Schema Markup for AI Citations: What the Data Actually Shows

In December 2025, SearchAtlas published a study concluding that schema markup has "no effect" on AI visibility. Their data showed domains with 100% schema coverage performing identically to those with 0%.

That same month, page-level experiments from Search Engine Land, Relixir, and Agenxus documented 28-60% citation rate improvements from proper schema. Both findings are correct - and the gap between them explains why most sites get schema wrong.

Why Does One Study Say Schema Doesn't Work While Others Say It Does?

SearchAtlas measured domain-level schema coverage - what percentage of a site's URLs contain any schema markup. They grouped domains into five buckets (0%, 1-30%, 31-70%, 71-99%, 100%) and compared visibility scores across OpenAI, Gemini, and Perplexity. No statistical difference between any bucket.

Page-level studies took a different approach. Search Engine Land tested three nearly identical pages differing only in schema quality. Only the page with well-implemented schema appeared in AI Overviews - and it ranked Position 3 versus Position 8+ for the others. Relixir's 50-site study measured 28-67% citation rate improvements depending on schema type. Agenxus found 40-60% more frequent citations for content with proper HowTo and FAQ schema.

Both findings are correct. They measure different things.

SearchAtlas didn't differentiate schema type, completeness, or quality. They acknowledged this directly: "The analysis measured the presence of schema markup, not the type, completeness, or quality of structured data." Page-level studies controlled for exactly those variables and found consistent improvements.

The distinction matters for how you invest time. "Adding schema to more pages doesn't improve visibility" is true at the domain level. "Adding quality schema to your best pages improves visibility" is true at the page level. Same system, different measurements, opposite-looking conclusions.

Which Schema Types Actually Improve AI Citations?

Not all schema types carry the same weight. Research from Relixir, Agenxus, and Frase.io points to a clear hierarchy based on measured citation impact.

Tier 1: Highest impact (40-60% improvement)

Tier 2: High value (25-85% improvement)

The FAQ paradox. A peer-reviewed study of 100,000 ChatGPT prompts from SE Ranking found that pages with FAQ schema averaged fewer citations (3.6) than pages without (4.2). That looks like FAQ schema hurts you.

It doesn't. Pages that carry FAQ schema tend to be thin - support docs, product FAQs, single-sentence answers. The schema isn't the problem. The shallow content that typically accompanies it is.

FAQ schema works when the answers are substantive. Agenxus's analysis of 5,000+ FAQ pages found that FAQPage schema with 150-300 word answers generates 35-50% more citations than QAPage structure. FAQPage signals definitive, authoritative answers. QAPage's multi-answer format introduces ambiguity that reduces citation confidence.

HowTo outperforms FAQ for procedural queries. Match your schema type to the query intent, not to your content management convenience. Someone searching "how to implement schema markup" needs steps, not Q&A. HowTo schema gives AI systems the sequential structure they can extract directly.

Here's what a properly implemented FAQPage schema looks like:

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "How does schema markup improve AI citation rates?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Schema markup helps AI systems extract content with higher confidence by providing explicit structure. FAQPage schema increases citation probability by 28-50% compared to unstructured content. The key requirement is substantive answers of 150-300 words. Generic single-sentence answers underperform regardless of schema quality."
      }
    }
  ]
}

And HowTo schema for procedural content:

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to Implement Schema Markup for AI Citations",
  "totalTime": "PT1H30M",
  "step": [
    {
      "@type": "HowToStep",
      "position": 1,
      "name": "Audit existing schema",
      "text": "Run Google's Rich Results Test on your top content pages. Identify syntax errors and missing required properties. Fix critical errors first."
    },
    {
      "@type": "HowToStep",
      "position": 2,
      "name": "Add Article schema with author credentials",
      "text": "Every content page needs Article schema with headline, author linked to Person schema, datePublished, and dateModified in ISO 8601 format."
    }
  ]
}

For a deeper look at how AI engines decide which content to extract and cite, see How AI Answer Engines Actually Select Content.

How Do Different AI Platforms Handle Schema?

Each major AI platform treats schema differently. A strategy optimized for Google AI Overviews may underperform on Perplexity, and vice versa.

Google AI Overviews show the strongest schema response. In Search Engine Land's controlled experiments, a page with proper schema appeared in AI Overviews while identical pages without schema didn't appear at all. Google favors nested entity schemas - Person linked to Organization linked to Location. Build relationships between entities, not isolated labels.

ChatGPT pulls from Bing's index, so schema that improves Bing ranking indirectly improves ChatGPT citations. But ChatGPT weights content substance more than structure. Pages with 19+ statistical data points averaged 5.4 citations versus 2.8 for data-sparse content. Expert quotes pushed citation rates from 2.4 to 4.1. Schema helps, but depth drives ChatGPT more than markup.

