Schema markup is necessary infrastructure for any site serious about search visibility — but on the current evidence, it is not a direct lever for AI citations. The standard advice that schema markup will get your content into ChatGPT, Perplexity, or Google AI Overviews is ahead of the data. Recent rigorous testing has overturned a lot of what the SEO industry has been saying about schema for AEO. Here’s what the research actually shows and what to implement anyway.
Key takeaways
- Three recent empirical tests — Ahrefs (May 2026), searchVIU, and Search Atlas — found no direct relationship between adding schema markup and AI citation rates on already-visible pages.
- AI engines (ChatGPT, Claude, Perplexity, Gemini, Google AI Mode) extract only visible HTML during real-time retrieval. JSON-LD, hidden Microdata, and hidden RDFa are ignored at fetch time.
- Microsoft has confirmed schema helps Copilot. Google’s John Mueller maintains that schema is not a ranking factor but helps systems understand content. The signals diverge by platform.
- What actually drives AI citations: content depth, topical authority, traditional search performance, entity strength, extractable answers, and original information.
- Schema is still worth implementing as foundational SEO and entity infrastructure — but it is not a citation magic spell, and the agencies selling it that way are ahead of the evidence.
The standard advice on schema for answer engines is ahead of the evidence
Most articles you’ll find on schema markup for AI search make confident claims — schema increases citation rates by 36%, 40%, even 73%. Search results have been dominated by this framing for the better part of two years. The narrative is consistent: implement FAQPage, Organization, Article, and HowTo schema and AI engines will reward you with citations.
The problem is that almost none of those claims survive scrutiny. Most “studies” come from agencies and software vendors that sell schema implementation as a service or compete with tools that do. As of May 2026, there are no peer-reviewed studies on schema’s impact on AI search visibility. The 2024 GEO study from Princeton and Georgia Tech that gets cited constantly in schema discussions did not actually test schema — it tested in-content strategies like adding citations and statistics. Those worked. Schema was never part of the experiment.
In the last six months, three separate empirical studies have all found that schema markup does not meaningfully drive AI citations. The most rigorous of those dropped on May 11, 2026.
What the May 2026 evidence actually shows
The most-cited recent test came from Ahrefs. Louise Linehan and Xibeijia Guan tracked 1,885 web pages that added JSON-LD schema markup between August 2025 and March 2026. They matched those pages against a control group of 4,000 similar pages that never added schema, with comparable pre-treatment citation levels. They measured citation changes across Google AI Overviews, Google AI Mode, and ChatGPT for 30 days before and 30 days after schema implementation.
The results: Google AI Mode citations rose 2.4%. ChatGPT citations rose 2.2%. Google AI Overviews citations declined 4.6%. The first two figures are statistically indistinguishable from zero — random noise across thousands of URLs. The AI Overviews decline is statistically significant, with Ahrefs reporting the odds of seeing a gap that large by pure chance at roughly 1 in 2,500.
Ahrefs cross-referenced a separate experiment from searchVIU, which tested whether five major AI systems — ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode — actually used schema markup when fetching pages in real time. The finding was uniform: during direct retrieval, every system extracted only visible HTML content. JSON-LD, hidden Microdata, and hidden RDFa were all ignored.
A separate study from Search Atlas examined schema coverage against LLM citation rates across OpenAI, Gemini, and Perplexity. It found no correlation. Domains with comprehensive schema performed no better than domains with minimal or no schema.
Three caveats are worth sitting with. The Ahrefs study only measured pages that were already being cited heavily by AI — every page in the dataset had received 100 or more AI Overview impressions before treatment. The effect of schema on pages not yet visible to AI is unmeasured by this methodology. The study also did not isolate per-type effects (FAQPage versus Organization versus Product, for example), did not measure time windows longer than 30 days, and did not test JavaScript-injected markup.
So the honest summary: for already-AI-visible pages, in the 30-day window after adding schema, JSON-LD does not appear to drive citations. Whether schema helps a new, unindexed page get discovered or initially cited remains an open question.
What major platforms have actually said about schema and AI
The schema-for-AEO narrative didn’t appear from nowhere. It rests on a handful of official platform statements that are technically accurate but routinely overinterpreted.
In March 2025, Microsoft’s Fabrice Canel confirmed at SMX Munich that schema markup helps Microsoft’s LLMs understand content for Copilot. That is the strongest pro-schema platform statement on record. In April 2025, Google’s Search team published guidance noting that structured data provides an advantage in search results — in search results, not specifically in AI citations.
John Mueller, Google’s Search Advocate, has stated multiple times that structured data is not an algorithmic ranking factor. In November 2025, Mueller clarified on Reddit that schema is not being phased out broadly. He noted that specific markup types are deprecated regularly while the core types remain valuable, framing it as markup types coming and going while a small set of essential ones is worth holding onto. Google deprecated several structured data types starting January 2026, including Practice Problem and most Dataset uses. FAQ rich results were further restricted on May 7, 2026 — they now appear only for government and health sites.
