Ahrefs finds JSON-LD schema does not boost AI citations directly

Schema’s biggest lesson is what it did not do

JSON-LD schema has become one of the easiest things to blame, or to celebrate, whenever AI search visibility shifts. Ahrefs’ latest test cuts through that habit fast: adding schema alone barely changed AI citations. The result is less a victory lap for markup than a warning about the causation trap that keeps misleading marketers.

The lure is obvious. Ahrefs found that pages cited by AI were almost three times more likely to have JSON-LD than pages that were not cited. That kind of gap can look like a simple optimization play, until you test the tactic directly. Once Ahrefs tracked 1,885 pages that added JSON-LD schema and compared them with 4,000 control pages across Google AI Overviews, Google AI Mode, and ChatGPT, the supposed advantage mostly vanished.

Why the correlation looked so persuasive

A lot of teams want schema to be the missing switch for AI visibility because it feels concrete, controllable, and tidy. If cited pages are disproportionately marked up, the temptation is to treat JSON-LD as the reason those pages win. Ahrefs’ earlier 6 million-URL analysis showed exactly why that story spreads so quickly: the overlap between markup and citations is real enough to notice, but that does not prove the markup itself is causing the citation.

That distinction matters because AI systems do not reward one isolated signal in a vacuum. Better-maintained sites tend to have cleaner structure, stronger editorial habits, and more disciplined site hygiene, which makes them more likely to use schema well. In that sense, schema can be a marker of overall quality rather than the source of it. The new study is useful precisely because it resists the easy conclusion and points instead to the broader trust and relevance signals that appear to matter more.

What the experiment actually showed

Ahrefs tracked pages that added JSON-LD between August 2025 and March 2026, then measured whether citation behavior changed after implementation. Across Google AI Overviews, Google AI Mode, and ChatGPT, there was no major uplift in citations. AI Overviews slipped slightly, while AI Mode and ChatGPT showed changes so small they were statistically indistinguishable from noise.

The clearest number in the study was a 4.6% relative decline in AI Overview citations for treated pages. Ahrefs described that as small but statistically significant, which is important because it keeps the conclusion grounded. This was not a dramatic penalty, and it was not evidence that schema hurts. It was evidence that schema, by itself, does not reliably move citation rates in the direction marketers hoped.

That makes the practical takeaway sharp: schema is a support layer, not a standalone lever. It can still help machines understand page structure, and it can still be useful for machine readability. But if a page is weak on substance, topical fit, or authority, adding JSON-LD does not appear to rescue it.

Google’s own guidance points in the same direction

The Ahrefs findings land harder because Google has been saying, in effect, that there is no special trick to chase. Google Search Central says there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode. Google also says these features may use a query fan-out technique, surfacing relevant links only when the AI answer is additive to classic Search.

That matters because Google is steering publishers away from the idea of a schema hack. In May 2025, Google launched AI Mode in the United States and described it as its most powerful AI search experience, powered by advanced reasoning and query fan-out. Google also said AI Overviews drove over a 10% increase in usage of Google in its biggest markets like the U.S. for queries that show the feature. In other words, the audience for AI-search visibility is growing, but the platform’s own guidance still points back to ordinary SEO fundamentals rather than special markup games.

What the shifting citation landscape says about AI search

The temptation to overread schema is partly a product of how unstable AI citation patterns have been. Ahrefs’ July 2025 study found that 76.10% of AI Overview-cited pages ranked in Google’s top 10, 9.50% ranked between positions 11 and 100, and 14.40% did not rank in the SERPs at all. That already suggested AI citations and web rankings overlap, but not perfectly.

Then the pattern shifted again. Ahrefs later reported that only 38% of AI Overview citations came from top-10 pages in a newer dataset, down from 76% a year earlier. That kind of swing is exactly why easy formulas break down. When the ecosystem itself is moving, a single technical change like JSON-LD is unlikely to explain much on its own.

The right reading is more cautious and more useful. AI citation behavior seems to be shaped by a bundle of signals: trust, content quality, topical relevance, and authority. Schema can support those signals, but it does not replace them. If a page is already strong, schema may be part of a well-run system. If the page is not strong, schema alone will not make it cite-worthy.

Why the Search Engine Land experiment still matters

There is one more layer that keeps this story interesting rather than tidy. Search Engine Land ran an experiment in September 2025 with three nearly identical pages, and only the page with well-implemented schema appeared in an AI Overview and achieved the best organic ranking. That result does not contradict Ahrefs so much as it complicates the story.

The difference is implementation and context. A page with strong, well-implemented schema may also be the page that is better maintained, better structured, and more likely to align with broader quality signals. That makes schema look powerful, but it may be acting as part of a larger bundle rather than as the main cause. The latest Ahrefs test pushes the field toward that more disciplined conclusion: markup may help, but it is not a dependable citation switch.

The takeaway for AI visibility work

The temptation in AI SEO is to hunt for the quickest visible lever. This study is a reminder that visibility often comes from the parts you cannot isolate as neatly as a line of JSON-LD. Schema still belongs in the toolkit, especially if it helps systems understand structure and context more clearly.

But the strategy cannot stop there. The pages most likely to earn AI citations are still the ones with stronger substance, clearer topical authority, and healthier overall site signals. Ahrefs has now shown both sides of that story: a strong correlation between citations and schema, and a direct test that shows schema alone barely moves the needle. That is the kind of result that should reshape the conversation, because it replaces a catchy shortcut with a more durable truth.