Schema Markup Impact on GEO Success: A 2026 Study

In generative engine optimization (GEO), schema markup is often treated as something that matters a great deal for getting a brand recommended by ChatGPT, Claude, Gemini, and Google AI Overviews. That claim is frequently overstated. Our independent analysts set out to test it.

We assembled a panel of 2,387 content-publishing domains across seven site categories; retrieved the homepage and up to four high-traffic content or product URLs monthly from January 2025 through June 2026; checked each page for schema markup, including the common JSON-LD format, and recorded which schema types were present; and separately scored how clearly the visible page laid out its facts. We measured commercial recommendation share across those four AI surfaces. We also compared sites that newly added schema during the study to similar sites that did not. We tracked 612 domains that added schema for the first time during the window against matched domains that did not.

The Schema Correlation Claim, Tested

This section compares sites with schema to sites without it, first in a simple cut and then after matching sites on page clarity and category. Sites that use schema also tend to publish clearer pages, run larger catalogs, and invest more in technical SEO, so a simple with-versus-without comparison overstates what the markup itself is doing.

• At first glance, sites with schema show up in commercial AI recommendations 7.2 percentage points more often than sites without it.
• When we compare sites with similar page clarity and in the same category, that gap shrinks to 0.7 points.
• That smaller gap is the same for regular sites and large product catalogs. For ecommerce, schema's clearer effect shows up later in how complete the product details are in AI answers.

What Happens When a Site Adds Schema

This section compares domains that added schema for the first time between January 2025 and June 2026 with matched domains that never added it. Of the 612 treated domains, 468 were non-catalog publishers and 144 were large product-catalog sites. Outcomes are measured as the change in share from the 90 days before the first observed schema to the 90 days after, relative to the matched control.

• On non-catalog sites, the change after adding schema sits between -0.1 and +0.2 percentage points across ChatGPT, Claude, Gemini, and AI Overviews. That is effectively no gain in commercial recommendations.
• Catalog sites show small positive lifts, peaking at +0.8 points on ChatGPT, still under one percentage point of recommendation share on most surfaces.
• Informational citation rate moves even less than recommendation share on non-catalog domains (+0.1 pts relative). The weak result holds beyond buyer-intent prompts alone.

Recommendation and Entity Results by Schema Type

This section compares the main schema families in the panel rather than treating "has schema" as one switch. Being recommended in a commercial AI answer and showing up in an entity or Knowledge Panel result are measured separately, because Organization markup and commercial recommendation are different jobs.

• Product and Offer domains lead on commercial recommendation share at 20.4%. That is expected for catalog sites and does not by itself prove the schema caused the lift.
• Organization and WebSite markup without Product sits at 12.6% recommendation share, close to the structure-matched no-schema baseline of 13.4%, while entity appearance rises to 15.2%.
• Article, FAQ, and HowTo markup stays near the no-schema baseline on recommendation share once page clarity is controlled. That matters when FAQ schema is sold as a GEO lever.

Product Attribute Accuracy With Schema, Tables, or Prose

This section isolates product pages and asks a narrower question: when an AI system answers about a specific product, how often are price, availability, brand, and key attributes correct. The comparison is Product and Offer schema versus clear HTML product tables versus unstructured prose.

• Clear HTML product tables without schema reach 79.1% attribute accuracy, within 2.3 points of pages that carry both tables and Product/Offer schema.
• Schema on an otherwise unstructured product page still beats pure prose (72.6% against 58.3%). Markup helps when the visible page does not already show the facts.
• The practical ecommerce reading is that explicit attributes matter. Schema is one way to package those attributes; clear tables do much of the same work.

Clear Structure Without Schema

This section cross-tabulates the panel by visible page clarity and whether schema is present. Structure scoring covered heading hierarchy, on-page attribute tables or definition lists, and consistent product or service facts in HTML, whether or not schema was present.

• Domains with high visible structure and no schema reach a 16.8% commercial recommendation share, nearly matching high-structure domains that also carry schema (17.4%).
• Schema-heavy pages with low structure scores underperform high-structure pages without markup by 5.9 points on recommendation share.
• Citation rate follows the same ordering. Facts that are easy to read on the page matter more for both recommendation GEO and informational citation than the presence of a schema block.

What Predicted GEO Success

This section ranks predictors of commercial recommendation share across the full 2,387-domain panel. Coefficients come from a category-stratified model that includes content, offsite, crawl, structure, and schema variables together, so each figure is the association after the others are in the model.

• Unique substantive content and third-party mentions dominate the model. Together they account for most of the explained variation in recommendation share.
• Crawlability and clear visible structure remain material after content and mentions are controlled. That fits systems that must fetch and parse a page before they can recommend it.
• Any schema and schema type count sit near noise at 0.06 and 0.04. Once the facts are already clear on the page, schema is optional.

A separate study posted to arXiv on September 28, 2026, by Ben Moore and Liam Dunne, reached a similar result with different data. The authors joined about 2 million citations from ChatGPT, Claude, Google AI, and Gemini, collected over six months, to 10,000 crawled pages from 19 B2B SaaS workspaces. Structured data, FAQ blocks, and Core Web Vitals looked helpful in the pooled numbers, and those effects reversed or fell to zero once the authors compared pages within the same domain. The strongest page-level predictor was word overlap between a page and the full set of prompts each workspace tracked. A domain's overall standing in AI answers carried about six times the weight of any page feature other than that overlap. That study counted citations, not commercial recommendations.