For more on ChatGPT's retrieval mechanics, see How ChatGPT Searches Differently Than Google.

Perplexity performs real-time web searches and weights freshness above almost everything else. 76.4% of its top-cited pages were updated within 30 days. Schema combined with genuine freshness signals - dateModified, visible "Last updated" text, and sitemap lastmod - creates high citation velocity. Practitioners report citations appearing within 2 hours of publishing properly structured content.

Bing Copilot shows the most direct schema-to-citation relationship. A documented case study found that adding HowTo schema to developer setup guides increased citation share from 9% to 24% in three weeks. LocalBusiness schema also drives citations for location-based "near me" queries.

To track how schema changes affect your citation rates across platforms, see How to Measure Your AI Visibility.

What Are the Implementation Mistakes That Kill Schema's Value?

Some implementation errors reduce citation probability to zero. Others cut it by 30-80%. These failures explain why SearchAtlas found no correlation at the domain level - many sites' schema implementations are broken in ways that provide no benefit to AI systems.

Critical errors (eliminates all benefit):

• Blocking crawlers from schema pages. Schema on a noindex page or behind a disallowed URL is invisible to every AI system. Check your robots.txt and meta robots tags before anything else.
• Semantic redundancy. Using both Microdata and JSON-LD for the same data with conflicting values - price marked $299 in Microdata and $279 in JSON-LD - causes AI parsers to trust neither. Pick JSON-LD exclusively and remove legacy formats.
• Fake freshness signals. Updating dateModified without changing actual content triggers credibility penalties. AI systems detect mismatches between claimed dates and real changes. Update dateModified only for substantive edits: new data, expanded sections, updated methodology. Not typo fixes.
• Missing required properties. FAQPage schema without question-answer text. Article schema without author or date. Incomplete schema is ignored entirely.

High-impact errors (30-80% reduction):

• Client-side rendered schema. If your schema is generated by JavaScript after page load, many AI crawlers never see it. View your page source (not DevTools inspect). If the JSON-LD block isn't in the raw HTML, it's invisible to crawlers that don't execute JavaScript.
• Stale data. Outdated prices, old publication dates, dead sameAs links. Each erodes trust incrementally.
• Incorrect nesting. Product schema nested inside Article schema without proper contextual linking causes entity misclassification.

If your site relies on JavaScript rendering, see Lazy Loading and AI Visibility for how AI crawlers handle client-side content.

A properly structured Article schema with E-E-A-T signals looks like this:

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Your Article Title Here",
  "datePublished": "2026-04-16T09:00:00Z",
  "dateModified": "2026-04-16T09:00:00Z",
  "author": {
    "@type": "Person",
    "name": "Your Name",
    "jobTitle": "Your Title",
    "affiliation": {
      "@type": "Organization",
      "name": "Your Company",
      "sameAs": "https://www.linkedin.com/company/your-company/"
    },
    "sameAs": ["https://www.linkedin.com/in/your-profile/"]
  },
  "publisher": {
    "@type": "Organization",
    "name": "Your Company",
    "logo": {
      "@type": "ImageObject",
      "url": "https://yoursite.com/logo.png"
    }
  }
}

Author links to Person. Person links to Organization via affiliation. Organization has sameAs to verified external profiles. This nesting lets AI systems verify entity relationships without inference.

Where Should You Start?

Don't implement schema across your entire site at once. Broad coverage without quality control is the trap the SearchAtlas data exposes - it produces zero benefit.

Week 1-2: Foundation

Identify your 10 highest-traffic content pages. Add Article schema to each with these required properties:

• headline matching the visible H1
• author linked to a Person schema with name, jobTitle, and affiliation
• datePublished and dateModified in ISO 8601 format with timezone
• publisher linked to your Organization schema

Validate every page with Google's Rich Results Test. Fix errors before expanding.

Week 3-4: High-impact additions

Pick schema type based on query intent:

• Question-based content gets FAQPage schema with 150-300 word answers
• Procedural content gets HowTo schema with all steps, tools, and time estimates
• Homepage and landing pages get Organization schema with sameAs links to LinkedIn, Crunchbase, Wikipedia

Week 5+: Monitor and adjust

Test in actual AI platforms. Ask ChatGPT, Perplexity, and Google your target queries. See whether you're getting cited. Compare before and after.

Schema changes show in AI citation patterns faster than most SEO work. Practitioners report shifts within 2-4 weeks. For Perplexity specifically, citation velocity can change within hours of publishing structured content.

CompetLab's AI Visibility tracking monitors how ChatGPT, Claude, and Gemini mention your brand over time - so you can measure whether schema changes actually move your citation rates across platforms, not just hope they do.