Your FAQ schema is technically valid. Google Search Console shows no errors. A featured snippet fired once in 2023. And every AI engine that matters right now, Claude, Perplexity, ChatGPT Search, Gemini, is citing your competitor instead of you.
This is the FAQ schema gap. It isn't a bug you can patch with a plugin update. It's a structural mismatch between what you built the markup to do and what AI retrieval systems actually reward.
The rules changed. Most service businesses are still playing the 2021 game.
Two retrieval models, one markup
Google's featured snippet algorithm is essentially a keyword-match proximity test. Your FAQ answer wins when it contains the query terms and sits close to a matching question string. Concise beats comprehensive. Google wants the shortest accurate answer it can display in the SERP without making the user click.
AI engines work the opposite way. Claude doesn't scan for keyword proximity. Perplexity doesn't award points for brevity. These systems run a semantic completeness check. They ask: does this source contain enough connected, attributable information to synthesize a trustworthy answer? Thin wins on Google. Thin loses everywhere else.
The comparison that fits here: optimizing FAQ schema for featured snippets and expecting AI citation is like formatting a Netflix pitch deck for a Blockbuster acquisition meeting. Same industry. Structurally incompatible audiences.
What your current markup actually says
Pull any FAQ page from a mid-market service business and the pattern is predictable. Question: "How much does [service] cost?" Answer: "The cost of [service] varies depending on your needs. Contact us for a custom quote." Perfectly useless. No number. No range. No methodology. No entity.
That answer is schema-wrapped nothing. It tells an AI engine that your page exists but gives the model zero citable substance. Perplexity won't cite it. ChatGPT won't surface it. It's the equivalent of a Yelp listing with no reviews, technically present, functionally invisible.
The second pattern is keyword stuffing dressed as an answer. "HVAC repair in Denver costs between $150 and $450 for most Denver HVAC repair jobs, and Denver HVAC companies typically charge. ." The keyword density signals fire. The factual density signals don't. AI models are trained on authoritative text. That answer reads like SEO copy, not expertise. It gets skipped.
The third pattern is actually the sneakiest: technically solid FAQ schema wrapped around a thin service page. The markup validates. The surrounding page context doesn't support it. AI engines don't evaluate the schema block in isolation. They evaluate the document. A credible FAQ block inside a shallow page is a good frame around a bad painting.
What AI engines actually cite
The AEO content research points to a consistent signal: specificity is the deciding variable. Not schema syntax. Not keyword density. Specific numbers, named processes, geographic context, and attributable claims are what generative models pull when they build an answer.
Compare these two FAQ answers to the question "How long does commercial roof replacement take?":
- Version A (built for Google): "Commercial roof replacement timelines vary depending on the size of the roof, weather conditions, and material availability. Most projects are completed within a few weeks."
- Version B (built for AI citation): "A typical commercial flat roof replacement on a 15,000 to 30,000 sq ft building runs 8 to 14 working days with a crew of six. TPO membrane installs faster than modified bitumen because it eliminates the torch-down step. Weather delays in high-humidity climates like Houston or Miami add 2 to 4 days on average. We track this on every project in our job management system."
Version A might win a featured snippet. Version B is what Claude quotes when a facilities manager asks it to help evaluate a roofing contractor. The schema markup on both pages is identical. The content is not.
“AI engines don't evaluate your schema block. They evaluate your document. Good markup around thin content is a good frame around a bad painting.”
This aligns with what Search Engine Journal reported this week: AI SEO that prioritizes specific content may be the primary driver of citation frequency. It isn't a formatting trick. It's a content depth requirement that most FAQ sections were never written to meet.
Schema markup has a second job in AI search that almost nobody is using it for. Beyond marking up the question-answer pair, structured data signals entity identity. Who are you? What do you do? Where? For whom?
AI engines build knowledge graphs. When Perplexity composes an answer about commercial roofing contractors in Denver, it's drawing from entities it has confidence in. If your Organization schema is missing, if your LocalBusiness markup doesn't include `areaServed`, `serviceType`, and a consistent NAP pattern, you don't have an entity. You have a page. Pages get indexed. Entities get cited.
Most FAQ implementations on service business sites look like this: a FAQ schema block dropped by a plugin onto a page that has no Organization markup, no author entity, no ServiceArea definition. The FAQ floats. There's nothing for an AI engine to anchor it to. Schema markup as citation currency only works when the full entity picture is present, not just the Q&A block.
The fix isn't complicated. It's layered. Your site needs a consistent entity spine: Organization on the homepage, LocalBusiness or MedicalBusiness or LegalService on the right pages, Person schema on author and team pages, ServiceArea and hasOfferCatalog where appropriate. The FAQ markup connects to that spine. Without it, the FAQ block is a widget, not a citation signal.
The rewrite sequence that actually works
The instinct is to fix the schema first. Wrong order. Schema is a wrapper. Rewrite the content first, then confirm the markup wraps the better content. Wrapping the same thin answers in cleaner JSON-LD is the self-checkout at CVS. You moved the labor around. Nothing actually changed.
Here's the sequence we run on SEO and GEO engagements when auditing a client's FAQ architecture:
- 01Audit for specificity gaps. Flag every FAQ answer that contains no number, no named process, no geographic reference, no attributable claim. These are the invisible answers.
- 02Identify the five questions your best clients actually ask during the sales process, not the questions that match head keywords. Rewrite those five with full factual density.
- 03Check entity completeness. Confirm Organization, LocalBusiness, and relevant service-type schema is present on the pages hosting the FAQ blocks. If the entity spine is missing, build it first.
- 04Add attribution signals inside the answers where possible. Named methodology, staff credentials, tool names, specific project data. These are the claims AI models pull when composing a cited answer.
- 05Validate with a live AI test, not just a schema validator. Run your rewritten questions through ChatGPT Search, Perplexity, and Gemini. Check if your page appears in citations. If it doesn't, the content still isn't dense enough.
The live AI test is the step most teams skip. Schema validators tell you the markup is syntactically correct. They tell you nothing about whether an AI engine finds the content worth citing. Syntax correctness and citation worthiness are two completely different outcomes.
One more thing on the rewrite: Google explicitly warned this week against building separate markdown pages optimized for AI. That's the wrong direction. The goal is one authoritative page that satisfies human readers, earns Google's confidence, and gives AI engines enough factual density to cite. Not three versions of the same page in different formats.
Where the gap is actually widening
McKinsey's 2026 data has more than 50% of consumers using AI-powered search. The marketing director at a 12-location dental group isn't waiting for the industry to figure this out. She's already watching Perplexity cite her competitor's FAQs on questions about implant costs and recovery timelines while her technically compliant schema earns nothing.
The gap compounds. AI engines learn citation patterns. Pages that earn citations get more citations. Pages that don't get skipped again. The brands that fix their FAQ content architecture now, not their schema syntax, are building a citation compounding advantage that gets harder to close every month.
This is the same dynamic we laid out in why AI engines cite your competitors while Google ignores you. The ranking game and the citation game have different mechanics. You can win one and lose the other simultaneously. Most service businesses are doing exactly that.
The FAQ schema gap isn't a technical debt problem. It's a content strategy problem with a structured data wrapper. Fix the content. Add the entity spine. Test against live AI engines. Then your schema does what it was always supposed to do: tell AI engines, with confidence, that your answers are the ones worth quoting.