Conclusion

After controlling for clear structure and site category, adding schema raised commercial recommendation share by less than one percentage point on non-catalog sites. In the 612-domain comparison, non-catalog relative lifts across ChatGPT, Claude, Gemini, and Google AI Overviews ranged from -0.1 to +0.2 points. On product pages, clear HTML attribute tables without schema reached 79.1% attribute accuracy in AI answers, against 81.4% for pages with both tables and Product/Offer schema, and 58.3% for unstructured prose alone. High-structure domains with no schema recommended at 16.8%, while schema-heavy, low-structure domains recommended at 10.9%.

AI systems need facts they can extract. Schema is one optional way to mark those facts. Clear structure, crawlability, distinctive content, and third-party mentions predicted GEO success in this panel. Schema, once those were controlled, did not move recommendation share in a meaningful way. Recommendation GEO and entity or Knowledge Panel work should be split: Organization and sameAs markup may clarify an entity, and that is a different job from being recommended for a commercial purchase prompt.

Our Methodology

Our independent analysts assembled a panel of 2,387 content-publishing domains across seven categories: ecommerce and product catalogs, B2B and corporate service sites, news and magazines, forums and user-generated content, reference, stock media, and government and education. For each domain they retrieved the homepage and up to four high-traffic content or product URLs every month from January 2025 through June 2026.

Schema was detected primarily through JSON-LD blocks and recorded by schema.org type. Visible clear structure was scored separately from markup, using heading hierarchy, on-page attribute tables or definition lists, and consistent product or service facts in HTML. Commercial recommendation share is the share of buyer-intent prompt runs in which the brand or a URL from the domain appeared across ChatGPT, Claude, Gemini, and Google AI Overviews. Informational citation rate uses category and brand prompts on the same surfaces. That comparison covers 612 domains that added schema for the first time during the study window, each matched to a control domain in the same category with a similar pre-period recommendation level that did not add schema.

Robots.txt and CDN blocks limit what a research crawl can see, and schema that is injected only after render can be missed when a fetch does not execute scripts. This study measures recommendation and citation GEO. It does not measure classic rich-result click-through rates, for which structured data remains a separate and still useful eligibility path.

The Sources list below is external context only. Those papers and documents are not part of the 2,387-domain panel, and their sample sizes and outcomes are not pooled into the tables above.

Sources

• Google Search Central. Optimizing your website for generative AI features on Google Search. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide (Structured data is not required for generative AI search, and there is no special schema.org markup required for those features; structured data remains useful for rich results.)
• Google Search Central. AI features and your website. https://developers.google.com/search/docs/appearance/ai-features (No additional technical requirements beyond ordinary Search eligibility for supporting links in AI Overviews and AI Mode.)
• Google Search Central. Introduction to Product structured data. https://developers.google.com/search/docs/appearance/structured-data/product (Product markup and Merchant Center feeds support shopping-related Search experiences; cited here for shopping-surface nuance, not as a generic GEO citation lever.)
• Schwartz, Barry. Google Again Says Structured Data Does Not Make Your Site Rank Better. Search Engine Roundtable, April 13, 2025. https://www.seroundtable.com/google-structured-data-ranking-39232.html (Reporting John Mueller: structured data will not make a site rank better; it is used for documented search features.)
• Linehan, Louise, and Xibeijia Guan. We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved. Ahrefs, May 11, 2026. https://ahrefs.com/blog/schema-ai-citations/ (External matched difference-in-differences study of pages that added JSON-LD; cited for context only, not as TGA panel data.)
• Moore, Ben, and Liam Dunne. What Drives Citations in Production Large Language Models? An Observational Multi-Method Study of Two Million AI Citations Across Ten Thousand Web Pages. arXiv, September 28, 2026. https://arxiv.org/abs/2609.35077 (Observational study of about 2 million citations across four AI engines; FAQ, structured data, and Core Web Vitals effects collapsed to zero or reversed under domain fixed effects.)
• searchVIU. Schema Markup and AI in 2025: What ChatGPT, Claude, Perplexity & Gemini Really See. Updated December 2, 2025 (tests October 26-30, 2025). https://www.searchviu.com/en/schema-markup-and-ai-in-2025-what-chatgpt-claude-perplexity-gemini-really-see/ (Direct-fetch experiment: prices present only in JSON-LD were not extracted by the tested systems; visible HTML prices were found more often.)
• de Segonzac, Olivier. Inside ChatGPT's retrieval stack: The index, cache, and pages it actually reads. Search Engine Land, August 17, 2026. https://searchengineland.com/chatgpt-retrieval-stack-index-cache-pages-485036 (Reports that ChatGPT's HTML-to-Markdown cache strips scripts, iframes, and JSON-LD on that read path.)
• Aggarwal, Pranjal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande. GEO: Generative Engine Optimization. arXiv, November 16, 2023. https://arxiv.org/abs/2311.09735 (Content rewrite tactics such as statistics, quotations, and citations improved generative visibility; schema markup was not among the tested methods.)