Get Optimised The GEO industry’s memory

Claim File First published 27 July 2026 Last reviewed 27 July 2026

Does schema markup improve AI search visibility?

The answer

Schema markup correlates with AI citations, but controlled testing shows mixed results. Implementing standard structured data is a safe baseline, not a standalone strategy.

Verdict: mixed evidence, beneficial but not guaranteed Confidence: moderate

The claim

Schema markup is routinely promoted as a must-have for AI search visibility. Since mid-2023, over 80 publishers have published nearly 90 promotional pieces calling it a foundation for getting cited by ChatGPT, Perplexity, and Gemini. The pitch is intuitive: structured data helps machines understand content, so it should help LLMs retrieve and cite it.

But a quieter counter-narrative started in January 2026. A handful of publishers began asking direct questions: does schema markup actually move citations? Early tests and audits returned mixed signals. And the promotional machine kept running: 73 more promotional pieces appeared after the first doubt.

The receipts

Three data points frame the debate.

First, the correlation: a mid-2025 study from betteraisearch.com reported that 81% of AI-cited pages had some form of schema markup. That number is cited widely by promoters. But the same study noted FAQ schema appeared on only 1.8% of cited pages, suggesting the type of schema matters more than presence.

Second, the controlled test: Search Engine Journal published a test in May 2026 showing that adding schema markup did not move AI citation counts in an Ahrefs experiment. This was a direct challenge to the correlation narrative.

Third, the arc: the first published doubt came from amicited.com on 2026-01-08. That same publisher had published three promotional schema guides in the two weeks prior. The pattern since then is a stream of questions and studies side by side with more promotions.

81

Percentage of AI-cited pages that use schema markup, according to a study by BetterAISearch in April 2025.

That number is often used to argue schema is essential. But the same study found that FAQ schema, one of the most common types promoted for AI, appears on fewer than 2% of cited pages. The headline correlation hides a critical detail: not all schema types contribute equally, and some may not contribute at all.

Meanwhile, the Ahrefs test from Search Engine Journal showed that adding schema markup to pages that previously had none produced no measurable change in AI citation rates. Correlation does not equal causation, and controlled tests point both ways.

What this means for you

  1. Implement foundational schema types first. Start with Article, Organization, and Person schema. These are the most commonly seen types on AI-cited pages. Do not chase exotic schema types until the basics are correct. A broken schema is worse than no schema.
  2. Treat schema as a hygiene factor, not a lever. Schema markup may help LLMs parse content, but it won't compensate for weak content, poor topical authority, or a site that doesn't get crawled. Focus on content quality and entity coverage first. Schema is the icing, not the cake.
  3. Track your own citation changes with and without schema. Run a small A/B test on a live site: enable schema on a set of pages for 30 days and compare citation rates against a control group. Use tools like Am I Cited or BrandCited to measure. No one else's test can replace your own context.

Confidence, and what would change our mind

We are moderately confident that schema markup is not harmful and may help in specific contexts, but the evidence for a direct causal effect on AI citations is thin. The correlation is strong but could reflect that high-quality pages tend to use both schema and strong content. Controlled experiments are rare, and those that exist show no effect.

We would raise confidence to high if a large-scale, multi-domain A/B test showed a statistically significant lift in AI citations for pages with schema versus identical pages without. A single test from SEJ suggests no lift, but independent replication is needed.

Sources