How to Write FAQ Answers That AI Search Engines Actually Cite

Most service businesses have a FAQ section. Most of those FAQ sections don't get cited by AI.

The failure isn't usually missing schema. It's the content itself -- answers written to reassure humans who are already considering hiring you, not structured as extractable citations. In our June 18, 2026 Phase 2 FAQ content specification, we traced three converging research findings that define the difference. The requirements are specific. Most businesses aren't meeting them.

Why Google's FAQ schema change doesn't mean FAQs matter less

In May 2026, Google removed FAQ rich results from search. The expanded Q&A dropdowns that used to appear inline in traditional results are gone. Several agencies have told clients to deprioritize FAQ schema because "the rich snippet benefit disappeared."

That's the wrong conclusion.

FAQPage schema in JSON-LD still does two things Google's display removal didn't touch. First, it strengthens your Knowledge Graph entry, which Gemini draws from when answering local business queries. Second, it marks your Q&A pairs as machine-readable structured content, which ChatGPT, Perplexity, and Claude use when extracting answers from your pages. Those platforms don't care that Google stopped showing rich snippets -- they're reading your HTML and finding labeled Q&A blocks they can extract cleanly.

The framing has to update. The reason to add FAQPage schema is no longer "get expanded SERP snippets." It's "make your answers extractable by AI retrieval systems."

The first sentence rule

This is the most common FAQ mistake and the most costly.

Our June 18, 2026 content specification identified what AI retrieval systems actually do with FAQ answers: they extract the first sentence as a candidate citation. If your first sentence contains the answer, the extraction works. If your first sentence is preamble -- a capabilities claim, context without the answer, or a "we'd be happy to help" construction -- the extraction fails. The page doesn't get cited for that question.

This plays out clearly when you compare a thin answer against an evidence-dense one, using an HVAC example from our June 18 research:

**Not cited:** "Our experienced technicians can assess your AC unit and recommend whether repair or replacement is right for you."

**Gets cited:** "A 12-year-old AC is typically near the end of its 15-20 year lifespan. If the repair cost exceeds half the unit's replacement value, replacement usually makes more economic sense. Call us for a free assessment -- we'll give you the repair vs. replace breakdown in writing."

The second answer opens with a specific, measurable claim that stands alone as a citation. The first returns a capabilities statement that could describe any HVAC company anywhere. AI retrieval systems don't find capabilities statements useful. Specific answers are useful.

Buyer incident language vs. service category language

Scout's June 17, 2026 analysis of page-level citation signals documented a finding from Discovered Labs 2026 that changes how you should write FAQ questions, not just answers.

Vocabulary alignment -- how closely your page language matches the exact words buyers use in their queries -- is the only page-level signal that survives domain fixed-effects controls in the Discovered Labs study, with a standardized coefficient of β=+0.37. That statistical framing matters: this effect doesn't disappear when you control for domain authority. It holds within a single domain, across different pages. Two service pages from the same site, identical in every other way except question vocabulary, produce different AI citation rates.

The definition is precise: buyer incident language, not service category language.

From our June 18, 2026 content specification:

| Service jargon (low alignment) | Buyer incident language (high alignment) | |---|---| | "What plumbing services do you offer?" | "How quickly can a plumber arrive for a burst pipe?" | | "What are your HVAC maintenance packages?" | "Is it worth repairing a 12-year-old air conditioner?" | | "What dental services do you provide?" | "Can I get a same-day appointment for tooth pain?" |

The left column describes what you offer. The right column describes the situation a buyer is in when they're searching. AI search engines retrieve on query language -- the right column language. A contractor whose FAQ asks and answers buyer-incident questions is capturing vocabulary alignment that a contractor with service-listing questions is missing entirely.

To find the right questions for your business: look at what queries an AI visibility audit ran against your category. The B and C query set -- category and service queries -- are the phrases buyers actually type into ChatGPT and Perplexity. If your FAQ questions don't use that language, the alignment gap is real.

The 40-80 word answer window

The GEO-SFE structural study (arXiv:2603.29979, Yu et al., March 2026), which measured citation lift across six AI platforms, found a 17.3% citation improvement from structural changes alone. The strongest effect came from what the researchers called meso-level chunking: one claim per paragraph, data in tables where applicable, and self-contained passages that can be extracted without surrounding context.

FAQ answers map directly onto this. An answer under 40 words doesn't have enough specificity to stand as a citation. An answer over 80 words gets truncated in retrieval, cutting off the specific details that gave the answer its value.

The working recommendation from our June 18 specification: four to eight questions per service page. Six is the working default. Each answer should be 40-80 words, with the complete answer in sentence one and one to two sentences of measurable, specific context after it. Timeframes, service areas, price thresholds, and named specifics are what make an answer evidence-dense enough to be cited.

What this means for independent contractors

The FAQ content requirements matter most for businesses that can't earn citations through press coverage or brand scale -- which is most independent contractors.

Our June 2026 knowledge file on home services AI citation data (updated June 15, 2026) tracks findings from the 5W HVAC and Plumbing AI Visibility Index 2026, which analyzed 65+ consumer-intent prompts across ChatGPT, Claude, Perplexity, and Google AI Overviews. The index found 87% of independent HVAC and plumbing contractors have effectively zero AI citation share in their own metro -- including businesses with 800+ five-star Google reviews. The same pattern extends to roofing, electrical, and other trades.

National chains win AI citations through IPO press coverage and trade media -- citation fuel independent contractors can't replicate. A plumber who answers "how fast can a plumber arrive for a burst pipe?" with a specific, 60-word evidence-dense answer using buyer language is doing the one thing the chains can't fake: accurate local specificity.

There are 145 million monthly plumbing queries in the US, and 55% of them are proximity-based. Those are buyers actively in need. The business whose FAQ directly addresses the incident is structurally positioned for that query. The business with a thin services list is not.

What to check first

Read your FAQ answers with the first sentence rule in mind. If you extracted only sentence one of each answer, would someone reading it know the answer to the question? If not, it fails the citation extraction test.

Then audit your question language. Compare your FAQ questions against how a homeowner, patient, or first-time customer would describe the situation they're in -- not how you would describe your service. Rewrite questions that describe your offering into questions that describe the buyer's situation.

Signal Check at sourcepull.ca runs live queries in your category across ChatGPT, Perplexity, Gemini, and Claude. If you have FAQ content and still aren't appearing in citations, the per-page breakdown shows whether the gap is in question language, answer density, schema, or something upstream in your entity infrastructure.