Schema Markup and AI Search: What Research Shows
The honest answer first: no published research has directly measured whether schema markup increases AI citations. What research does show is that AI systems like ChatGPT, Perplexity, and Google AI Overviews draw from web content in ways that schema may help — but the evidence is indirect.
If you are deciding whether to add schema to your pages for AI search reasons, the answer is: it cannot hurt, it probably helps, and we do not have a number to put on the impact. What we do have is a clear picture of what schema does, how AI systems read web content, and where the two intersect.
For the basics of schema markup itself — what it is, which types matter, how to add it — see the existing schema markup guide. This post focuses specifically on the AI search angle.
Quick Navigation
- What Research Says About Schema and AI Citations
- How AI Systems Read Web Content
- Schema's Indirect Path to AI Citations
- Which Schema Types Matter Most for AI
- What Schema Cannot Do for AI Search
- A Practical Approach for Brass-SEO Users
- Frequently Asked Questions
What Research Says About Schema and AI Citations
No published research has isolated schema markup as a variable affecting AI citation rates — citation research has measured content patterns, not metadata.
The Princeton GEO study (Aggarwal et al., 2024) tested nine content modifications and found expert quotes (+41%) and statistics/citations (+30-40%) drove the largest citation gains. It did not specifically test schema markup. Adam Gnuse's practitioner research from 2025 analyzed 7,500 ChatGPT referral sessions across 15 domains and identified answer capsules (72.4% of cited posts) and original data (52.2%) as defining structural traits. Schema markup was not among the variables measured.
This is not because researchers think schema is irrelevant. It is because schema is hard to isolate as a variable. Pages with schema tend to be technically well-built pages overall — they often have good headings, faster load times, more comprehensive metadata, and clearer content structure. Attributing AI citations to schema specifically would require a controlled experiment that nobody has published.
What we do know comes from indirect evidence: how AI systems source content, what platforms have said about AI Overview sourcing, and how schema is processed by web crawlers including the ones that feed AI training data.
How AI Systems Read Web Content
AI systems read web content in two modes — training and retrieval — and schema markup may influence the second more than the first.
When AI models are trained, they ingest large web crawls and learn patterns from the text. Schema markup is part of the HTML those crawlers retrieve, so schema data is technically present in training data. Whether models meaningfully attend to it during training is undisclosed by OpenAI, Anthropic, and Google.
When AI systems use retrieval-augmented generation (RAG) — Perplexity for every query, ChatGPT in Browse mode, Claude with web search — they fetch live web pages and parse them in real time. This is where schema is most likely to matter. The parsing step has to identify structured information quickly, and schema gives it a machine-readable shortcut.
Google has been the most explicit publicly. Its documentation on AI features in search states that AI Overviews surface content from Google's existing index and that pages with strong technical foundations are favored. That is as close as any major platform has come to a public statement on the role of structured data in AI citations.
Schema's Indirect Path to AI Citations
Schema does not cause AI citations directly — it reinforces three properties the research does correlate with citations: extractability, authority signals, and clear topic identification.
Extractability. AI systems prefer content they can quote without ambiguity. FAQPage schema labels question-answer pairs explicitly. HowTo schema labels steps. Article schema identifies the headline, byline, and date. When an AI system has to decide whether a passage is a complete answer, schema removes guesswork. This aligns with the answer-capsule research, which found 72.4% of ChatGPT-cited blog posts had self-contained, extractable passages.
Authority signals. Article and TechArticle schema include author, datePublished, publisher, and (for academic content) citation fields. These map onto the same E-E-A-T signals Google's quality raters look for and the same authority cues the Princeton study identified as driving the +41% expert-quote effect.
Topic identification. Schema tags content with explicit types (LocalBusiness, MedicalCondition, Product, Recipe). AI systems trying to determine relevance for a query benefit from explicit type information versus inferring it from the page text.
None of these are proven to increase AI citation rates in a controlled experiment. But all three correlate with citation behavior in the practitioner research that does exist.
Which Schema Types Matter Most for AI
Four schema types have the clearest theoretical link to AI citability based on how AI systems extract and quote content.
FAQPage. AI systems frequently cite Q&A content because user prompts are themselves questions. FAQPage schema explicitly labels the question and answer in machine-readable form. The Brass-SEO blog generates FAQPage schema automatically on every post that uses the ## Frequently Asked Questions heading.
