Schema Markup for AI Search: What the 2026 Data Actually Shows

Start with the correlation, because it’s real. In a 2026 analysis of roughly 6 million URLs, Ahrefs found that pages cited by AI systems were almost three times more likely to have JSON-LD schema than pages that weren’t cited. Taken alone, that number has been used across the industry as proof that schema drives citations.

So Ahrefs ran the obvious follow-up test.

They identified 1,885 pages that added JSON-LD between August 2025 and March 2026, matched them against roughly 4,000 control pages with similar prior citation levels, and measured the change 30 days before and after schema went live: a matched difference-in-differences design, the standard way to isolate cause from coincidence. Across Google AI Mode and ChatGPT, the change sat close enough to zero that Ahrefs classified it as noise. For AI Overviews the measured change was actually a small statistically significant decline, but both the treated and control pages were already trending down before any schema was added, so schema wasn’t the cause there either.

The explanation for the correlation is straightforward once you see it: schema tends to live on better-maintained, more technically sophisticated sites. Those same sites publish stronger content, earn more links, and build more authority: all the things AI retrieval actually rewards. The schema rides along with quality; it doesn’t manufacture it. Ahrefs’ own summary put it plainly: if you’re already doing the rest of the SEO work well, JSON-LD isn’t what is holding you back, and if you’re not, schema by itself won’t make up for it.

Do AI engines even read your schema?

During live retrieval, mostly no, and this is the finding that reframes everything. A 2026 experiment by searchVIU tested whether five major systems (ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode) actually used structured data when fetching a page in real time. When those systems pulled a page directly, every one of them extracted only the visible HTML content. Hidden JSON-LD, Microdata, and RDFa were ignored. In other words, the markup a human never sees is often invisible to the model at the exact moment it’s deciding what to quote.

That doesn’t mean schema is meaningless to AI. It means the value lands earlier in the pipeline. Microsoft’s Fabrice Canel confirmed on stage at SMX Munich in March 2025 that Bing uses schema markup to help its LLMs understand content for Copilot. The reconciliation is that structured data helps engines crawl, parse, index, and build a durable understanding of your content over time, even if it isn’t the deciding factor during a single live answer. Understanding and retrieval are different steps, and schema clearly touches the first one.

The pitch says schema gets you quoted. The data says schema helps machines understand you — and understanding is worth having, even when it isn’t the thing that wins the citation.

Where does schema markup still matter?

Schema still earns its place: just for reasons that predate the AI-citation hype and remain well documented. Even the Ahrefs study that found no citation lift was careful to list what structured data still does reliably:

• Rich results in Google Search. Star ratings, FAQs, recipes, event details, and product pricing all depend on schema. These improve how your listing looks and can lift click-through from traditional results — which still send meaningful traffic.
• Entity recognition and the knowledge graph. Schema like Organization and Person markup helps engines connect your brand to a stable, machine-readable identity. That entity understanding feeds directly into how AI systems describe you, which is a separate and durable payoff. We go deep on this in building the brand entity and knowledge graph AI search trusts.
• Getting crawled, parsed, and indexed in the first place. Ahrefs explicitly flagged that its study measured pages already being cited heavily — it couldn’t speak to whether schema helps a brand-new or thin page get discovered and understood initially, where clean labeling plausibly helps more.
• Voice assistants and downstream systems. Plenty of machine consumers of your content beyond the headline AI engines still lean on structured data to answer questions cleanly.

None of that is glamorous, and none of it shows up as “we added schema and citations spiked.” It shows up as infrastructure that quietly keeps you legible to machines: the same reason you’d fix crawl errors or a broken sitemap. For the broader version of this discipline, see our guide to whether your website is actually AI-ready.

Which schema types are worth implementing?

Focus on the handful that map to how engines identify and organize your content, and don’t over-engineer the rest. A practical priority order for most business sites:

1. Organization (and Person where relevant). This is your entity anchor — name, logo, profiles, and relationships. It’s the markup most tied to being recognized as a consistent brand across search and AI.
2. Article / BlogPosting. Attribution for editorial content: author, publish and update dates, headline. Useful for content attribution and for signaling freshness, which AI systems do weigh.
3. FAQPage. Cleanly structured question-and-answer pairs that mirror how people phrase things to an AI engine — use it only where the Q&A is genuinely on the page, not bolted on.
4. Product / Service. For commercial pages, this labels the thing you sell, its price, and its availability — core to rich results and to being understood in a buying context.
5. BreadcrumbList. Helps machines understand your site structure and how a page fits into the broader topic, reinforcing the architecture around it.

One honest limitation from the research: the controlled study pooled all these types together, so it couldn’t isolate whether, say, FAQPage behaves differently from Organization for AI specifically. Treat the priority order above as sound technical hygiene mapped to documented functions, not as a ranked list of citation boosters, because the data doesn’t support that framing for any single type.

So should you still add schema markup?

Yes, as hygiene, not as a growth strategy. Add it because it powers rich results, strengthens your entity, and keeps you legible to machines, and set your expectations accordingly. The mistake is spending weeks marking up every page in the belief that citations will follow, when that time belongs on the things retrieval actually rewards: authoritative, comprehensive content and real third-party recognition. A sane sequence:

1. Implement the core types once, correctly. Organization, Article, and the commercial or FAQ markup that fits your pages. Validate it, then move on — this is a one-time build, not an ongoing campaign.
2. Put the freed-up effort into content and authority. What earns citations is being a source engines already trust. See how LLMs actually choose their sources for what those signals are.
3. Keep schema accurate as pages change. Stale or mismatched markup is worse than none. Treat it like any other part of technical SEO — maintained, validated, and left to do its quiet job.