The pattern matters. Google’s statements consistently say schema helps systems understand content, not that it drives AI citations. Mueller has been careful to distinguish ranking from understanding. Microsoft’s statement is more direct about Copilot, but the empirical tests of Perplexity, ChatGPT, and Google’s own AI products show those systems do not extract from JSON-LD during real-time retrieval.
The most coherent reading of all of this: schema may contribute to how AI systems build their internal knowledge during training and indexing. It does not appear to influence which sources get cited during answer generation.
Why schema is still worth implementing
If schema does not directly drive AI citations, why bother with it at all?
Four real reasons.
Traditional SEO still benefits from structured data. Rich results, knowledge graph signals, and entity disambiguation all run through schema. The practical case from Google’s own documentation is consistent over time: schema helps systems understand content, and content that gets understood properly tends to surface better. Rich results have narrowed — many types are deprecated or restricted — but the ones that remain still drive real CTR improvements.
Microsoft Copilot does use schema markup. Canel’s confirmation was specific. If Bing and Copilot are meaningful surfaces for your business, schema is contributing to that visibility even if it is not contributing to Perplexity or ChatGPT citations. The fragmentation of AI search means optimizing for the platforms that confirm schema use is still worth doing.
Entity disambiguation is increasingly important as AI systems build their knowledge graphs. Organization schema with sameAs links to authoritative profiles (Wikidata, LinkedIn, Crunchbase, official social) signals which specific entity your site represents. AI systems that can confidently resolve who you are are more likely to cite you accurately when they do cite you. That citation accuracy matters as much as citation frequency.
Schema may help during initial discovery and indexing for pages not yet AI-visible. The Ahrefs methodology cannot answer this question — it only tested already-cited pages. The downside risk of implementing schema correctly is essentially zero. The upside, on this specific dimension, is unmeasured but plausible.
So the conclusion holds in both directions. Schema is necessary as part of a strong SEO foundation. It is not a magic spell to ensure citations or mentions. Those are different claims and the evidence supports the first one, not the second.
The schema types worth implementing in 2026
Here are the types we implement at KDGS, and the rationale for each that survives the current evidence.
| Schema type | What it signals | AI utility on current evidence | When to implement |
|---|---|---|---|
| Organization | Brand entity, identity, contact | Strong — entity disambiguation | Once on homepage, referenced via @id elsewhere |
| Article | Content type, authorship, recency | Moderate — extraction context | Every long-form content page |
| Person | Author identity, E-E-A-T | Moderate — entity resolution for authors | All bylined content |
| FAQPage | Q&A structure | Contested — possible extraction aid | Pages with genuine Q&A content |
| HowTo | Step-by-step format | Low to moderate — content type signal | Step-by-step instructional content |
| Product | Product entity, price, ratings | Strong for ecommerce surfaces | Ecommerce product pages |
| LocalBusiness | Location, hours, contact | Moderate — local AI recommendations | Service-area businesses with physical locations |
| BreadcrumbList | Site structure | Low — navigation context | All non-homepage pages |
| Speakable | Voice-suitable content | Limited evidence | Skip unless targeting voice |
| Practice Problem | (Deprecated January 2026) | None | Remove |
A few notes on the highest-leverage types.
Organization schema is the foundational entity declaration. Implement it once, correctly, on the homepage, and reference it from page-specific schema via @id. A minimal but complete implementation:
{
"@context": "<https://schema.org>",
"@type": "Organization",
"@id": "<https://yoursite.com/#organization>",
"name": "Your Company",
"url": "<https://yoursite.com>",
"logo": "<https://yoursite.com/logo.png>",
"sameAs": [
"<https://www.linkedin.com/company/yourcompany>",
"<https://www.crunchbase.com/organization/yourcompany>",
"<https://www.wikidata.org/wiki/QXXXXXXX>"
],
"contactPoint": {
"@type": "ContactPoint",
"telephone": "+1-XXX-XXX-XXXX",
"contactType": "customer service"
}
}
Article schema tells systems this is narrative content, attributes it to a named author with a publication date, and supports the freshness signal Google has explicitly confirmed:
{
"@context": "<https://schema.org>",
"@type": "Article",
"headline": "Your Article Title",
"author": {
"@type": "Person",
"@id": "<https://yoursite.com/team/author-name/#person>"
},
"publisher": {
"@type": "Organization",
"@id": "<https://yoursite.com/#organization>"
},
"datePublished": "2026-05-22",
"dateModified": "2026-05-22",
"image": "<https://yoursite.com/article-image.jpg>"
}
Person schema for authors connects content to a named expert. Include name, jobTitle, sameAs links to authoritative profiles, and worksFor referencing the Organization entity. E-E-A-T signals work better when they are machine-readable, not just inferred from a byline.
FAQPage schema is the contested type. Rich results in the SERP are effectively gone for almost all sites since May 7, 2026. But the markup itself still has a purpose — single-agency tests have suggested it continues to feed AI citation extraction even if the Ahrefs aggregate did not pick up the effect. Implement it where you genuinely have Q&A content. Do not manufacture FAQ blocks just to add the schema. Google has specifically flagged inauthentic FAQ implementations.
