Does Schema Markup Actually Help AI Search?

Most AEO providers sell structured data as the main lever. A controlled study of 1,885 pages says otherwise. Here is the evidence, and what to do with it.

This contradicts most of what's currently being sold as answer engine optimization, so it's worth walking through the evidence carefully.

What the Ahrefs study found

Ahrefs started with a broad correlational analysis of roughly 6 million URLs. Pages cited by AI were almost three times more likely to have JSON-LD markup than pages that weren't. That figure spread quickly and now appears in a large share of AEO pitch decks as proof that schema drives AI visibility.

Then they ran the causal test: track pages that added schema during a defined window, compare them against matched controls that didn't, and measure the change in citations across Google AI Overviews, AI Mode, and ChatGPT.

The effect disappeared. Across thousands of URLs, the movement was small enough to be noise.

Why does everyone believe schema drives citations?

This is a textbook confounding variable, and it's an easy mistake to make in good faith. It's also a convenient one to make in bad faith: schema audits are fast to produce, easy to demonstrate, and can be billed as a retainer. "We added structured data to 40 pages" is a satisfying deliverable. It just isn't the thing that moves the number.

So should I still use schema?

Yes — for the right reason. Schema's job is entity definition: telling answer engines who you are, what you do, and that the various profiles bearing your name all refer to one business. Models need to classify you with confidence before recommending you, and structured data removes ambiguity from that process.

That makes it necessary but not sufficient. Table stakes, not a growth lever.

What's worth implementing:

• Organization schema on every page, with sameAs pointing at every verified profile you control. This is the highest-value field — it's how models corroborate your identity.
• Person schema for the author or founder, linked to the organization.
• Article schema on content pages, with real author attribution and an accurate dateModified.

What isn't worth much effort: exhaustively marking up every page type, chasing deprecated rich-result formats, or treating structured data as an ongoing program. It's largely a one-time job done correctly.

What actually correlates with AI citations?

Add to that: brands are roughly 6.5x more likely to be cited via third-party sources than their own domain, and 97% of AI Overview citations come from the top 20 organic results.

The pattern is consistent. What moves AI visibility is what the rest of the internet says about you, not what your own markup asserts.

What should I do instead?

1. Verify AI crawlers aren’t blocked. Free, five minutes, and the most common silent failure there is.
2. Get the entity right once. Consistent name and description everywhere, Organization schema with complete sameAs. Then stop optimizing it.
3. Baseline your visibility. Twenty prompts, four engines, weekly, before you change anything else.
4. Publish video. Highest measured correlation, and almost no small business competes there.
5. Earn mentions you don’t control. Podcasts, press, directories, community participation.
6. Make content extractable. Direct answers, question headings, attributed data.

Roughly 30% of your effort belongs on your own site. The other 70% belongs everywhere else.