Ahrefs found schema markup raised AI Mode citations by only 2.4 percent: Reveneau

Why it mattersIf your team's GEO strategy assumes schema markup brings in AI citations, this data shows no measurable increase, so budget that was going to schema work would do more if spent on title alignment with the sub-queries an answer engine actually asks.

Search Engine Journal reported on 9 September on new Ahrefs data that undermines two of the most common working assumptions in generative-engine optimisation. Matt G. Southern wrote the piece around research Ahrefs published on its own blog. The two findings a marketing team can act on today are worth stating plainly.

The schema markup number

Ahrefs tracked 1,885 pages that added JSON-LD schema markup between August 2025 and March 2026, and compared them to 4,000 control pages that did not. The pages in the sample already carried at least 100 AI Overview citations before the change, so this measures the extra gain on pages the answer engines already knew about, not first-time discovery.

The increase was small enough to be random variation. Citations from Google AI Mode rose 2.4% compared to controls. Citations from ChatGPT rose 2.2%. Ahrefs marked both as statistically insignificant. Citations from Google AI Overviews fell 4.6%, roughly 12 fewer citations per page per day on pages that were already receiving hundreds, in a period when both groups were declining anyway. Ahrefs said it cannot tell whether schema caused the decline or something else did.

The AI Overview ranking overlap

Across 863,000 keywords and about 4 million AI Overview URLs, 37.1% of the URLs cited in an AI Overview also appeared in Google's organic top 10 for the same query. Another 26.2% ranked between position 11 and 100. The remaining 36.7% were not in the top 100 at all. So organic ranking correlates with citation, and also fails to explain more than a third of it.

Ahrefs' second finding helps explain that gap. The company's ChatGPT analysis found that the titles of cited pages aligned more closely with the sub-queries ChatGPT generates from a prompt than the titles of non-cited pages did, and the alignment was closer to the sub-queries than against the original prompt. In plain terms: if a reader types "compare X and Y", ChatGPT breaks that into a set of narrower questions, and pages whose titles match those narrower questions get cited more often.

What a team should change now

Two things. First, if a schema markup project is on the roadmap on the theory that it will bring in AI citations, revisit the theory: on pages that answer engines already know about, this data shows no measurable increase on the two biggest AI search products. Schema is still useful for rich results in classic search, but that is a different argument.

Second, the finding about sub-queries (the smaller questions ChatGPT breaks a prompt into) gives a team something real to act on. Tracking the ranking of one main search term will miss a citation earned by a page that does not rank for that term but happens to answer a sub-query clearly. A tracker that only reports mentions or only reports rankings hides the cause when a score changes. Splitting the measurement into mentions, citations and rankings, and keeping them separate, is what lets you tell whether a drop is a positioning issue, a title-alignment issue, or a platform glitch.

Source

Brands Are Tracking AI Visibility, But Are They Measuring The Right Things?, Search Engine Journal, 9 September 2026
Primary source: Schema markup and AI citations, Ahrefs
Analysis of 863,000 keywords and AI Overview citations, Ahrefs

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.