The schema worth skipping in 2026
A few categories of structured data that get recommended in agency content but don’t earn their implementation cost.
Speakable schema. Marketed for voice assistant optimization. Limited evidence of actual impact. Skip unless you are specifically targeting voice search and have data showing it matters for your audience.
PracticeProblem, Dataset (general use), and other January 2026 deprecations. Google has retired support. Stop maintaining them. Remove them from existing pages.
Aggressive nested schema for the sake of nesting. Some guides recommend connecting every entity to every other through @id references. The signal-to-noise ratio drops fast. Implement Organization, Article, and Person with clean cross-references. Don’t engineer a complete graph.
Schema on pages where it doesn’t match the primary content. Google’s March 2026 core update specifically penalized this — schema describing peripheral or supplementary content on pages whose primary purpose was something else. Schema must match what the page is actually about.
What actually drives AI citations
If schema isn’t the lever, what is?
Topical authority. Pages on sites that have demonstrably covered the topic in depth, with internally linked cluster content, get cited more often than isolated articles on otherwise unfocused sites.
Traditional search performance. Pages that rank well in Google’s organic results are more likely to be in the candidate pool that AI engines draw from. Google AI Overviews in particular pull primarily from the existing organic index.
Entity strength. Whether the AI system can confidently resolve who or what the source is. This is where Organization and Person schema do contribute — they make entity resolution easier, which makes citation more likely.
Extractable answers. Direct answers in the first sixty words of a section, question-formatted H2s, claim-style sentences that survive being lifted from context. AI engines extract sentences, not paragraphs. For more on how to structure content for AI extraction, see our breakdown of answer engine optimization.
Original information. Unique data, primary research, named sources, specific statistics. AI systems prefer to cite sources that contain information not available everywhere else.
Recency. dateModified and datePublished signals matter. Stale content gets cited less, and AI engines often prefer recent sources for time-sensitive queries.
Of these six, only entity strength has a direct schema component. The rest are won through content strategy, topical depth, editorial discipline, and original work.
Implementation and validation
JSON-LD in the document head remains the preferred format for all schema. Microdata and RDFa work but JSON-LD is what platforms have standardized on and what their documentation assumes.
Two validation tools are worth using.
Google’s Rich Results Test (search.google.com/test/rich-results). Checks whether your markup is eligible for any of Google’s rich result types. Useful for confirming you have not broken anything, but limited in scope — it only validates types Google supports for SERP features.
Schema Markup Validator (validator.schema.org). More forgiving and broader. Validates against the full schema.org vocabulary regardless of whether Google supports the type for rich results. Use this to catch errors in types beyond Google’s rich-result set.
Common implementation mistakes worth catching before they go to production:
- Schema describing content that is not visible on the page. AI systems and Google both check for consistency between markup and visible content. Mismatch is a red flag.
- Outdated dateModified on content you have actually updated. The freshness signal only works if it is accurate.
- Aggressive FAQ schema on what is actually marketing copy. Google has specifically called this out.
- Multiple conflicting Organization schemas across a site. Pick one canonical version and reference it via @id from page-specific schema.
- Schema injected via JavaScript that doesn’t execute before crawl. JavaScript-rendered markup is inconsistently picked up. Server-side rendering or static markup is more reliable.
Frequently asked questions
Do I need to add schema to every page?
No. Implement Organization once on the homepage. Add Article or BlogPosting on every long-form content page. Add FAQPage, HowTo, Product, or LocalBusiness only on pages where the content genuinely matches those types. Schema on pages where the markup does not match the primary content can hurt you under Google’s March 2026 update.
Will schema get me into AI Overviews?
Probably not directly. The most rigorous May 2026 test (Ahrefs, 1,885 pages) found no citation lift on AI Overviews — actually a small statistically significant decline — and no lift on AI Mode or ChatGPT. Schema may still help during indexing or for not-yet-cited pages, but it is not a reliable lever for AI Overviews citation specifically. Topical authority, content depth, and traditional ranking performance matter more.
Can incorrect schema hurt me?
Yes. Mismatched schema — markup describing content that is not actually on the page — can trigger manual actions in Google Search Console and suppress rich result eligibility. Google’s March 2026 update specifically penalized sites using schema on peripheral content. Validate before deploying and audit existing schema for content-markup alignment.
Which schema validator should I use?
Both. Google’s Rich Results Test confirms eligibility for Google-specific rich results. The Schema Markup Validator at validator.schema.org is the broader check against the full schema.org vocabulary. Run both before pushing to production. If they disagree, the schema.org validator is more permissive — use the Rich Results Test as the binding check for anything Google supports.
Kim Schmutzler is the founder and director of SEO at KDGS Works, a content and SEO agency in Overland Park, Kansas. Implement schema markup as part of a complete SEO and AEO program at kdgsworks.com.
