Your Schema Markup Is Probably Hurting Your AI Visibility
A 730-citation study found generic schema performs 18 points worse than no schema at all. Here's what actually works for AI citations.
Your Schema Markup Is Probably Hurting Your AI Visibility
Every SEO checklist says the same thing: add schema markup. It helps search engines understand your content. It’s good for rich snippets. And now, with AI search, it’s supposed to help you get cited by ChatGPT and Google AI Overviews too.
There’s just one problem. A new study of 730 AI citations found that generic schema markup doesn’t just fail to help. It actually makes things worse.
Pages with minimally-populated schema got cited 41.6% of the time. Pages with no schema at all? 59.8%. That’s an 18-point penalty for doing what every best-practice guide tells you to do.
The catch: attribute-rich schema, the kind with fully populated pricing, ratings, and specifications, hit a 61.7% citation rate. So schema can help. But only if you do it right. And most sites aren’t.
The Study That Changes the Conversation
The research comes from Growth Marshal, published in February 2026. The team analyzed 1,006 pages across 75 commercial queries, tracking which pages got cited in ChatGPT and Gemini responses.
They tested three scenarios:
| Schema Type | Citation Rate | vs. No Schema |
|---|---|---|
| Attribute-rich (Product, Review with full data) | 61.7% | +1.9 points |
| No schema at all | 59.8% | baseline |
| Generic/minimal schema | 41.6% | -18.2 points |
The generic schema category includes the usual suspects: Article, Organization, and BreadcrumbList markup with minimal attributes filled in. The type of schema that most CMS plugins generate automatically. The type sitting on millions of websites right now.
The attribute-rich category means Product and Review schema with populated pricing, ratings, specifications, and other detailed fields. Not just the schema type, but complete, thorough data within it.
Why Generic Schema Backfires
This seems counterintuitive. How can adding structured data make things worse?
The Growth Marshal team found a statistical answer: generic schema had an odds ratio of 0.678 (p = .296) for citations. Not statistically significant on its own, but the pattern is clear when you look at the mechanism.
Here’s what’s likely happening. When an AI system encounters a page with schema markup, it uses that structured data as a signal about the page’s information quality. If the schema is sparse (an Article type with just a headline and publish date, or an Organization type with only a name), it signals that the page’s structured information is thin. The AI weighs that signal alongside everything else it knows about the page.
A page with no schema doesn’t send that negative signal. The AI evaluates the content on its own terms. But a page with half-empty schema actively communicates: “We started to describe this content formally, and there wasn’t much to say.”
Think of it like a product listing. A product page with a price, 47 reviews, detailed specs, and comparison data looks authoritative. A product page with just a name and category looks like a placeholder. No product page at all? At least you’re not setting low expectations.
The Authority Gap Makes It Worse
The penalty hits harder if you’re not already a well-known domain. The study broke results down by Domain Rating:
Lower-authority domains (DR < 60):
- Attribute-rich schema: 54.2% citation rate
- Generic schema: 31.8% citation rate
- Gap: 22.4 percentage points
High-authority domains (DR > 75):
- Minimal difference between schema types
- Authority signals dominate citation decisions regardless
This makes sense. High-authority domains have enough trust signals that schema quality becomes a minor factor. But for mid-tier and smaller brands, schema quality is one of the few signals that can differentiate you. And right now, generic schema is actively working against them.
If your domain authority is below 60, this isn’t a nice-to-have optimization. It’s a visibility leak you’re probably not tracking.
What Attribute-Rich Schema Actually Looks Like
The difference between schema that helps and schema that hurts comes down to completeness. Here’s a concrete example.
Generic schema (hurts):
{
"@type": "Product",
"name": "Project Management Tool",
"description": "A tool for managing projects."
}Attribute-rich schema (helps):
{
"@type": "Product",
"name": "ProjectFlow Pro",
"description": "Project management platform for teams of 5-500",
"brand": { "@type": "Brand", "name": "ProjectFlow" },
"offers": {
"@type": "AggregateOffer",
"lowPrice": "12",
"highPrice": "49",
"priceCurrency": "USD",
"offerCount": "3"
},
"aggregateRating": {
"@type": "AggregateRating",
"ratingValue": "4.6",
"reviewCount": "1847"
},
"category": "Business Software > Project Management",
"operatingSystem": "Web, iOS, Android"
}The second version gives AI systems concrete, comparable data points. Pricing tiers, review counts, platform availability, and team size context. When ChatGPT is deciding which project management tools to mention, it can actually extract useful information from this schema.
