FAQ schema appears in 1.8% of the pages AI systems actually cite. It features in nearly every SEO best-practice checklist. Those two facts point in different directions, and the data from 9,000 AI citation sources tells you which one to follow.
What the data comes from
AccuraCast published a study in September 2025 analyzing 9,000 AI citation sources drawn from 2,000 prompts across ChatGPT, Google AI Overviews, and Perplexity. The study mapped which structured data types appeared on cited pages versus a control set of non-cited pages.
The finding: 81% of cited pages included some form of schema markup. But the distribution across schema types was not what most SEO advice would predict.
The FAQ schema problem
FAQ schema is one of the most widely recommended structured data types. The logic makes intuitive sense: if your content answers questions, mark it up as questions and answers so machines can read it easily. AI systems are question-answering machines. The recommendation feels obvious.
The AccuraCast data does not support it. FAQPage schema appeared in 1.8% of the 9,000 cited sources. That is not a rounding error. It is a signal that the recommendation and the outcome are disconnected.
Two reasons explain this gap. First, Google has officially restricted FAQ rich results to government and health sites. For commercial content, FAQPage schema produces no rich result in Google Search, which means it generates no additional signal in the systems that feed Google AI Overviews. Second, AI crawlers appear to rely primarily on visible content rather than metadata. A Writesonic study of 62 webpage elements across 6 AI crawlers found that JSON-LD scored zero out of six for readability across all crawlers tested. The markup is there; the crawlers are not reading it.
Person schema is the dominant signal
The schema type that actually appears in cited pages is Person schema, which identifies the author of a piece of content: their name, credentials, and a URL that uniquely identifies them as an entity.
58.9% of all AI-cited pages in the AccuraCast dataset include Person schema. For ChatGPT specifically, that number rises to 70.4%.
This correlates with a separate finding from Presence AI, which tracked 1,200 pages and 3,600 queries across four AI platforms over 90 days. Pages with expert authors and documented credentials achieved a 72% AI citation rate. Pages with no author attribution achieved 25%. That is a 2.4x difference driven entirely by author attribution.
Person schema is not causing the citation. It is reflecting the underlying fact that AI systems favor content produced by identifiable, credentialed humans. The schema makes that fact machine-readable and consistent across platforms.
| Schema type | Share of AI-cited pages | ChatGPT share |
|---|---|---|
| Person | 58.9% | 70.4% |
| Article | Correlated with citation (exact % not published) | n/a |
| FAQ Page | 1.8% | n/a |
Source: AccuraCast, 9,000 citation sources, Sep 2025
What JSON-LD actually does for AI crawlers
Google recommends JSON-LD as the preferred schema implementation format. Traditional search crawlers read it reliably. But AI crawlers work differently.
Writesonic tested 62 webpage elements across 6 major AI crawlers in March 2026. JSON-LD scored zero out of six. Meta descriptions scored zero. Open Graph tags scored zero. The title tag was the only metadata element with strong cross-crawler readability at five out of six.
The practical implication: JSON-LD schema may still benefit your visibility in traditional Google Search and in Google AI Overviews, which builds on Google Search infrastructure. But for OpenAI, Anthropic, and Perplexity crawlers, visible content is the primary input. Schema is not the channel those crawlers use.
What to implement instead
The evidence points toward two priorities.
First, add Person schema to every published page. Google officially recommends includingauthor.url in Article structured data as a URL that uniquely identifies the author. This creates a persistent entity signal connecting your content to a real, credentialed person. It is the schema type with the strongest observed correlation to AI citation.
Second, make the author visible in the content itself, not just in metadata. A named byline, a short credentials statement, and a consistent author page all contribute to the underlying signal that schema is trying to communicate. When AI crawlers cannot read the JSON-LD, they can still read the paragraph that says who wrote the piece and why they are qualified.
FAQ schema is not worth prioritizing unless your site qualifies for Google FAQ rich results (government or health). Article schema is worth including because it provides structural context that benefits traditional crawl parsing. But if you are choosing where to invest time, Person schema and visible author credentials produce the observable outcome.
The bottom line
81% of AI-cited pages use structured data. But schema markup is not driving citations; author credibility is. Person schema appears in 58.9% of cited pages because cited pages tend to be written by identified, credentialed authors. The schema makes that signal consistent. FAQ schema appears in 1.8% of cited pages because it does not reflect anything AI systems are weighting in their retrieval decisions.
Optimize the thing that is actually correlated: the person behind the content, made visible both in structured data and in the content itself.