The schema tag is very common on pages cited by AI. But a new Ahrefs report found that adding it didn’t lead to a clear increase in citations.
Ahrefs tracked 1,885 web pages that added the JSON-LD schema. Each page was compared to control pages that never added schema, and citation changes were measured across Google AI Overviews, AI Mode, and ChatGPT.
Neither platform showed a meaningful citation increase after the schema was added.
Discovered by Ahrefs
The report analyzed 6 million URLs and found that pages cited by AI were almost three times more likely to include JSON-LD. This gap was seen as evidence that schema improves AI visibility. However, Ahrefs tested whether this is true when separated from other signals, as sites with schema tend to invest in better content and gain more links.
They used a controlled comparison, matching each schema page with three control pages from different domains with the same citation levels that did not add JSON-LD. Citation changes were measured 30 days before and after program addition.
Using its Brand Radar tool and Agent A, Ahrefs performed a similar differential analysis to track platform trends. Here are the findings.
- Google AI overview: −4.6% (a small but statistically significant decrease relative to controls)
- Google AI mode: +2.4% (too small to distinguish it from a random variable)
- ChatGPT: +2.2% (too small to distinguish it from a random variable)
Three additional tests were conducted in parallel with the main comparison, and four did not find a clear positive or negative effect.
Degradation AI Overview
I −4.6% going down in the AI overview section deserves context. Ahrefs reports that both managed and controlled pages were already declining before the schema was added. Treated pages declined quickly, but the difference was small, with about 12 fewer daily citations per page in the sample where most pages received large ones.
The report notes that the decline may reflect a small negative impact from the schema, or it may be a coincidence. It does not foreclose the conclusion in any way.
What the Report Does Not Include
Each page in the dataset has 100+ AI Overview citations before any schema is added. These pages were already in a set of considerations, crawled and exposed.
The report acknowledges this limitation. For pages not yet visible to AI, the schema may still help crawl, parse, or index, but the data cannot confirm this.
The report also notes other limitations. Pages that add JSON-LD often change other elements, making it difficult to isolate the schema effects of those changes. All schema types were compiled, so some may behave differently. The 30-day window may miss slower results.
The searchVIU test mentioned in the report tested whether five AI systems used the schema tag when fetching pages in real time. No one does; they only output visible HTML, ignoring JSON-LD, Microdata, and RFa. This was a direct retrieval test, not evidence for a role of schema during training, identification, or retrieval.
Why This Matters
Schema markup is often recommended for AI visibility. However, Ahrefs’ data complicates this. Although the schema supports rich results and knowledge graphs, adding JSON-LD does not increase AI citations for already indexed pages.
The data shows a correlation: pages with a schema are cited more often by AI, but Ahrefs interprets this as a sign of the overall quality of the site rather than a direct influence of the schema.
Looking Forward
The report cannot determine whether the schema is valid for unindexed pages, which are a separate group of pages that require further research. If the pages are visible to AI, JSON-LD will probably not improve the citations.
Featured Image: Roman Samborsky/Shutterstock