The Eight Schema Types That Matter
Not all schema types carry equal weight for AI citations. Based on the Growth Marshal findings and additional research from SE Ranking and SALT.agency, here are the types worth investing in:
| Schema Type | AI Citation Value | Key Attributes to Populate |
|---|---|---|
| Product | High | Price, ratings, specs, availability, brand |
| Review / AggregateRating | High | Rating value, review count, author |
| FAQPage | Medium-High | Question/answer pairs (real ones, not filler) |
| Organization | Medium | Founded date, employees, location, sameAs links |
| Person | Medium | Job title, affiliation, credentials, sameAs links |
| Article | Medium | Author, datePublished, dateModified, citations |
| LocalBusiness | Medium | Hours, address, geo, price range |
| Service | Medium | Provider, area served, pricing, offers |
The critical detail: every one of these types can be either helpful or harmful depending on how thoroughly you populate the attributes. An FAQPage with two generic questions performs worse than no FAQ schema. An FAQPage with twelve specific, well-answered questions helps.
The sameAs Problem
Here’s a related gap most sites ignore. The SALT.agency analysis of 107,352 URLs found that fewer than 4% include sameAs links to Wikidata or Wikipedia entities.
Why this matters: sameAs is how schema connects your organization or product to the broader knowledge graph. When your Organization schema includes "sameAs": ["https://www.wikidata.org/wiki/Q12345", "https://en.wikipedia.org/wiki/Your_Company"], you’re giving AI systems an unambiguous entity reference.
Without it, “Acme” could be a coyote supply company, a software firm, or a cleaning brand. AI systems have to guess from context. With a Wikidata link, there’s no ambiguity. This is especially useful for brands with common names or brands competing in crowded categories.
If your company has a Wikipedia page, link to it via sameAs. If it doesn’t, consider whether a Wikidata entry is warranted. Our earlier analysis found that Wikipedia presence correlates with AI visibility, and proper entity linking through schema is part of that connection.
AI Overviews Are Getting Less Predictable
This schema research arrives at an interesting moment. A February 2026 Ahrefs study of 863,000 keywords found that only 38% of pages cited in Google AI Overviews now rank in the traditional top 10 for the same query. That’s down from 76% just seven months earlier.
Where are the other citations coming from? Roughly 31% from positions 11-100, and another 31% from pages that don’t rank in the top 100 at all. Google’s Gemini 3 model is running what the industry calls “fan-out queries,” splitting a user’s question into multiple sub-queries and pulling sources from across those results.
This matters for schema strategy because it means you don’t need to be a page-one result to get cited. But you do need your content to be machine-readable when AI systems find it through those sub-queries. Attribute-rich schema gives AI systems structured, extractable data to work with, even when your page shows up in a long-tail sub-query where the AI has dozens of candidates to choose from.
As we covered in our analysis of Google rankings and AI visibility, the relationship between traditional search position and AI citations is getting more complex. Schema quality is becoming one of the tiebreakers.
A Schema Audit in Five Steps
Here’s a practical process to fix your schema implementation:
1. Inventory your current schema. Use Google’s Rich Results Test or Schema Markup Validator on your top 20 pages. List every schema type and count the populated attributes for each.
2. Score attribute completeness. For each schema instance, calculate how many optional attributes you’ve filled versus how many exist for that type. If you’re below 50% completion, you’re probably in “generic” territory.
3. Kill or complete. For any schema instance with fewer than five populated attributes, either remove it entirely or fill it out. Based on the Growth Marshal data, removing thin schema is better than leaving it. That’s not a typo. Deleting your sparse schema will likely improve your AI citation rate.
4. Add entity links. For Organization and Person schema, add sameAs links to Wikipedia, Wikidata, LinkedIn, and Crunchbase where applicable. This costs almost nothing and closes the entity ambiguity gap.
5. Prioritize Product and Review schema. If you sell products or services, these two types showed the strongest citation advantage when properly populated. Make sure pricing, ratings, and specifications are complete and current.
What This Means for Your AI Visibility Strategy
The schema findings reinforce a broader pattern in AI search optimization: half-measures don’t work. Partial content updates, minimal structured data, surface-level optimization. AI systems are surprisingly good at distinguishing between thorough work and checkbox compliance.
The practical takeaway is clear. Audit your schema today. If you have sparse, auto-generated schema markup on your pages, you have two options: make it complete or delete it. Both are better than what you’re doing now.
For brands with domain authority under 60, this is one of the highest-leverage technical changes available. A 22-point citation rate gap between generic and attribute-rich schema is significant for any channel, and it’s fixable in a week.
Want to know what AI platforms say about your brand? Try RivalHound free and find out.