns the topic, as H2 sections with the question as the heading. Keep a separate FAQ page only for questions that fit nowhere else. FAQPage schema will not make an AI assistant cite you.

That is the short version. The rest of this post is the data behind it, because until recently the standard advice, including ours, was to publish a FAQ page with ten buyer questions. A practitioner who does GEO for clients told us the models prefer H2 sections and avoid FAQ hubs. Instead of arguing, we measured.

What we measured

LLM Audit runs buyer questions live against ChatGPT, Gemini and Claude and stores the full answers. Some of those answers include URLs. We took every live answer from the last 60 days, pulled out every URL the models cited, and classified each one by its path: FAQ or help-center pages, blog or news articles, the home page of a domain, or any other page (product, pricing, services).

Over 60 days to August 2026:

• 505 live answers to buyer questions
• 438 URLs cited, all of them by ChatGPT (322) and Gemini (116)
• 400 of 438 pointed at a home page (91 percent)
• 38 pointed at a product, pricing or services page
• 0 pointed at a FAQ or help-center page
• 0 pointed at a blog or news article
• Claude cited no URLs at all in this sample

Two honest caveats. The classification is by URL pattern, so a FAQ living at /support without the word in the path would be counted as "other page"; that bucket is 38 URLs, so even if every one of them were a hidden FAQ, FAQ pages would be under 9 percent against 91 percent for home pages. And citing a URL is not the same as having learned from it: a model trained on your FAQ can answer from it without ever linking to it. What the data does say, clearly, is that when a model points a buyer somewhere, it points at the page that owns the brand or the product. Not at the FAQ hub.

The script is a few lines of classification over our own database; we re-run it before we change what the audit recommends, and the audit copy now reads "answer these questions inside the product pages first, as H2 sections". What the models cite drove that sentence, not opinion.

Why the page that owns the topic wins

Think about what the model has to do with "does [brand] invoice in pesos" or "is [brand] worth it for a team of three". It needs a passage that names the product and answers the question in one place. A product page with an H2 that reads "Invoicing in Argentine pesos" followed by two sentences gives it that passage, on the page whose URL already carries the brand's authority. A FAQ hub gives it the same two sentences on a page that is about nothing in particular and that competes with forty other questions for the model's attention.

There is a second, quieter effect. The entity the model is learning about is your product, and everything that describes the product on the product page strengthens that association. The same text parked on /faq is evidence about a page called FAQ. Extraction and association both favor the page that owns the topic.

This is the same principle as our earlier guidance on getting cited: lead with the answer, keep the block short, phrase the heading the way the buyer asks. The addition here is where the block goes.

What FAQPage schema does and does not do

Two facts from Google, both verifiable on its own pages.

First, in August 2023 Google reduced FAQ rich results to "well-known, authoritative government and health websites" (Google Search Central, FAQPage). A SaaS or a local business marking up its FAQ no longer gets the expanded result in classic search.

Second, Google's guidance for AI Overviews and AI Mode says: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add" (Google Search Central, AI features).

For the other assistants we have no comparable statement and no data of our own that FAQPage markup moves an answer. So our position is plain: keep structured data where it describes something real (Organization, Product, Article, a truthful FAQ block), because it costs nothing and helps machines confirm what the page is about. Do not expect it to produce a recommendation, and do not let a tool tell you "missing FAQ schema" is why you are invisible. Our own audit is not allowed to claim a schema is missing unless it actually fetched the page and looked.

For the record, posts on this blog carry a short FAQ block with FAQPage markup, and the questions live on the post that owns the topic. That is the rule applied, not an exception to it.

How to move a FAQ page into the product pages

Take a worked example: a nail salon in Buenos Aires with a "Preguntas frecuentes" page holding twelve questions. Three of them decide bookings: do you come to my home, how much is a gel manicure, which neighborhoods do you cover. On a FAQ hub those three answers sit on a page the models have no reason to cite. On the services page, as three H2 sections with a two-sentence answer and the price each, they sit on the page the models point at when they name the salon.

The procedure:

1. List the questions on the FAQ page. Mark the ones a buyer asks before deciding (price, fit, coverage, integrations, migration). Those move.
2. For each one, find the page that owns the topic: pricing questions go to pricing, "does it work with X" goes to the integration or features page, "who is it for" goes to the home or the audience page.
3. Add an H2 phrased as the question, or as the short noun phrase the buyer would type, then answer in 40 to 60 words. Lead with the answer. Numbers and names, no hedging.
4. Leave on the FAQ page only what belongs nowhere else: account, billing, support logistics. That page stays useful for humans; it just is not where you send the buyer questions.
5. Keep the old FAQ URL alive, or redirect it to the page that took the most questions.

Then measure it. Ask the same buyer questions again a few weeks later, and compare against the cells where nobody was named before. Re-measuring on the same questions is the only way to tell whether the move worked, rather than assuming it did.

If you would rather not build the question list by hand, the free audit reads your site, writes the buyer questions, asks ChatGPT, Gemini and Claude each one live, and shows which questions nobody is answering yet. Those are the H2 sections to write first.