Ahrefs' Schema and AI Citations Study, Digested

Ahrefs, the SEO data company, published the study on 11 May 2026 under the title "We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved." It asks one question: when a page adds schema markup (JSON-LD structured data), do AI platforms cite it more often afterwards?

The question matters because schema is one of the most commonly sold GEO recommendations. Many audits list "add schema" or "add FAQ schema" as a quick win for AI answers. Until this study, that advice rested on correlation and vendor anecdote, not on a controlled comparison.

The study measured citations on three surfaces: Google AI Overviews, Google AI Mode and ChatGPT. The data window ran from August 2025 to March 2026.

Most GEO studies look at pages that are cited and count what they have in common. That approach can't separate cause from coincidence, because big, well-known sites tend to have schema and tend to get cited for reasons unrelated to markup.

Ahrefs used a matched-control design instead:

1. Treated group. 1,885 pages that added schema during the study window.
2. Controls. For each treated page, three comparable pages that did not add schema.
3. Comparison. Citation counts before and after the change, treated versus controls.

The controls absorb whatever else moved during the period: model updates, seasonal demand, changes in how often AI answers trigger. If schema caused more citations, treated pages should pull ahead of their controls. If they don't, the change in citations came from something else.

This is the same logic as a clinical trial with a comparison group. It is a stronger design than anything else published on schema and AI answers, which is why it carries weight. For background on why studies in this field reach different conclusions, see why GEO studies disagree.

The headline finding, in Ahrefs' words: "Adding schema produced no major uplift in citations on any platform."

The per-platform results:

The only statistically significant result runs the wrong way for anyone selling schema. AI Overviews citations fell on treated pages relative to their controls. The small rises on AI Mode and ChatGPT are not presented as meaningful lifts.

This lines up with what Google says first-hand. Google states that structured data "isn't required for generative AI search", and its generative AI guide says you don't need new markup to appear in its AI features. The digest of Google's generative AI optimization guide covers that document in full. A related change: Google deprecated FAQ rich results in May 2026, so FAQ markup no longer earns a special Google display either.

Ahrefs disclosed its own boundary, and it is the most important caveat on the page: "Every page in the dataset had 100+ AI Overview citations in February 2025, before any schema was added." In its own summary, "we studied pages that were already being cited heavily by AI."

That shapes how far the result travels:

• It tests pages already inside the set. The study shows schema does not lift pages that AI already cites often.
• It does not test entry. Whether schema helps a page with zero citations get considered at all is untested. Untested is not the same as refuted.
• Many brands sit outside the sample. A mid-market company that AI never names is exactly the population Ahrefs excluded.
• It does not show harm from keeping markup. The 4.6% fall is about adding schema during the window, measured on AI Overviews. Google says existing markup does no harm.

So two over-readings are both wrong. "Schema gets you cited" is contradicted by the best evidence available. "Schema is useless, rip it out" isn't supported either. The honest line is narrower: schema is not a lever for AI answers.

The schema result is one of several nulls. A July 2026 critical survey of 45 GEO studies concluded that no reviewed technique shows a stable, cross-platform causal effect on whether content gets discovered. C-SEO Bench found most citation-oriented rewriting methods ineffective or harmful. A 252,000-trial factorial found formatting-only edits have little impact.

What does hold up is access. Most AI crawlers download JavaScript but don't run it, according to Vercel's crawler study, so content that appears only after client-side rendering may be invisible to them (see the Vercel AI crawler study). Robots.txt permission is set per bot, and a firewall or CDN can block a crawler that robots.txt allows. These are publisher-controlled gates with documented consequences.

One on-page point does have a positive experimental signal: stating a specific fact, like a price or a one-line definition, in fetchable text gives the engine a sentence it can quote. The comparison of structured data and plain-text facts walks through the difference.

Treat the study as permission to stop spending on schema as an AI tactic, not as a reason to delete anything.

• Keep valid markup you already have. It does no harm, and it can still serve non-AI search features.
• Make sure markup matches the page. If your JSON-LD states a price or product name the visible page contradicts, fix the contradiction. The free JSON-LD checker compares the two.
• Drop "add schema for AI" from your roadmap. If an agency or tool lists it as a quick win for AI answers, ask for evidence that controls for brand size.
• Move the effort to access. Check that key facts are in the HTML before JavaScript runs, that OAI-SearchBot, PerplexityBot and Claude-SearchBot are allowed in robots.txt, and that your CDN doesn't return 403 to them.
• Write the facts plainly. Put prices, plan names and definitions in visible text on pages crawlers can fetch.

If you want to see where you stand before changing anything, the free checker on this site shows whether AI platforms name you for up to three buyer questions. It is a snapshot, not a promise of what fixes will change. For the broader argument, read does schema markup help AI citations.