Last updated August 2026
The question AEO practitioners keep asking
Structured data is supposed to help search engines understand your content. The promise for answer engine optimization extends that logic: if you label your FAQs, your how-to steps, and your product reviews correctly, AI engines will surface your content as the cited answer.
That premise is intuitive. And the correlation data looks encouraging. But a rigorous controlled study published in 2026 tells a more complicated story, one that every serious AEO practitioner needs to understand before they put schema at the center of their citation strategy.
This piece lays out what the evidence actually shows: the correlation, the contradiction, and where schema fits in a realistic optimization framework.
What the correlation data shows
The AirOps team analyzed more than 12,000 URLs across 900 ChatGPT queries in 15 industries. The structural patterns between cited pages and top-ranked Google pages were striking.
| Signal | Pages cited by ChatGPT | Top-ranked Google pages |
|---|---|---|
| 3 or more distinct schema types | 61% | 25% |
| Sequential heading structure (H1 to H3) | 68.7% | 23.9% |
| At least one structured list | ~80% | ~29% |
| Comparison tables (3 or more HTML tables on comparison pages) | Higher citation rate (+25.7%) | Baseline |
(Source: AirOps, “Structuring Content for LLMs,” July 2025. Single-vendor study, ChatGPT citations only, not independently replicated.)
The gap is large. Pages cited by ChatGPT are more than twice as likely to use rich schema and sequential headings as the pages Google ranks highest for the same queries. Nearly four out of five cited pages include at least one structured list, compared to fewer than one in three top-ranked Google pages.
This is the data that has driven the widespread AEO recommendation to “add schema to win citations.” It is not wrong. But it is incomplete.
The contradiction: the Ahrefs controlled study
In May 2026, Ahrefs published a study that tracked 1,885 pages before and after they added JSON-LD schema. The study ran from August 2025 to March 2026, with 4,000 control pages that made no schema changes.
The result: adding schema produced no meaningful uplift in AI citations across any platform tested.
| Platform | Change in citations after adding schema |
|---|---|
| Google AI Mode | +2.4% (within margin of error) |
| ChatGPT | +2.2% (within margin of error) |
| Google AI Overviews | -4.6% (slight decline) |
(Source: Ahrefs, “We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved.” May 2026. Important caveat: all pages in this study already had 100 or more AI Overview citations before treatment, meaning the results may not apply to pages starting from zero citations.)
The Ahrefs team concluded that schema does not causally drive AI citation visibility for pages already receiving substantial citations. The +2.4% and +2.2% changes for Google AI Mode and ChatGPT were statistically indistinguishable from zero. The -4.6% for Google AI Overviews was small but negative.
This does not mean schema is useless. It means schema is not a reliable citation lever for pages that already have baseline authority and citation activity.
How to read the evidence honestly
Two things can both be true:
- Pages with rich schema are more likely to be cited.
- Adding schema to your page does not reliably cause more citations.
This apparent paradox resolves when you think about what cited pages have in common beyond schema. They tend to be authoritative sources on their topics, with clear heading structure, direct answers to specific questions, and content written to match how people ask questions, not just how they search for keywords.
Schema use may be a marker of overall content hygiene, not an independent variable. Teams that build pages with correct FAQPage markup, Article schema, and Product/Review schema also tend to be teams that structure their content carefully. The schema and the citation may both be effects of the same underlying content quality, not a direct cause-and-effect relationship.
A UC Berkeley arXiv preprint (Kumar and Palkhouski, 2025) found structured data to be the third-strongest predictor of AI citation likelihood in their analysis of 1,100 URLs across 1,702 citations from Brave, Google AI Overviews, and Perplexity, with an associated +39% lift. But even that finding is an association, and the study audited a narrower dataset than the Ahrefs controlled experiment.
The Princeton, Georgia Tech, IIT Delhi, and Allen Institute for AI team, in a peer-reviewed study at ACM KDD 2024, found that adding statistics to content improved AI visibility scores by roughly 40% on their Position-Adjusted Word Count metric. That is not schema. That is content substance.
