Does schema markup get you cited by AI? Less than you were told
The best controlled evidence so far suggests that adding JSON-LD schema alone does little for AI citations. A 2026 study comparing pages that added structured data against matched control pages found roughly no change in ChatGPT and AI Mode citations. What did correlate with citation was visible, specific information: real prices, ratings and facts on the page itself. Schema still earns its place for rich results in classic search and for making entities unambiguous, so keep it, but don't expect it to carry AI visibility on its own.
“Add structured data so AI can understand your site” has become standard advice. It sounds plausible: schema is machine-readable, and AI systems are machines.
In 2026 people finally tested it properly. The results are worth knowing before you pay for a schema project.
What the controlled study found
The most rigorous test published so far, by Ahrefs, followed pages that added JSON-LD markup between August 2025 and March 2026 and compared them with a matched set of control pages that had similar citation levels beforehand.
The reported effect on citations in ChatGPT and Google AI Mode was roughly zero. For AI Overviews, the treated pages did slightly worse than the controls.
The same analysis looked at a narrower case. Pages with Product or Review markup that contained real, visible prices and ratings were cited more often than pages with generic markup. The pattern points to the visible facts, not the markup, doing the work.
Why other studies said the opposite
You will find case studies reporting dramatic gains after a schema rollout. Most of them are before-and-after comparisons without a control group. A site that rolls out schema usually also fixes titles, adds content, improves speed and earns links in the same period. The gain is real; the attribution is not.
Correlational research, such as academic work finding structured data among the signals associated with citation, has the same limitation. Sites that implement schema carefully tend to do many other things carefully too.
Why visible content wins
When an AI system answers a question, it retrieves pages and works with what they say. Experiments that placed facts only inside markup, with nothing on the page, found the systems did not pick them up.
That matches how these systems are built. They are trained and grounded on text. Markup is a label on content; it cannot stand in for content that isn’t there.
What schema is still good for
Keep it, for the right reasons:
- Rich results in classic search. Product price and availability, review ratings, breadcrumbs and other enhancements still appear where Google supports them, and they improve click-through.
- Entity clarity. Organization markup with consistent name, logo, address and profile links helps search engines connect your site to the rest of your footprint.
- Data quality discipline. Writing accurate markup forces you to have accurate prices, hours and details on the page, which is the part that does matter.
What actually moves AI visibility
The work that has support is less technical:
- Put the specifics on the page. Prices or ranges, timelines, locations served, named clients where you are allowed to name them.
- Make every page answer one question clearly, starting with the answer.
- Keep your company described the same way everywhere: site, profiles, directories, review platforms.
- Earn independent mentions. AI systems corroborate, and third-party sources carry weight.
A similar story applies to another popular fix: do you need an llms.txt file. And if you want the broader picture of optimizing for answers rather than rankings, start with GEO and SEO are not the same thing.
Common questions
Should I remove schema markup from my site?
No. It still powers rich results in classic search, such as product prices, review stars, FAQs and breadcrumbs where Google displays them, and it helps search engines identify your organization unambiguously. It is useful infrastructure. It is just not a shortcut to AI citations.
Why would AI systems ignore schema?
When an AI system retrieves a page to answer a question, it typically works with the visible text. Tests that placed information only in markup, and not on the page, found the systems didn't use it. Markup describes content; it doesn't replace it.
Which schema types are still worth adding?
Organization or LocalBusiness on the homepage, Product with real price and availability for stores, Article for editorial content, BreadcrumbList for structure, and FAQPage where the questions are visible on the page. All of them should mirror what a visitor can see.
