Few topics in AI SEO attract bolder claims than structured data. Agencies promise that a JSON-LD overhaul will make ChatGPT and Google’s AI features cite you more often. Some of that advice rests on real correlations. Very little rests on proof that schema causes citations. This article traces where the claims came from, what the best test found, and how to decide whether schema deserves your time.
Where the hype came from
The most persuasive early evidence was correlation. Ahrefs analyzed 6 million URLs and initially found AI-cited pages were almost three times more likely to contain JSON-LD. That sounds decisive, but pages with schema tend to belong to larger, better-maintained, more authoritative sites. Schema may simply travel with the traits that earn citations. strategicrevenue
Another popular number deserves suspicion. One analysis traced the widely repeated claim of a 112% citation increase from FAQ schema to a study by the agency TechCognate about Gemini. A later, larger test pointed elsewhere: SE Ranking found FAQ content blocks produced roughly 11% more citations (4.9 versus 4.4), not 170%, and the benefit came from the FAQ content structure rather than the markup.
The controlled test
The strongest evidence so far is an experiment, not a correlation. Ahrefs published a study on May 12, 2026 that tracked 1,885 pages adding JSON-LD between August 2025 and March 2026, matched against 4,000 control pages, across AI Overviews, AI Mode, and ChatGPT.
The results were close to nothing. AI Overviews citations fell 4.6%, while AI Mode rose 2.4% and ChatGPT rose 2.2%, the last two statistically indistinguishable from zero. The AI Overviews dip needs context: the authors caution against reading it as harm, since both treated and control pages were already declining before schema was added.
The study has limits, which a fair reading must include:
- The sample consisted of pages that were already receiving substantial citations, and several schema types were pooled together.
- Schema could still play a role earlier in the pipeline, such as indexing or training data, which a test of direct citations cannot rule out.
The evidence is not unanimous
Other studies point in different directions. One roundup reports that an OtterlyAI test found AI Overviews citations rising by 1,500% after markup, while citations fell on ChatGPT, Gemini, and Copilot and Perplexity showed no impact. That summary appears in a Shopify-focused guide, and I did not examine the original methodology, so treat the figure cautiously. The useful takeaway is that results vary by platform and design, which is exactly why single-study headlines mislead.
What Google and Microsoft say
Official statements are consistent with the weak experimental result. Google states that structured data is not required for generative AI search and that there is no special schema.org markup to add. Microsoft sounds somewhat warmer. Bing’s Fabrice Canel said at SMX Munich 2025 that schema markup helps Microsoft’s LLMs understand content. Understanding is not the same as citing more often, but it suggests schema has some role in how pages are interpreted. strategicrevenuewhitehat-seo
So should you bother with schema?
Yes, but for the right reasons. One analyst’s conclusion is that schema is still worth using for rich results and entity clarity, and should be treated as cheap infrastructure rather than the thing that gets you recommended.
At Sarkar SEO, we use a three-question test, which is our own working rule rather than an industry standard:
- Does it unlock a visible feature? Product, review, event, and similar markup can still support rich results in classic search. That value is real and independent of AI.
- Does it clarify who or what you are? Organization details, author profiles, and product attributes help systems connect facts to the right entity. Treat this as a hygiene benefit, not a guaranteed citation lever.
- Can you keep it accurate? Markup that drifts out of date is worse than none, because it creates contradictions between your code and your page.
If the answer to all three is no, spend the time elsewhere.
How to test schema on your own site
Rather than trust anyone’s number, copy the logic of the Ahrefs design on a small scale:
- Pick 10 to 20 similar pages that are already indexed and sometimes cited.
- Add markup to half of them, and leave the rest alone as a control.
- Run the same 20 prompts in several AI tools before and after, and log citations over a few weeks.
- Compare changes between the two groups, not simply before and after.
Your sample will be small, so treat results as directional. A control group still guards against the biggest error, which is crediting schema for improvements that came from elsewhere.
Implementation rules that avoid trouble
- Match the visible page. Google’s policies say not to mark up content that is not visible to readers, so FAQ schema needs visible FAQ content.
- Write real answers. One source argues dedicated FAQ pages with full-text answers outperform thin Q&A pairs. That is a vendor view, but sensible.
- Validate and maintain. Run markup through testing tools, then review it whenever prices, authors, or services change.
- Don’t change five things at once. If you add schema, rewrite copy, and refresh dates in one release, you cannot tell what worked.
- Be cautious with sales pitches. If a proposal centers on a schema overhaul as an AI citation strategy, ask for controlled evidence, not correlation.
