Does schema markup help you get cited in AI search?
Adding schema markup does not reliably earn more AI citations. Ahrefs tracked 1,885 already-cited pages that added JSON-LD and found no meaningful lift on ChatGPT or Google AI Mode, and a small but statistically significant 4.6 percent drop on Google AI Overviews against matched controls. We still add clean schema because it helps machines read a page correctly, but the citation work sits in the answers on the page and in what third parties say about you.
Key takeaways
- Ahrefs compared 1,885 pages that added JSON-LD with 4,000 matched control pages across three AI surfaces.
- The 4.6 percent AI Overview dip was relative to controls, on heavily cited pages, and the study does not explain it.
- Whether schema helps a page with no AI citations earn its first one is a question this data cannot answer.
- In IT services, a company's own pages were 2 percent of top AI citation sources, per Trendos data.
What did Ahrefs' schema study measure, and on which pages?
It measured how AI citations changed for 1,885 pages that added JSON-LD schema between August 2025 and March 2026, against 4,000 matched control pages. Every page in the set was already heavily cited, with 100 or more Google AI Overview citations in February 2025, before any schema went on.
The study, written by Louise Linehan and reviewed by Ryan Law, was published by Ahrefs on May 11, 2026. It tracked citations on Google AI Overviews, Google AI Mode, and ChatGPT. The control group is why we take it seriously, because AI answers shift for everyone from month to month.
Several pages ranking for this query report large AI visibility gains from schema with nothing to compare against. Ahrefs' test is one of the few with a control group.
Read every number here as relative to the controls. A 2.4 percent rise on AI Mode means the schema pages did 2.4 percent better than their matched pages, and Ahrefs treats both rises as statistically indistinguishable from zero.
| Surface | Change against matched controls | How Ahrefs reads it |
|---|---|---|
| Google AI Overviews | Down 4.6 percent | Small but statistically significant |
| Google AI Mode | Up 2.4 percent | Indistinguishable from zero |
| ChatGPT | Up 2.2 percent | Indistinguishable from zero |
Why did Google AI Overview citations fall after schema was added?
The study cannot tell us why. It records a small, statistically significant 4.6 percent drop relative to matched controls, on pages that were already among the most cited, during a period when the treated pages and the controls were both losing AI Overview citations.
So the schema pages fell a little further than pages that were already falling, and we would not repeat that as proof that schema hurts.
Two explanations fit the data, and neither is proven. Pages starting with 100 or more citations had a lot to lose, and may have drifted down for reasons unrelated to markup. Sites that added schema in that window may also have changed templates or copy at the same time, which a schema count would not separate out.
The effect could also be real and specific to how AI Overviews picks its sources. Nothing in the study confirms or rules that out, and we are not going to guess at it here.
Is schema still worth adding for pages AI tools do not cite yet?
We add it, for the job schema is built to do, which is telling machines plainly what a page is, who published it, and what it offers. Whether it helps a page with no AI citations earn its first one is a separate question, and the Ahrefs data cannot answer it.
Every page in the study started inside the set AI Overviews already drew from. A page with zero citations might respond to markup differently, or it might respond exactly the same way. The study did not test that group.
In practice we treat JSON-LD the way we treat a correct title tag. It costs little and stops a crawler from misreading the page, so it goes on every page we build. We do not sell it as a citation lever.
That leaves the question standing. For a company with no AI citations yet, it is the more useful of the two schema questions, and the evidence so far was collected on someone else's pages.
What earns AI citations if markup does not?
Clear answers on your own pages, and mentions in the places AI tools already cite. In one breakdown of 107 million AI answers, a company's own pages were a small share of what got cited.
Trendos analyzed 107 million AI answers across ChatGPT, Perplexity, Gemini, and Google AI Overviews, as Trendos co-founder Gintare Rimolaityte wrote in Search Engine Journal on August 27, 2026. In IT and solutions services, brand-owned sources such as a company's own pages were 2 percent of the top 10 citation sources. Community and user generated content made up 51 percent, and independent editorial and reference sources made up 47 percent.
That is one industry, and other categories may split differently. In that one, though, 98 percent of the top sources were written by someone other than the brand.
On your own pages, the work is putting the answer a buyer asks for in the first lines, in plain sentences an assistant can quote. Off your pages, it is being worth mentioning in the forums, reviews, and editorial sites your category's answers draw on. Both take one to three months to show movement, and neither comes with a promise that any assistant will cite you.
Where the facts in this piece come from
FAQs about Does schema markup help you get cited in AI search
1. Does schema markup hurt AI citations?
The Ahrefs study does not show that in general. It found a 4.6 percent drop on Google AI Overviews relative to matched controls, on pages already heavily cited while both groups were declining, and no meaningful change on ChatGPT or AI Mode.
2. Should I remove schema markup from my site?
We would not. A small, unexplained dip on already cited pages is not a reason to strip markup that helps machines read your pages correctly.
3. How long does AI search work take to show results?
Movement usually shows in one to three months, which is slower than paid ads. We take a baseline before changing anything, so a later rise or fall can be checked against where you started.
About Corvan
Corvan makes Corvan AI, AI agents that run your marketing: they plan your campaigns, write every post in your voice, and show you which posts booked meetings. The team also runs go-to-market services: consultation, ad operations, and AI search ranking. Every piece under Resources is written by the team that runs the work, with sources. Meet the team on the team page, or read what it returned for Drive Me Barcelona.
Keep reading
Do G2 and Capterra reviews help you get cited by AI?
Almost every tool ChatGPT names has a G2 and Capterra profile, yet review volume tracks slightly negative against ranking and the pages earn no citations.
Read more →
Does a Wikipedia page help you get cited by AI tools?
ChatGPT cites Wikipedia often, but studies count it in different ways and disagree, and Wikipedia's own rules mean no company can buy its way onto a page.
Read more →
Do ChatGPT and Google AI Overviews cite the same sites?
ChatGPT and Google AI Overviews name similar brands about a third of the time and cite almost entirely different sources. Plan for each engine on its own.
Read more →