Schema Markup for GEO: What JSON-LD Does for AI Citations

When Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026 and compared them with 4,000 matched pages that never did, ChatGPT citations moved by 2.2% and Google AI Mode citations by 2.4%, both within statistical noise, while AI Overview citations fell 4.6%. Schema markup for GEO is therefore not a citation lever for pages AI engines already cite. It still earns its place as JSON-LD that describes your organization, authors and content accurately, because search engines use it to understand entities, and it must match what the page visibly says.

What the Ahrefs schema study found

The study started from a striking correlation. Across 6 million URLs, Ahrefs found that pages cited by AI were almost three times more likely to carry JSON-LD than pages that were not. That is the statistic behind most "add schema to get cited" advice. The authors then tested whether adding JSON-LD actually caused more citations, using a matched difference-in-differences design that strips out platform-wide trends.

The answer was no on every platform. The 4.6% drop on AI Overviews was statistically significant but small, about 12 daily citations per page on pages that usually had hundreds, and both groups were already declining before the change. The main limit is the sample: every page already had more than 100 AI Overview citations, so the result says nothing about pages that AI engines have never picked up.

Schema markup helps machines read what a page already says. It does not make the page more worth citing.

Why AI assistants may never see your markup

Part of the explanation is mechanical. In a test cited in the same Ahrefs report, searchVIU checked whether five AI systems (ChatGPT, Claude, Perplexity, Gemini and Google AI Mode) used schema when fetching a page live. None did: every system extracted only the visible HTML. When an assistant reads your page at answer time, the facts that count are the ones written in the text.

Markup injected by JavaScript has a further problem, since most AI crawlers do not run scripts, as explained in whether AI crawlers render JavaScript. JSON-LD that only exists after a tag manager fires may never reach them at all. Shipping it in the server-rendered HTML is the safe default.

Where schema still does real work

Schema matters earlier in the chain, when search engines build their understanding of entities. Microsoft's Fabrice Canel said at SMX Munich in March 2025 that schema markup helps Microsoft's LLMs understand content, which matters for Copilot. Google's position is more cautious: its guidance on AI features says no special optimization is required for AI Overviews and AI Mode, and that the usual SEO fundamentals apply.

The lasting value is identity. Organization markup with sameAs links ties your site to your LinkedIn, Crunchbase, Wikidata and review profiles, which helps systems resolve which company a name refers to and feeds the knowledge graph that engines draw on. Our guide to Wikipedia, Wikidata and brand entities covers the other half of that work.

Which JSON-LD types to keep for GEO

Four types cover most sites. Organization goes on the homepage, with name, URL, logo, a factual one-sentence description and sameAs links. Article or BlogPosting goes on editorial pages, with headline, author as a Person with a profile URL, datePublished and a dateModified that changes only when the content does. Person goes on author pages. Product with Offer and Review markup goes on product pages, where it still drives visible results in classic search.

FAQPage needs a more careful call. Google stopped showing FAQ rich results on 7 May 2026, and FAQPage remains a valid Schema.org type that can stay on pages with a visible FAQ. What a question-and-answer section does for AI answers is a separate question, covered in our analysis of FAQ schema and AI search. Whatever the type, the markup should repeat facts that appear in the visible text, never add facts that exist only in the code.

How to check schema, and how to test its effect

Validate syntax with Google's Rich Results Test or the Schema.org validator, then check consistency by hand: the company name, description and author in the markup should match the page and your external profiles word for word. For broader technical checks, our GEO audit guide lists what AI crawlers need to read a page.

Measuring the effect on AI answers needs a control group, which is the most useful lesson of the Ahrefs work. A plain before-and-after comparison showed AI Mode citations up 43% on the pages that added schema, but the matched control pages had grown almost as fast, and the true effect shrank to 2.4%. Track the same prompts on pages you changed and on comparable pages you left alone, using the method in how to track brand mentions in ChatGPT, Gemini and Perplexity, before crediting any change in citation rate to the markup.