TLDR
Question and answer pairs are the native format of AI systems, which makes a well-built FAQ one of the most quotable assets on your site.
Google retired FAQ rich results in May 2026, so the dropdown snippet is gone. The markup remains valid, and the answer-shaped writing matters more than ever.
Source questions from real customers, answer in the first sentence, and keep the markup matched to the visible page.
Why do FAQs matter for AI visibility?
Because assistants answer questions, and a FAQ hands them the question and the answer already paired. When your page asks "how long does a roof replacement take" and answers it cleanly in the first sentence, you have produced exactly the unit of content an AI system can lift and reuse.
Generic FAQs waste that power. The value comes from answering the specific questions your market actually asks, with the specificity a national content site cannot fake.
What happened to FAQ rich results?
Google retired them. FAQ rich results stopped appearing in search in May 2026, after being restricted to government and health sites since 2023. So the old tactic of adding markup to win a dropdown under your listing is finished.
The markup itself is a different story. FAQPage remains a valid type in the schema.org vocabulary, Google says existing markup causes no problems, and other engines and AI crawlers still parse it. Keep it as machine-readable labeling, not as a widget play.
Where do good questions come from?
Three sources beat all guessing. First, the phone: the questions your office answers every week are the questions your market types and speaks. Second, sales objections: whatever makes customers hesitate deserves a public answer. Third, the assistants themselves: ask ChatGPT or Gemini about your trade in your area and note the follow-up questions they raise.
Collect twenty real questions before writing anything. The collection step is where most FAQ pages are actually won.
How should you write the answers?
Answer first, explain second. The opening sentence should resolve the question on its own, because that sentence is what gets quoted. Then add the context a homeowner needs, in plain language, at two to five sentences total.
Be concrete where competitors are vague. Ranges, timelines, and process steps for your area carry the local authority that people-first content guidance rewards. They are also the details assistants prefer to cite.
Where should FAQs live and how are they marked up?
Put questions on the pages they belong to, so the water heater questions live on the water heater page. A single giant FAQ page is fine as a hub, but on-page relevance is what feeds rankings and answers. Keep one FAQPage markup block per URL, listing every visible pair on that page, per Google's structured data documentation.
The iron rule: markup mirrors the page. Every question in the code must be visible to a human reader, word for word. Drift between the two is the most common way FAQ schema quietly rots.
How do you maintain the asset?
Review quarterly. Add the new questions the office started hearing, update any answer with changed prices or timelines, and edit the markup in the same sitting as the visible text. A FAQ that tracks reality for two years becomes one of the most cited assets in your market, because most competitors let theirs go stale in six months.
Frequently asked questions
Should I remove my FAQ schema now that the rich result is gone?
No. Google states unused structured data causes no problems, and the markup still labels your questions for the systems that read pages. Remove it only where the underlying FAQ content was thin bait to begin with.
How many questions should one page carry?
Enough to cover what real customers ask about that topic, which is usually four to eight. Past that point, split by subtopic instead of stacking, so each page stays focused.
Can I reuse the same FAQ across many pages?
No. Duplicated blocks dilute relevance and read as filler. Each page earns its own questions, even when a few overlap naturally between related services.
Xavier writes about structured data and the technical work of being read by machines. FoundLocal is where the full playbook lives.