Schema Markup for AI Search: What the Data Says in 2026

Schema markup for AI search is structured data, almost always in JSON-LD format, that describes a page so machines can read it without guessing. It helps Google understand entities and qualify pages for rich results. What it does not do, on the evidence, is lift AI citations: a 2026 Ahrefs study of 1,885 pages found adding schema produced no meaningful gain in Google AI Mode or ChatGPT, and a small decline in AI Overviews.

That finding matters because schema is sold hard as an AI visibility lever, and teams pour hours into markup expecting citations to follow. The data tells a more useful story about where structured data helps and where your effort actually moves the needle. This piece walks the evidence and lands on a plan you can run.

Schema markup for AI search, defined

Schema markup for AI search is code you add to a page, almost always in JSON-LD, that labels what the page is about so search engines and language models can parse it cleanly. It maps your content to shared Schema.org vocabulary: an article, a product, an organization, a set of steps. Google reads that markup to understand entities and relationships and to qualify pages for rich results (Google Search Central).

The confusion in 2026 is about what schema buys you in the AI layer specifically. Structured data earns its keep for machine readability, and that job is real. The claim worth testing is whether it moves the needle on getting quoted in AI answers. Our answer engine optimization work starts from evidence rather than assumption, and the 2026 AEO playbook sets the wider context this piece narrows into.

The evidence on schema and AI citations

Adding schema to a page that is already visible does close to nothing for AI citations. Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, matched each against control pages that never added it, and measured citations across Google AI Overviews, AI Mode, and ChatGPT. The result: AI Mode rose 2.4 percent, ChatGPT rose 2.2 percent (both statistically indistinguishable from zero), and AI Overviews fell 4.6 percent (Ahrefs). Four separate statistical tests pointed the same way.

There is a correlation here that gets misread. In a first pass across 6 million URLs, pages cited by AI were almost three times more likely to carry JSON-LD, and 53 percent of AI-cited pages run schema (Ahrefs). That gap reflects the kind of site that adds schema: technically maintained, authoritative, well linked. Strip schema out and those signals still carry the page. It echoes the lesson in our guide to off-page SEO for AI search, where the citation follows the authority rather than the tag.

Visible content is what AI crawlers extract

During live retrieval, the major AI systems read the visible page and skip the hidden markup. An experiment by searchVIU, cited in the Ahrefs study, tested ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode fetching a page in real time; every system pulled only visible HTML content, while JSON-LD, hidden Microdata, and hidden RDFa were ignored (searchVIU, via Ahrefs). When a model reads the words on the page to build its answer, the words are where your effort belongs.

The practical takeaway lands on your copy. State facts plainly, answer the question early, and structure sections so a machine can lift a clean claim. A machine-readable content foundation supports this, which is why we treat structured data as one layer alongside the plain-language signals in our llms.txt guide rather than a substitute for them.

Structured data still earns its place

Schema still does real work outside AI citations, and dropping it would be a mistake. It qualifies pages for rich results in classic search, feeds Google's entity understanding and Knowledge Graph, and gives voice assistants and downstream systems a clean read of your facts (Ahrefs). Google is explicit that structured data helps it understand a page even where no special rich result applies (Google Search Central).

One 2026 change reset expectations. On May 7, 2026, Google deprecated FAQ rich results, so FAQ markup no longer produces the expandable question dropdowns in search, though the FAQPage type stays valid and harmless to leave in place (Google Search Central). Knowing which schema types still return a visible benefit is now part of doing the job well, a check we fold into the technical pass in our 2026 website audit checklist and the delivery standards behind our digital marketing and technology programs.

The schema types worth implementing in 2026

Concentrate on the types that still return a benefit rather than marking up everything. Article or BlogPosting, Organization, Product, Breadcrumb, and HowTo remain worth implementing because they either qualify for a live rich result or sharpen entity clarity in Google's index (Google Search Central). Keep FAQPage where it fits your content, with the understanding that it no longer earns the SERP dropdown it once did.

Implementation quality matters more than coverage. Use valid JSON-LD, keep the markup consistent with what the visible page actually says, and validate before shipping so a malformed block does not undercut the signal. This is the same evidence-first discipline we bring to the ranking-versus-answer tradeoffs in GEO vs SEO: implement what the data supports, skip what it does not.

Putting schema in its place in an AI search plan

Treat schema as maintenance, and put your AI visibility strategy elsewhere. The Ahrefs data is clear that structured data will not lift citations on a page that is already surfacing, so the effort that moves AI visibility goes into strong content, entity consistency, and third-party validation. Ahrefs' separate analysis of 17 million citations found AI assistants lean toward fresher content, which points the work at publishing and updating rather than tag hygiene (Ahrefs).

Sequence it simply: keep valid schema on the types that still pay off, then invest the real hours in citable content and the off-site presence engines actually read. If you want that program owned end to end alongside your existing site work, our Next Best Action team runs it as an ongoing roadmap, and you can start a conversation to scope where schema fits your stack.