“Add schema and you’ll show up more in AI answers” is a recommendation repeated often in GEO content. But Google’s own statement, plus two independent controlled studies, largely disprove that claim: generic schema has no measurable effect on AI Overviews, AI Mode, or ChatGPT citations. The one place an effect does show up is specific schema types carrying concrete data — Product and Review, in particular.
This post takes the single-line checklist item from the pillar post — “structured data should match visible content” — and goes deeper, using three sources.
Google’s Official Position: No Special Schema
Google Search Central’s AI Features documentation addresses this directly:
“You don’t need to create new machine readable files, AI text files, or markup to appear in these features. There’s also no special schema.org structured data that you need to add.”
Google’s only general recommendation is that structured data match the visible text on the page — a longstanding general rule, not something AI-specific. Google’s message, in short: AI Overviews draws from the same index as organic Search, meaning pages already eligible for organic results; there’s no special markup type like “AIPage” or “LLMOptimized.”
Ahrefs’ Controlled Experiment: Close to No Effect
Ahrefs’ study, published May 11, 2026 (by Louise Linehan and Xibeijia Guan, reviewed by Ryan Law), is the most comprehensive experimental test of this question. Methodology:
- 1,885 pages that added JSON-LD schema between August 2025 and March 2026 were identified.
- These pages were matched against 4,000 control pages that never added schema, with similar pre-period citation levels.
- A difference-in-differences analysis was run alongside three additional statistical tests: a two-sample t-test, an event study, and a sensitivity analysis.
- The measurement window covered 30 days before and 30 days after the schema was added.
Results:
| Platform | Change after schema addition |
|---|---|
| Google AI Overviews | -4.6% |
| Google AI Mode | +2.4% |
| ChatGPT | +2.2% |
The researchers note that the AI Mode and ChatGPT gains are “statistically indistinguishable from zero,” while the AI Overviews decline is significant but points in the opposite direction of what schema advocates would expect. There’s a critical limitation: every page in the dataset already had more than 100 AI Overview citations as of February 2025, before any schema was added. So the study only answers “does schema make a difference on pages that are already being cited?” — it doesn’t test whether a page invisible to AI systems today gets discovered for the first time because of schema. Ahrefs flags this limitation itself.
The researchers also note that the high correlation between schema and AI citations isn’t causal: AI systems already favor well-maintained, technically mature, authoritative sites, and those sites also tend to already use schema. The correlation may come from both variables sharing a common root — site quality — rather than from schema’s own effect.
The Academic Study: Generic Schema Doesn’t Work, Concrete Data Fields Do
Kurt Fischman’s study, “Does Schema Markup Predict AI Citation?” (February 20, 2026, preprint on SSRN), takes a different angle than Ahrefs: cross-platform and by schema type.
Methodology: 730 AI citations were collected across 75 commercial queries in five categories — SaaS/Technology, Health, Finance/Insurance, Professional Services, and How-To/DIY — using ChatGPT (GPT-4o with web browsing) and Gemini (1.5 Pro with search grounding). The same queries were used to collect Google’s top 10 results via SerpAPI, yielding 1,006 unique control pages.
The findings come in two layers:
- Generic schema has no measurable effect. Pages with generic, CMS-default schema had a 41.6% citation rate, versus 59.8% for pages with no schema at all — meaning generic schema was associated with a lower citation rate than having no schema. The researcher attributes this not to a causal harm from schema, but to generic schema use correlating with lower-effort, less original pages.
- Schema with concrete data fields makes a difference. Pages using Product or Review schema with concrete fields — price,
aggregateRating, specs — reached a 61.7% citation rate, versus 41.6% for generic Article/Organization schema. The author notes this gap is especially pronounced on lower-authority domains, suggesting schema can partly compensate for a lack of authority.
Reading Both Studies Together
Ahrefs and Fischman use different platform mixes, different query sets, and different time windows, so the numbers can’t be added together directly. Still, the shared conclusion is clear:
- “Add schema and AI will notice you” hasn’t been proven. Neither study found a meaningful positive effect for generic or general-purpose schema.
- Concrete, verifiable data fields are a different category. Product/Review schema that turns visible page information — price, rating, stock status — into machine-readable data made a measurable difference in at least one study.
- Schema’s role is likely in discovery, not citation. Ahrefs’ own note supports this: schema provides no added benefit on pages already being cited; but on pages that haven’t been crawled or parsed yet, schema playing a role during crawling and indexing is still a plausible hypothesis — just not the question these two studies tested.
The Current Status of FAQPage and HowTo Schema
A point that often gets confused in GEO discussions: people say FAQPage and HowTo schema have been “removed,” but what’s removed is the visual search-result feature, not the markup. Google fully removed the FAQ rich result (the search appearance, the rich result report, and Rich Results Test support) as of May 2026; Search Console API support ends in August 2026. The HowTo rich result was similarly phased out across desktop and mobile between 2023 and 2025.
The important distinction: you don’t need to remove FAQPage or HowTo schema from your pages — the markup isn’t invalid or penalized, only the related visual rich result no longer displays. Ahrefs’ and Fischman’s studies also don’t measure through these two types specifically; they measure general JSON-LD presence and data-dense types like Product/Review.
Practical Takeaways
- Treat schema as a tool for accurately conveying content that’s already visible to a machine, not as an “AI visibility hack.”
- Don’t expect a citation boost from adding generic, template-sourced schema alone — neither independent study found evidence for that.
- Prioritize Product/Review schema on pages with concrete, verifiable data — price, rating, specs — since it can make a more noticeable difference on lower-authority sites.
- Always keep structured data matched exactly to the visible text on the page — that’s Google’s only AI-specific warning.
- Don’t remove FAQPage/HowTo markup, but stop expecting it to guarantee a rich result or an AI citation.
- Position schema as a complement to the visibility, accuracy, and business-outcome three-layer measurement — not a substitute for crawlability and content clarity work.
Sources
- Google Search Central: AI Features and Your Website
- Ahrefs: We Tracked 1,885 Pages Adding Schema — AI Citations Barely Moved
- Kurt Fischman / Growth Marshal: Does Schema Markup Predict AI Citation?
- SSRN preprint: Does Schema Markup Predict AI Citation?
- Google Search Central: Mark Up FAQs with Structured Data