Does Schema Help AI Citations? The Honest Answer

1Does schema help AI citations? The short answer

On the best evidence available today, adding schema does not meaningfully increase how often AI answer engines cite you. The lift in the controlled tests is small and does not point consistently in one direction, which is what you would expect from noise rather than a real effect. That is the finding. Everything else in this article is the nuance around it, because the finding is easy to over-read in either direction.

The wrong takeaway is "schema is a scam, rip it out." The right takeaway is that structured data is machine-readability hygiene you should get right and then stop obsessing over, while the work that actually moves citations happens at the level of what your content says and where it is said.

2What did the controlled test actually find?

The strongest data point comes from Ahrefs, which ran a difference-in-differences analysis on pages that added structured data and compared them against pages that did not, measuring the change in AI citations across three surfaces. Difference-in-differences is the useful part of the method here: instead of just checking whether pages with schema get cited more (they might, for reasons that have nothing to do with the schema), it isolates the change caused by adding schema by netting out the trend everything else was on anyway.

Source: Ahrefs, "Does schema markup affect AI citations?"

Read those numbers together, not one at a time. One surface went slightly down, two went slightly up, and none of the moves is large. If schema were a real citation lever you would expect a consistent, sizeable lift across engines; instead you get a wash. Ahrefs' own conclusion is that adding schema had essentially no effect on how often the pages were cited. That is a single study, from one dataset, and it deserves the usual caveats. But it is a genuinely controlled test, and it points the opposite way from the checklist consensus.

3The one-variable test that says the same thing

Large studies are one kind of evidence; a clean single-variable experiment is another, and the two lining up is more convincing than either alone. A practitioner running a week-by-week citation test documented exactly that. They held everything constant, changed one thing, and watched the result:

Zero citations before the schema, zero after. This is a tiny sample and the author would be the first to say so, but the shape of the result matches the large-scale finding: isolate schema, change nothing else, and the citation needle does not move. It is also a small masterclass in method. By changing one variable at a time, they could actually attribute the (non-)result to the schema rather than guessing. That discipline is the whole point of the last section of this article.

4Why is the schema-equals-citations belief so sticky?

If the data is this unambiguous, why is structured data still near the top of nearly every AEO guide? Because it has three properties that make advice spread regardless of whether it works.

• It is concrete. "Add FAQPage schema" is a task with a clear start and finish. "Become the source a model trusts" is not. Concrete tasks win checklist real estate.
• It is fully in your control. You can ship schema without anyone else's cooperation. Earning third-party mentions or restructuring content to answer a question directly is slower and messier, so the controllable task gets over-weighted.
• It is familiar. A decade of SEO trained everyone that structured data unlocks rich results, so extending that intuition to AI citations feels obvious. It just happens not to hold up when tested.

None of that is a knock on the people repeating the advice. It is a reminder that intuitive, tidy, controllable recommendations survive in the wild whether or not the evidence supports them, which is exactly why a measured test beats a plausible-sounding checklist.

5What is schema still genuinely good for?

Here is the balance the contrarian headline can obscure: schema is still worth adding. The evidence says it is not a citation lever, not that it is worthless. Those are different claims, and conflating them leads teams to strip out markup that is quietly doing useful work.

So do it. Mark up your pages, keep the markup valid, keep your entity data consistent. Just file it under correctness and machine-readability, not under "how we get cited." The moment schema starts consuming the hours that should go to content and third-party presence, it has been mis-prioritized.

6What actually correlates with getting cited?

If not markup, then what? The studies that look at what cited pages have in common keep landing on content-level signals, not technical ones. Three show up repeatedly and are worth internalizing because they are where your effort compounds.

Sources: Search Engine Land citation research, Surfer's key-facts study, and Ahrefs on freshness.

The pattern is consistent. Answer engines lift passages that answer the question directly, that are dense with the specific facts the question needs, and that are recent enough to trust. Search Engine Land found a clear answer capsule, a self-contained block that states the answer up front, in 72.4% of cited posts. Surfer found cited pages carry noticeably higher key-fact coverage than uncited ones, 31% against 24%. Ahrefs found AI assistants systematically prefer fresher content. Notice what is absent from that list: the presence of a JSON-LD block.

7Where should you spend the effort instead?

The reason this debate matters is not academic; it is about where a finite amount of time goes. If your team spends its AEO hours adding and re-adding schema, those hours are not going to the work the data says actually moves citations. The alternative is a loop, run one buying question at a time, that treats every change as a hypothesis to be tested rather than a checklist item to be ticked.

This is the same discipline that let the single-variable schema test reach a trustworthy verdict: isolate the change, then re-measure. It is also the honest posture for a field where no one controls the output. You are not guaranteeing a citation; you are shifting the odds and verifying the shift. If you want the fuller method, the guides to getting cited by Perplexity and ChatGPT and getting recommended by AI walk the whole loop end to end.

Add valid schema once, keep it healthy, and move on. Then put your real hours where the evidence points: a clear answer up top, the facts the question needs, content kept current, and presence on the third-party sources answers are built from. Structured data is a good habit. It is not the shortcut it is sold as.

Frequently asked questions