TechArticle. Used for technical or instructional content. The schema includes proficiencyLevel, dependencies, and experienceRequired fields that may help AI systems filter content by expertise level. The Brass-SEO GEO guide uses TechArticle schema.
HowTo. AI systems frequently surface step-by-step instructions in responses. HowTo schema labels each step explicitly with HowToStep items and supports totalTime, tool, and supply fields.
Article and BlogPosting. Provide author, publication date, and publisher information. These authority signals matter for AI systems that weigh source credibility, especially for time-sensitive topics.
For local businesses, LocalBusiness schema also matters but for a different reason — AI systems answering local queries (for example, "best plumber in Denver") may surface businesses with strong structured data alongside traditional local SEO signals like Google Business Profile.
What Schema Cannot Do for AI Search
Schema markup cannot rescue thin content, force a citation, or compensate for missing answer capsules.
The Princeton research is explicit that fluency optimization and cosmetic changes had negligible impact on AI citations. Schema markup is closer to a structural signal than a content signal — it describes content but does not substitute for content. A page with FAQPage schema but vague Q&A answers will not be cited just because the schema is there.
Schema also cannot overcome the dominant citability factors identified in the research. The 72.4% answer-capsule rate and the +41% expert-quote effect both reflect content patterns, not metadata. If your page lacks those patterns, schema markup will not make up for them.
There is also a practical limit on how much schema can do for AI. Schema fields are read by parsers, but AI responses are generated from the visible text of the page. The visible text still has to be quotable. Schema helps the AI find the quotable passages — it does not generate them.
A Practical Approach for Brass-SEO Users
For most small business sites, the right schema strategy for AI search is the same as the right strategy for traditional SEO — with one addition.
The same: add schema to pages that have content the schema can accurately describe. LocalBusiness for your homepage. FAQPage on any page with a real Q&A section. Article on blog posts. Product on product pages.
The addition: consider schema as part of an AI citability audit, not just a rich-results audit. Pages that already have answer capsules, original data, and expert citations gain more from schema than pages that lack those elements. Schema markup amplifies an already-citable page.
The Brass-SEO AI Citability button audits pages for ten citability factors. Structured data is one consideration among them, but the higher-impact factors (answer capsules, link-free formatting, original data) come first. If you are short on time, fix those first and add schema to the pages that already pass the structural tests.
For the basics of schema setup, see the existing schema markup guide. For the full GEO framework, see the Generative Engine Optimization guide. For the deep dive on extractable answer formatting, see Answer Capsules: The Content Trait LLMs Cite Most.
Frequently Asked Questions
Does schema markup increase AI citations?
No published research has directly measured this. Schema's effect on AI citations is most likely indirect — it reinforces extractability, authority signals, and topic identification, which the Princeton (2024) and Gnuse (2025) studies correlated with citations. Treat schema as one part of a citability strategy, not a standalone driver.
Which schema type matters most for AI search?
FAQPage schema has the clearest theoretical link to AI citations because AI prompts are usually questions. Q&A content matches the prompt format directly, and FAQPage schema labels the structure explicitly for machine parsing. TechArticle and HowTo also have strong cases for technical and instructional content.
Will adding schema today change my AI referral traffic?
Probably not on its own. Schema's effect on AI citations is amplifying, not generative — it makes already-citable pages more findable for AI parsers. If your pages lack answer capsules, original data, or expert citations, schema will not produce AI traffic by itself.
Can schema markup hurt AI citation rates?
Only if it violates schema.org or Google's guidelines. Marking up content that is not on the page, faking reviews, or using inappropriate types can result in Google ignoring your schema or applying a manual action. Honest, accurate schema cannot reduce AI citations.
Do I need schema markup if I use Brass-SEO?
The Brass-SEO blog adds BlogPosting and FAQPage schema automatically when you write posts with the standard structure (FAQ heading, frontmatter). For your other site pages, schema is added through your CMS or platform — see the existing schema guide for setup steps. Brass-SEO does not automatically add schema to non-blog pages on your customer site.
How does schema relate to llms.txt?
Schema and llms.txt are complementary. Schema is page-level structured data inside HTML. llms.txt is site-level guidance for AI crawlers in a separate file. Schema describes "what is this page about." llms.txt describes "what is this site about and where should AI look first." Both are signals for AI systems, but they answer different questions.