Schema Barely Moves AI Citations: Ahrefs Data

What did the Ahrefs schema study actually find?

Ahrefs tracked 1,885 pages that added JSON-LD schema against about 4,000 control pages from August 2025 to March 2026. The result was sobering: no meaningful citation growth in AI Mode or ChatGPT, and a small 4.6% decline in AI Overviews that they could not definitively pin on schema itself.

How was the study designed?

Ahrefs used a difference-in-differences approach, comparing pages that added schema against similar control pages that did not, over the same period. Notably, every studied page already had 100 or more AI Overview citations before schema was added. So the finding speaks to pages engines already saw, not brand-new ones.

Why didn't schema lift AI citations?

Because for pages already being cited, the markup added little new signal. Schema helps machines parse and categorize content, but it does not manufacture authority or relevance an engine was missing. When a page is already visible and trusted, layering on JSON-LD did not, in this data, push citation rates meaningfully higher.

Do AI engines even read JSON-LD on retrieval?

Often not at the moment of answering. Ahrefs cited tests where ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode extracted only visible HTML on direct retrieval, ignoring JSON-LD, Microdata, and RDFa. That means the words a reader actually sees on the page can matter more to an answer than hidden structured markup.

Should I stop adding schema to my site?

No, do not rip it out. Schema still supports traditional search features, helps engines parse your content, and carries little downside when done right. The takeaway is narrower: do not expect schema alone to boost AI citations on pages already being seen. Keep it as hygiene, just not as your main citation strategy.

When does schema still genuinely help?

It helps engines understand and categorize content, especially on a first crawl, and it still powers some traditional rich results. This is exactly why my HowTo schema tutorial frames it as a parsing aid, not a citation guarantee. Use schema where it clarifies meaning, and pair it with strong visible content for the best effect.

Which factors do move AI citations?

Authority, relevance, and freshness do most of the work. Pages that already rank well and read clearly tend to get cited, since about 38% of AI Overview citations come from Google's top ten. Clear, well-structured visible content and genuine topical authority move the needle far more than markup alone ever will.

How much does content freshness matter?

A lot, by the available data. SE Ranking found content updated within the past three months is about twice as likely to be cited by ChatGPT as older content, and Ahrefs found AI-cited pages run roughly 25.7% fresher than traditional organic results. Keeping content current is a stronger lever than adding schema.

Where should I spend effort instead of schema?

Put your hours into strong, clearly-written content, genuine authority, solid rankings, and regular freshness updates. Make answers easy to extract by writing them plainly in visible HTML. Schema stays a sensible baseline, but the bigger returns come from the fundamentals that engines actually reward when choosing whom to cite in an answer.

Can I test this on my own pages?

Yes, and you should. Pick a set of pages, add schema to some and leave comparable ones unchanged, then track AI citations over several weeks. Replicating the difference-in-differences idea on your own site tells you what works for your content far better than trusting any blanket rule, including this study's averages.

Spending effort where it counts? Pair this with my Organization schema tutorial, why ranking #1 no longer wins citations, and why AI citations vanish in 6 months. Let's chat.