An FAQ built for AI retrieval is not an FAQ built for humans. The difference is structure. A prose answer makes an engine work—it has to parse, infer, assemble. A marked-up question-and-answer pair gets cited as-is. This is why FAQPage schema is the citation cheat code almost nobody is using.
ChatGPT, Perplexity, and Google AI Overviews all prefer retrieved facts over synthesized ones. If your answer is already formed, quotable, and factually isolated, the engine pulls it whole. It does not rewrite. It does not blend. It cites you by name and domain.
Why Prose Loses to Structure
A paragraph answering "How much does a crown cost?" looks like this:
"The cost of a dental crown varies depending on the material used and the complexity of the tooth preparation. Porcelain crowns typically range from $800 to $1,500, while zirconia crowns may cost between $1,000 and $2,000. Gold crowns are more expensive, usually running $2,000 to $3,000, but they last the longest. Your insurance may cover part of the cost."
An engine reading this has to: 1. Extract the price ranges (three different ones). 2. Decide which one to cite. 3. Infer which answer is most relevant to the query. 4. Attribute it or hedge it.
It might cite you. It might blend your answer with three competitors' answers. It might rewrite.
The same answer, marked as FAQPage, looks like this:
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "How much does a dental crown cost?",
"acceptedAnswer": {
"@type": "Answer",
"text": "A porcelain crown costs $800 to $1,500. A zirconia crown costs $1,000 to $2,000. A gold crown costs $2,000 to $3,000 and lasts the longest. Insurance may cover part of the cost."
}
}
]
}
Now the engine has: 1. A named question. 2. A pre-formed, isolated answer. 3. No ambiguity about what to cite. 4. Permission to pull the text directly.
It cites you. It does not rewrite.
One Answer Per Question
The most common mistake is nesting multiple questions into one answer block. "What does a crown cost? How long does it last? Will insurance cover it?"
Do not do this. One question. One answer. One fact per block.
If your actual answer is compound, break it into separate questions:
- "How much does a crown cost?"
- "How long does a crown last?"
- "Does insurance cover a crown?"
Each gets its own FAQPage entry. Each can be cited independently. An engine retrieves the one that matches the prompt.
Answer Length and Specificity
Keep answers between one and four sentences. A sentence here means a complete statement, not a parenthetical aside or a clause.
"A porcelain crown costs $800 to $1,500." — one sentence. Cite-able.
"A porcelain crown, which is made of ceramic material bonded to a metal or zirconia base (depending on whether you choose high-noble gold or a tooth-colored alternative), typically costs between $800 and $1,500, though some practices charge more in high-cost-of-living areas like Brooklyn or Manhattan." — one sentence. Too many qualifiers. An engine will strip it or hedge it.
The rule: if you would italicize a parenthetical, delete it.
How to Structure the Question
Write the question as the user would ask it. Not "What is the cost of a porcelain dental crown?"—that is a label. Write "How much does a porcelain crown cost?"
Use the phrasing that appears in actual search behavior. If most users say "How do I schedule an appointment?" not "How do I make a booking?", use the first one. The question field is where retrieval begins.
Questions should also be specific. "What services do you offer?" is weak. "Do you offer same-day crowns?" is strong. Specificity triggers retrieval on specific prompts.
The Citation Layer
Once you have marked-up FAQ sections, your content becomes retrievable at sentence level, not page level. That changes where the citation points.
A prose answer gets cited as a page or a domain. A marked-up answer gets cited as a quotation with your business name attached.
Brooklyn BJJ Lessons built its entire early citation strategy around Person and LocalBusiness schema, and that resulted in a ChatGPT citation within 41 days. The same principle applies to FAQ. Marked structure is cheaper to cite than unmarked prose.
Testing Your FAQ Markup
Use Google's Rich Results Test to validate your markup. Paste your page URL. If your FAQPage markup appears in the results, it is well-formed and retrievable. If it does not, fix the JSON before publishing.
The test does not guarantee retrieval—it only confirms that the markup is valid. But invalid markup guarantees you will not be retrieved. Start there.
Nesting FAQ Into Your Site Architecture
Do not create a separate FAQ page. Nest FAQPage blocks into the pages where the questions naturally belong.
A dentist's crown page should have FAQ sections about cost, longevity, materials, aftercare. Those blocks live on the page itself, not on a separate /faq route.
Why? Because the page is already the authority for that topic. Adding FAQ markup signals to the engine that specific facts on that page are retrievable. It does not dilute the page's topical weight.
If you have a dedicated FAQ page, that's fine—add FAQPage markup there too. But do not treat it as your only FAQ layer.
The Maintenance Rule
FAQ markup decays if you do not update the answers. If you change your crown pricing, update the FAQ answer the same day. If you stop offering a service, delete that question entirely.
Engines notice inconsistency. A marked answer that contradicts your GBP hours or your homepage text looks like bad maintenance. You lose trust.
Treat FAQ markup as a living data structure, not a static content block.
What This Means for Your Business
FAQPage markup is a direct signal to an engine that your answers are quotable facts, not marketing prose. It is the fastest path to being cited by name in generative engine optimization results.
The investment is small. Most businesses can audit and mark up their FAQ sections in a few hours. The payoff is disproportionate: you stop competing on page ranking and start competing on citation.
Start with the five questions you hear most often from customers. Mark them. Test them. Watch where they get cited. Then build from there.