Does schema markup get you cited?

"Add structured data so the AI understands your page." It sounds technical, it sounds correct, and it is the kind of task an agency can bill for. When someone actually ran the experiment, the effect landed on zero.

Schema markup — usually JSON-LD, a block of code in the page's source that labels what things are: a product, a price, an author, a business address. Search engines have used it for years to build rich results, the star ratings and FAQ dropdowns you see in Google. The GEO pitch extends that logic: label your content clearly and the language models will understand it better, so they will cite you more. The logic is clean. The data does not cooperate.

The strongest-looking evidence for schema is a correlation. Ahrefs looked at six million URLs and found that pages cited by AI were almost three times more likely to carry JSON-LD than pages that weren't; across cited pages, 53% had schema (Ahrefs). That is the stat that ends up on conference slides. It is also exactly the kind of number that tells you nothing about cause.

Why the correlation is a trap

Schema tends to live on well-run sites. The same teams that add structured data also write good content, earn links, fix their technical SEO and maintain their pages — all the things that actually get a page cited. So of course cited pages have more schema. The question a small business actually needs answered is different: if I add schema to my page, will I get cited more? A correlation cannot answer that. An experiment can.

So Ahrefs ran one. They found 1,885 pages that added JSON-LD between August 2025 and March 2026, matched each against three control pages from other domains with similar citation levels beforehand — about 4,000 controls in total — and measured citations for 30 days before and after. Matching against controls strips out the platform-wide trend, so what's left is closer to the effect of the schema itself.

Ahrefs checked the result four different ways, including the naive before-and-after that vendors tend to quote. Run that way, AI Mode citations looked like +43% — impressive, until you notice the control pages that added no schema gained almost as much. AI Mode was simply booming for everyone that period. Strip the trend out and the +43% collapses to +2.4%. That gap between the two numbers is the whole game: it is the difference between measuring your work and measuring the weather.

There is a reason the effect is so flat. A separate experiment tested whether five major systems — ChatGPT, Claude, Perplexity, Gemini and Google AI Mode — actually read schema when fetching a page live. None of them did. During direct retrieval, every system pulled only the visible HTML; the hidden JSON-LD, Microdata and RDFa were ignored (searchVIU, cited in the Ahrefs study). The models read what a human reads. The label in the code is for a different audience.

The honest caveat

One thing the study cannot tell you, and it matters. Every page it measured was already heavily cited — 100+ AI Overview citations before any schema was added. These pages were already inside the machine's field of view. What the study shows is narrow and firm: for a page the AI already sees, adding schema will not push it higher. It does not prove schema is useless for a page the AI has never discovered, where structured data might still help a crawler parse and index it in the first place. That question is open.

If you have five employees

Do the schema once, through your website plugin or a free generator, for the Google rich-result benefit. Then stop thinking about it. Do not let it sit near the top of a GEO to-do list, and do not pay a monthly retainer that is mostly "structured data optimization" — you are paying for a wrapper around something that runs itself.

The thing that actually moves citations is duller and harder to bill: content on the visible page that answers a real customer question in plain words, on a page a crawler can reach. The models cite what they can read and lift as a clean answer. Schema is a note in the margin they mostly skip. Your job is to make the main text worth quoting.

As always, the test is simple. Ask whoever recommends a task to show you the controlled result — the version where the thing they changed was measured against the thing they didn't. Schema has that result now, and it says: fine to do, not the reason you'll get cited.