The schema types most associated with cited pages
Despite the causation uncertainty, schema still belongs in your AEO workflow. Pages without it are at a structural disadvantage: engines cannot parse what is a question, what is an answer, or what is a review if that structure is not declared.
The types most consistently associated with AI-cited pages:
- FAQPage schema: The most direct signal for question-answering engines. Label your FAQ section with FAQPage and individual question-answer pairs with
QuestionandAnswerproperties. This is the schema type most clearly aligned with how AI engines retrieve and quote content. - Article schema: Helps engines identify author, publisher, publish date, and content type. Important for establishing freshness and authority signals.
- HowTo schema: Structured steps with clear names and descriptions. Works best when your content genuinely walks through a process. Do not use it on pages that are not actually instructional.
- Product and Review schema: For e-commerce and comparison content, Product schema with Review and AggregateRating properties is consistently associated with higher citation rates on commercial queries.
The AirOps data suggests that using three or more of these types together matters. 61% of ChatGPT-cited pages used three or more distinct schema types, versus 25% of top-ranking Google pages. Single-type implementations may not cross the threshold that AI parsers use when deciding whether a page is a well-structured source.
Where schema fits in your AEO workflow
Schema is a foundation, not a lever. Here is how to position it in your broader optimization effort:
Step 1: Audit for missing or broken schema
Pages with no schema at all are at a baseline disadvantage. Start with an audit. Tools like Surfer SEO include content structure analysis that flags missing schema types. Semrush’s Site Audit also identifies schema errors and missing markup. Writesonic generates FAQ sections with FAQPage schema built in as part of its content creation workflow.
Step 2: Implement the schema types that match your content format
Match schema type to content format. FAQ sections get FAQPage. Process pages get HowTo. Product comparisons get Product plus Review. Articles get Article. Do not add schema types that misrepresent what is on the page: engines notice the mismatch.
Step 3: Focus optimization effort on the signals with stronger evidence
The evidence for content clarity, heading structure, and direct-answer formatting is more consistent than the evidence for schema as a citation driver. After schema is in place:
- Front-load your direct answer in the first 40 to 60 words of each section.
- Use sequential heading structure (H1 to H2 to H3) throughout.
- Include at least one structured list on every major page.
- Add statistics with named attribution. The Princeton and Georgia Tech KDD 2024 research found that adding statistics improved AI visibility scores by roughly 40%, a stronger signal than schema alone in that study.
Step 4: Monitor citation rates, not just schema coverage
Schema implementation tells you nothing about whether your pages are being cited. Use a dedicated tracking tool to measure actual citation rates across the engines you care about. Temso tracks citation visibility across eight AI engines (ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode, Grok, Microsoft Copilot, and Meta AI) starting at $89/mo, and shows you which pages are cited and which are not. Profound and otto-seo offer deeper citation attribution for enterprise teams that need per-prompt granularity. Without this data, you are optimizing blind.
The honest AEO practitioner’s summary
Schema markup is a necessary but not sufficient condition for AI citation. The data from AirOps shows that cited pages use richer schema than top-ranked Google pages. The data from Ahrefs shows that adding schema does not reliably produce more citations.
Both findings are credible. They point toward the same conclusion: schema is hygiene. You need it. It will not carry you on its own.
The stronger evidence points to content structure, directness, and substance. Pages that answer the question clearly in the first paragraph, use proper heading hierarchy, include structured lists, and cite statistics are more likely to be cited than pages that are technically well-marked up but structurally vague.
Get your schema right. Then invest most of your effort in the content itself.
What to read next
- Full AEO tool ranking: /rankings/aeo-tools
- How to measure AI answer citations: /blog/how-to-measure-ai-answer-citations
- The AEO schema checklist: /blog/aeo-schema-checklist-9-json-ld-types-add
- Glossary: /glossary
- Methodology: /methodology
Track which pages are actually being cited. Schema optimization without citation measurement is guesswork. Temso tracks your citation rates across eight AI engines for $89/mo, and shows you exactly which pages are appearing in answers and which are not. See the full AEO tool rankings to compare your options.