FAQPage Schema: The Fastest Path Into AI Answers

If you do one structural thing for AI visibility, make it this one.

AI assistants are answering questions. A page that already contains a clean question with a clean answer underneath requires no interpretation to reuse — it is pre-formatted for the exact job the system is doing.

Why this works better than anything else on the page

Consider what an assistant does when a homeowner asks "how much does water heater replacement cost in Tampa."

It retrieves pages. It reads them. It needs a passage that directly answers the question, ideally short, ideally unambiguous, ideally attributable.

A 1,500-word services page about water heaters contains that answer somewhere. Maybe. The system has to find it, decide it is the answer, and extract it cleanly.

A page containing:

How much does water heater replacement cost in Tampa?

Most Tampa water heater replacements run $1,400 to $2,600 installed for a standard 50-gallon gas unit. The spread depends on whether venting and gas line meet current code, whether an expansion tank is required, and access to the install location.

...has done all of that work already. The question matches. The answer is bounded. It is forty-two words.

This is the entire mechanic. Question in, answer out, no interpretation required.

Getting the questions right

The questions must be real. Invented questions written to hold keywords are obvious and produce nothing.

Sources for real ones:

Your phone. Whoever answers it hears the same twenty questions every week. Write them down for a week, verbatim, in the caller's words.

Your quotes and invoices. What do customers ask before approving work? What do they ask after?

Google Business Profile Q&A. People ask there directly.

Search suggestions. Type your service and city into Google and read the autocomplete and the People Also Ask box.

Reddit and forums. Search your trade plus your city. Read what people actually ask each other.

The test for a good FAQ question: would a real customer type this sentence? If it reads like "What are the best plumbing services in Tampa FL?", no. If it reads like "Do I need to replace both water heaters if only one failed?", yes.

Writing answers that get cited

Four rules.

Answer in the first sentence. No preamble, no restating the question, no "Great question!" The first sentence is the answer.

Keep the core answer to 40–60 words. That is the extractable unit. Elaborate after if useful, but the first block should stand alone.

Be specific and use numbers. "It depends on several factors" is not an answer and will never be cited. "$1,400 to $2,600, depending on venting, gas line, and expansion tank requirements" is.

Be willing to say the unprofitable thing. "If your unit is under eight years old and the failure is the thermocouple, repair it — that is a $180 job, not a $2,000 one." Assistants strongly favor sources that qualify their answers honestly, and customers remember it.

The schema

Wrap the visible questions and answers:

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "How much does water heater replacement cost in Tampa?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Most Tampa water heater replacements run $1,400 to $2,600 installed for a standard 50-gallon gas unit. The spread depends on whether venting and gas line meet current code, whether an expansion tank is required, and access to the install location."
      }
    },
    {
      "@type": "Question",
      "name": "How long does a water heater replacement take?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "A straightforward tank swap takes two to four hours. Converting to tankless, relocating the unit, or bringing venting and gas up to code can push it to a full day."
      }
    }
  ]
}

Rules that matter:

• The name must match the visible question text.
• The text must match the visible answer. Do not put a longer or different answer in the schema.
• Plain text in text. Basic HTML is tolerated but plain is safer.
• One FAQPage block per page.

The Google rich result question

Google significantly reduced FAQ rich result display in search results, and a wave of advice followed telling people to remove the markup.

That was the wrong conclusion. The rich result was a side benefit. The structured question-answer pairs are the actual asset, and they became more valuable as AI answers grew, not less.

Keep the markup. The reason to have it changed; the reason to have it did not go away.

Where to put FAQs

Service pages. Five to eight questions specific to that service, at the bottom.

Cost pages. The highest-value placement. Cost questions are the most-asked and least-answered.

Location pages. Questions specific to that area — response time, service radius, permit requirements in that jurisdiction.

Blog posts. Any article naturally ends with questions it did not fully cover.

Not: a single sitewide FAQ page with fifty unrelated questions. That was a 2015 pattern. Distributed, topical FAQs work far better because each one sits on a page about that topic.

Common mistakes

Hidden FAQs. Marking up content inside a collapsed accordion is fine — it is in the DOM and visible when expanded. Marking up content that is not on the page at all is a policy violation.

Answers that do not answer. "Contact us for a free quote" as an answer to a cost question is worse than having no FAQ.

Duplicated blocks sitewide. Dilutes everything.

Schema that disagrees with the page. The two must match exactly.

Too many questions. Thirty questions signals padding. Ten real ones signal knowledge.

What to do this week

1. Spend one week writing down every question a customer asks on the phone, verbatim.
2. Pick your five most important pages.
3. Add five to eight relevant questions to each, answer-first, 40–60 words.
4. Add FAQPage schema matching the visible text exactly.
5. Validate with Google's Rich Results Test.

Then check the rest

FAQ structure is one of fifteen on-page signals we check. Our free scan runs all of them alongside twenty-five real customer questions across five AI assistants and shows you exactly what is missing.

About fifteen seconds, free, no account.