Does Schema Markup Actually Help You Get Cited by AI? What a 2026 Study Found

A large part of the AEO advice circulating in 2025 and early 2026 rested on one assumption: add schema markup, and AI tools will cite you more. It made intuitive sense. Structured data helps machines parse a page, so more structure should mean more citations. In May 2026, Ahrefs ran the test properly, with a control group, and the assumption did not hold up.

The study

Ahrefs analyzed 1,885 web pages that added JSON-LD schema markup between August 2025 and March 2026, matching each one against control pages that never added schema, drawn from a base of 6 million URLs. Using a matched difference-in-differences analysis, a standard method for isolating the effect of one change from everything else happening at the same time, the study measured citation changes across a 30-day window before and after schema was added. The results: Google AI Mode moved 2.2 percent, ChatGPT moved 2.4 percent, both statistically indistinguishable from zero, and Google AI Overviews moved negative 4.6 percent, a small but real decline (Search Engine Journal, on the Ahrefs study, 2026).

Ahrefs' own conclusion was blunt: adding schema produced no major uplift in citations on any platform tested. The finding was independently corroborated in coverage from Search Engine Roundtable, which reported the same directional result across the three platforms (Search Engine Roundtable, 2026).

Why the correlation confused everyone

Here is where it gets interesting. The same dataset showed that pages already being cited by AI were roughly three times more likely to have JSON-LD schema in place than pages that weren't. That correlation is exactly what convinced a lot of marketers that schema was the lever. But a controlled test isolates cause from coincidence, and what the correlation actually reflects is site quality. Businesses that invest in structured data tend to be the same businesses investing in stronger, more organized, more complete content overall. The schema was a symptom of quality, not the source of it (Search Engine Journal).

It's also worth flagging a real limitation in the study: every page in the dataset already had over 100 AI Overview citations before schema was added, meaning these were pages already on AI's radar. The study says nothing definitive about whether schema helps a page that has never been cited at all get noticed for the first time. What it does say clearly is that schema is not the growth lever for a page that's already visible.

This distinction matters practically. If a business has strong, verifiable content that AI systems already find and cite occasionally, adding schema to that content is unlikely to change much. If a business has never been cited at all, the honest answer is that no study has isolated whether schema helps at that starting line, and betting a limited marketing budget on it anyway is a weak use of resources compared to options with actual controlled evidence behind them.

Google says the same thing, in its own words

This lines up exactly with what Google itself published weeks later in its official guide to generative AI search features: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add" (Google Search Central, June 2026). We go deeper on that full guide in our breakdown of how Google's AI Overviews actually decide what to cite. Between Ahrefs' controlled test and Google's own guidance, the technical-markup theory of AI visibility is about as directly contradicted as a marketing theory can be.

What actually moves the needle

If not schema, then what? A separate and much larger body of research answers that question with more consistency. In the paper that introduced the term Generative Engine Optimization, researchers from Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi tested nine content strategies against 10,000 real queries. Adding quotations from relevant sources produced a 41 percent lift in visibility. Adding statistics produced a 31 percent lift. Citing sources directly produced a 28 percent lift. Keyword-based tactics, the closest thing to old-school technical SEO tricks, produced no comparable gain (Aggarwal et al., "GEO: Generative Engine Optimization," arXiv).

Read those three findings together and a pattern appears: a real quote, from a real person, attached to a real number, is close to the exact content shape this research found AI systems reward. A customer testimonial with a name, a title, and a specific outcome checks all three boxes in a single piece of content. No markup required.

What this means if you were about to invest in schema

None of this means schema markup is a waste of time. It still supports rich results in regular Search and helps search engines and AI systems alike understand what a page is about. What it means is that schema should not be the centerpiece of an AI visibility plan, and any vendor pitching it as the primary lever for AI citations is selling a theory the data does not support. The actual lever, backed by controlled testing and Google's own words, is publishing content built on real, specific, attributable evidence, most reliably in the form of customer stories that carry a name, a role, and a checkable result.

Businesses considering where to actually spend time and budget on AI visibility should weigh the two options against what's been tested, not what's easiest to sell. A developer can add JSON-LD schema to a site in an afternoon, which is part of why it got oversold as a quick fix. Collecting and publishing a real customer story takes longer and requires an actual process, which is exactly why the businesses that do it consistently end up standing out from the ones still looking for a shortcut.

That is the entire premise behind the Trust Flywheel: real stories, published consistently, compound into trust that both people and AI systems can verify. You can see the pattern across more than 1,500 real businesses at our case studies page, and read the fuller picture of how AI recommendation actually works in How AI Actually Recommends Businesses.

FAQ

Should a business remove schema markup after reading this?

No. Schema still serves its established purpose for regular Search rich results and general machine readability. The finding is narrower: adding schema does not reliably increase AI citations on its own.

Did the Ahrefs study test every kind of schema?

It tested JSON-LD schema additions broadly, comparing 1,885 pages that added it against a much larger control group. It did not isolate every individual schema type, so it's possible certain narrow types perform differently, but the aggregate result found no meaningful uplift.

Why did AI Overviews actually decline slightly after schema was added?

The study found a small, statistically significant decline of 4.6 percent for Google AI Overviews specifically, while ChatGPT and Google AI Mode showed no meaningful change. The researchers did not identify a clear causal mechanism for the decline, only that it was measurable.

If schema doesn't drive citations, what does?

Research on Generative Engine Optimization found that quotations, statistics, and direct source citations produce meaningful visibility gains. A specific, attributed customer story naturally contains all three.

Is this specific to AI citations, or does it apply to regular SEO too?

This finding is specific to AI citation behavior. Schema still plays an established, useful role in traditional SEO, particularly for rich results eligibility, which is a separate question from AI Overview or chatbot citations.