Several studies report large AI citation lifts from schema markup. The study that tested cause and effect found almost none. Here's why the numbers disagree.
A SaaS marketing team analyzed 16,851 queries and found pages with JSON-LD schema got cited by ChatGPT 38.5% of the time versus 32.0% without it. A separate study of 1,000 Google AI Overviews found schema-marked pages cited 2.3 times more often than unstructured equivalents. An academic paper measuring 1,702 citations across three AI engines found structured data carried a statistically significant 39% lift. A fourth, presented at BrightonSEO, reported a sitewide schema rollout across 2,000-plus URLs that pushed Google AI Overview citations up 1,500%.
Four independent teams, four different methodologies, all pointing the same direction. That's the kind of agreement that normally settles a debate. Then Ahrefs ran the one study designed to answer the question everyone else was dancing around, and the result barely moved at all.
What the other four studies measured
Every one of those headline numbers comes from comparing pages that already have schema against pages that don't, then looking at which group gets cited more. AirOps partnered with Kevin Indig on the 16,851-query ChatGPT analysis, finding BreadcrumbList at a 46.2% citation rate and FAQPage at 45.6%. Digital Applied's 1,000-AI-Overview study compared 4,243 cited URLs against roughly 50,000 non-cited controls for the same queries and reported the 2.3x figure after controlling for domain authority. The arXiv paper behind the "GEO-16" framework found structured data correlated with citation likelihood at r=0.63, translating to a reported 39% lift.
All of that is real data, honestly reported. None of it can tell you what happens if you personally add schema to a page tomorrow, because none of it added schema to anything. Each study took a snapshot of the web as it already existed and compared pages that happened to have schema against pages that happened not to. That's a correlational design, and correlational designs can't separate "schema causes citations" from "the kind of site that bothers installing schema also tends to do a dozen other things that cause citations."
The one study that added schema and watched what happened
Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched each one against three control pages from different domains with similar citation levels that never added schema, then measured citation counts in Google AI Overviews, Google AI Mode, and ChatGPT for the 30 days before and after the schema went live. That before-and-after, treatment-versus-control structure is what makes it a genuine test of cause and effect rather than a comparison of two different populations.
The result: Google AI Mode citations moved 2.4%, ChatGPT moved 2.2%, both statistically indistinguishable from random noise. Google AI Overviews declined 4.6%, a result Ahrefs says is statistically significant but too small and too entangled with a pre-existing downward trend to confidently blame on the schema itself. Ahrefs ran four separate statistical tests on the same data specifically to check whether the finding would hold up, and all four landed in the same place: adding schema to a page that's already being cited does not meaningfully change how often it gets cited.
Ahrefs' own explanation for why their result conflicts with the correlational studies is the cleanest statement of the problem: pages that already run schema markup also tend to invest in technical SEO, publish stronger content, earn more links, and maintain their pages more carefully. Strip the schema out and most of those other signals are still there, still doing the work. Across their broader 6-million-URL sample, AI-cited pages were almost three times more likely to carry JSON-LD than non-cited pages, which is nearly the same ratio Digital Applied reported as a causal lift. Ahrefs found that gap and specifically built a second study to test whether it was real before publishing it as one.
Even the correlational studies don't agree on direction
The BrightonSEO-presented sitewide rollout is worth sitting with for a different reason: Google AI Overview citations reportedly rose 1,500% and AI Mode rose 377%, but citations on ChatGPT, Gemini, and Copilot dropped on the same rollout, and Perplexity showed no change at all. If schema markup were a straightforward citation lever, it should move every platform's citation count in the same direction, even if by different amounts. A rollout that helps two Google surfaces while apparently hurting three competitors is a pattern that fits "something else changed on this site at the same time" at least as well as it fits "schema worked."
Where this leaves a directory listing specifically
Ahrefs flags one honest limitation of their own result that matters directly for a listing page: every page in their 1,885-page dataset already had more than 100 AI Overview citations before any schema was added. Their study measured what happens when you add schema to a page already inside the AI consideration set. It says nothing about what schema does for a page nobody's citing yet, which describes most new directory listings in their first weeks of existence far better than it describes an established publisher's archive.
That's not a loophole that rescues the "44% more citations" style claims circulating in SEO content. It's a genuine, unanswered question, and the honest answer for a brand-new listing page is that the best available evidence doesn't cover that case either way. What the evidence does rule out is the specific claim that adding Organization or ItemList schema to a page Google already indexes and already surfaces is likely to meaningfully change how often an AI system cites it. If a vendor's pitch for a schema-markup service leans on a citation-lift percentage as the reason to buy, ask whether their number comes from a before-and-after test or a snapshot comparison, because on the only evidence that isolated cause from correlation, the before-and-after answer is close to zero.
What schema markup is still worth doing for
None of this makes schema markup pointless. It still helps Google parse what an entity is, how it connects to other data on your page, and whether your pricing, ratings, and availability information is accurate enough to trust for rich results, voice assistants, and knowledge graph entries. SearchVIU's separate test of five AI systems fetching pages in real time found none of them read the JSON-LD at retrieval time, every one extracted only the visible HTML a human would see. That's consistent with Ahrefs' result: schema is plumbing Google and other crawlers use to understand a page correctly, not a line item an AI model reads when deciding what to quote.
If a backlink directory listing is accurate, complete, and structured correctly, that's worth doing on its own terms. Just don't budget for it as the lever that gets a listing cited by ChatGPT.
