The First Controlled Schema Study: What It Means for AI Visibility

For the past two years, AEO content has been full of confident claims about schema markup. "Sites with proper schema get cited by Perplexity 67% more often." "Appear in ChatGPT 3.2x more frequently with structured data." These numbers circulate without methodology, without sample sizes, without primary sources.

In May 2026, Ahrefs ran the first controlled study. The results don't match the claims.

What the Ahrefs study actually tested

The methodology is the part worth reading carefully. Ahrefs tracked 1,885 pages that added JSON-LD schema markup between August 2025 and March 2026. They matched these against 4,000 control pages from different domains with similar pre-treatment citation histories that never added schema. Then they ran a difference-in-differences analysis to isolate schema addition as the variable.

Platforms measured: Google AI Overviews, Google AI Mode, and ChatGPT. Measurement window: 30 days before and 30 days after schema addition.

The results by platform:

| Platform | Citation change | Significance | |---|---|---| | Google AI Mode | +2.4% | Not statistically significant | | ChatGPT | +2.2% | Not statistically significant | | Google AI Overviews | -4.6% | Statistically significant (negative) |

No meaningful uplift on any platform. The AI Overviews negative result is significant at roughly 1-in-2,500 probability of being chance -- meaning it's real, and it goes the wrong direction.

In our session 26 research (2026-05-23), we documented this study in detail and updated our internal knowledge base. The conclusion: schema markup should not be framed as a citation frequency booster. The data doesn't support it.

The critical caveat nobody's running with

The Ahrefs study has a methodological constraint that matters for how you interpret the finding.

Every page in the dataset already had 100 or more AI citations before schema was added. These pages were already in AI systems' consideration set -- crawled, known, cited. The study tests whether adding schema pushes already-visible pages to be cited more. It does not test whether schema helps completely unknown pages get noticed in the first place.

This distinction is important for most local businesses. If your business is being cited by AI platforms regularly and you want to be cited even more -- schema addition probably won't move that number. If your business is essentially invisible to AI platforms, the Ahrefs study doesn't answer your question.

What the study does kill: the idea that schema markup is a citation frequency lever for pages with any meaningful existing AI presence. At that point, it's noise.

What schema actually does

Our methodology rec filed after reviewing this study (2026-05-23) requires a framing update for how we present schema recommendations. The frame that survives the evidence:

Schema markup is entity establishment infrastructure, not a citation booster.

For a local business that AI systems don't recognize yet, LocalBusiness schema with accurate name, address, telephone, and sameAs links to verified directories gives the model a structured, unambiguous description of what your business is and where it operates. This is a prerequisite for accurate citation -- not a driver of citation frequency. The distinction matters.

FAQ schema is still worth implementing on informational content pages. The correct frame: "helps AI systems extract and present your Q&A format accurately," not "increases how often you appear."

Entity fragmentation: the schema mistake with real consequences

There's a schema implementation pattern we track in our audits that does have measurable consequences -- but it's about what schema does wrong, not what it adds.

Our research vault documents what we call entity fragmentation (documented in our internal knowledge base from session 9, 2026-04-30). The mechanism: if your Organization schema `name` field says "Sourcepull Inc." but your website header says "Sourcepull" and your Product Hunt listing says "Sourcepull - AI Audit Tool," an AI model may treat these as three different entities. Each description competes against the others, and no single description reaches the confidence threshold for reliable citation.

The inverse is also true: if G2, Capterra, Crunchbase, and LinkedIn all describe your business using the same semantic terms, the model's confidence that these references point to the same real entity increases. Consistency across four or more sources outperforms completeness in any single source.

Stanford AI Index 2024 puts the baseline problem in context: 18%+ of LLM outputs involving brand entities contain either hallucinations or entity misattributions. For local businesses and small SaaS brands, the rate is higher -- they're underrepresented in training data, which increases the probability the model defaults to whatever entity is most familiar. Schema inconsistency compounds this.

The practical audit finding: pick a canonical business name and use it exactly across your website, schema, Google Business Profile, and every directory listing. This is where schema implementation failure creates real damage -- not in whether you have schema, but in whether your schema contradicts itself and your other sources.

What actually drives citation frequency

The Ahrefs study's own post-study commentary -- and the critique from iloveseo.net that documented the sample selection issue -- both point to the same set of drivers for citation frequency.

Topical authority expressed through multiple pieces of content. When a business has answered the full set of sub-questions a searcher might ask in a given category -- the how-to, the comparison, the explainer, the cost guide -- AI platforms have more surface area to cite across multiple query angles. A single service page ranks for fewer citation triggers than a cluster of topical content.

Third-party citations in the specific directories each AI platform trusts for the relevant category. Our research from SE Ranking (November 2025, verified) found that domains listed on Trustpilot, G2, Yelp, and similar platforms had a 3x higher ChatGPT citation rate than domains without those listings. Schema is not in this finding -- directory presence is.

This means for most local businesses, the fix priority is inverted from what AEO marketing content suggests. Schema establishes the entity. Directory presence drives the citation. These are different levers, and treating them as equivalent costs clients time on the lower-impact fix.

The practical takeaway

Implement LocalBusiness schema with your canonical business name, accurate address, and sameAs links to any directories where your business is verified. This is entity registration with AI systems -- it sets the baseline for accurate attribution.

After that, adding more schema types or refining JSON-LD is not where you'll see citation frequency move. The higher-leverage next step is consistent directory presence: same business name, same description, same category across Yelp, BBB, industry directories, and whatever category-specific platforms AI systems trust for your service type.

If you want to see where your business currently stands across ChatGPT, Perplexity, Gemini, and Claude -- and which of these levers are underbuilt for your specific profile -- Signal Check runs a live audit in under 60 seconds.