TL;DR

Adding schema markup to pages already cited by AI produced no citation uplift in a controlled Ahrefs study of 1,885 pages; Google AI Overviews citations actually dropped 4.6% relative to matched controls. Google's own documentation states no special markup is needed for AI features.

The FAQPage rich result was deprecated in May 2026. Bottom line: schema remains useful for indexing and rich results, but there is no evidence it boosts AI citations — prioritize other SEO investments.

Ask any GEO consultant how to get cited by Google's AI Overviews, AI Mode, or ChatGPT, and "add schema markup" is usually near the top of the list. It's a clean, actionable-sounding recommendation, and it happens to be easy to sell as a service. The problem is that the best controlled evidence available says it doesn't do what the pitch claims.

This isn't a case of "it depends" hand-waving. There's a real, large-scale study on this exact question, and its results are specific enough to actually change how you prioritize engineering time.

The study that actually tested this

In May 2026, Ahrefs published a controlled study tracking 1,885 pages that added JSON-LD schema markup for the first time between August 2025 and March 2026. Rather than just correlating schema presence with citation rates (which is trivially confounded — well-resourced sites add schema and do everything else right), Ahrefs ran a matched difference-in-differences design: each treated page was compared against three control pages from different domains with similar pre-existing AI citation levels that never added JSON-LD, for a control group of roughly 4,000 pages. They measured citation changes in the 30 days before and after schema was added, across Google AI Overviews, Google AI Mode, and ChatGPT.

The results, as reported by Search Engine Journal and Search Engine Roundtable:

  • Google AI Mode: +2.4% — statistically indistinguishable from noise
  • ChatGPT: +2.2% — statistically indistinguishable from noise
  • Google AI Overviews: −4.6% — a small but statistically notable decline relative to matched controls

Ahrefs's own conclusion: adding schema produced no major citation uplift on any platform tested. The AI Overviews decline is the one number worth sitting with, because it runs directly counter to what most GEO vendor content claims. Ahrefs is appropriately cautious about over-reading it — both treated and control pages were already trending down before the schema was added, and a 30-day window could miss slower effects — but it is not evidence for the "schema helps you get cited" narrative either.

Two caveats matter for how you apply this. First, every page in the dataset already had 100+ AI Overview citations before schema was added — this is a study of pages already inside the consideration set, not pages trying to break in from zero. Second, the study pooled Article, FAQ, Product, HowTo, and Organization schema together rather than isolating effects by type, so it can't tell you "FAQPage specifically does X" — only that schema as a category, added to already-cited pages, didn't move the needle.

Ahrefs also found that pages cited by AI are about three times more likely to carry JSON-LD than uncited pages — but they explicitly frame that as correlation with overall site quality (technical maturity, content depth, link profile), not causation from the markup itself.

Google's own documentation agrees

Direct answer: This isn't just a third-party study contradicting Google — Google's own current guidance says the same thing. Its documentation on AI features in Search states plainly:

"You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add."

That's about as direct a rebuttal of the "schema is an AI citation hack" pitch as you'll find from the source itself. Google still recommends keeping structured data accurate and in sync with visible page content as part of general SEO hygiene — but that's a rich-results and indexation argument, not an AI-citation argument.

The FAQPage nuance: a rich result actually got killed

Direct answer: Separate from the causation question, there's a concrete platform change worth knowing: Google deprecated FAQ rich results from Search. Its FAQPage structured data documentation now carries a notice that the feature stopped appearing in Search results as of May 7, 2026, with supporting tooling (the FAQ search-appearance filter, the Rich Results Test entry, and related Search Console API fields) phased out over the following months, as Search Engine Journal covered.

This had been coming for a while — Google restricted FAQ rich results to a narrow set of government and health sites back in 2023, and the 2026 change finished that withdrawal for the few remaining eligible sites. FAQPage as a schema.org type is not dead; it's still valid markup and other consumers of structured data (including whatever various AI crawlers parse) may still read it. What's gone is Google's SERP-level visual treatment. If your organization was building a content strategy around "add FAQ schema, get the SERP snippet," that specific payoff no longer exists.

Schema type by type: what's actually proven

Schema typeEffect on AI citationStill worth implementing?
Article / NewsArticleNo causal evidence of citation lift; supports content classification and indexingYes — low cost, standard practice, helps indexing and eligibility for other features
FAQPageNo citation evidence either way; Google's own SERP rich result for it is goneOnly if the FAQ content is genuinely useful on the page; don't add it solely for a rich result that no longer exists
ProductNo isolated AI-citation study; helps price/availability accuracy in shopping surfacesYes for ecommerce — this is a correctness/indexation case, not a citation hack
OrganizationNo citation evidence; feeds entity/knowledge-panel understandingYes — cheap, supports brand entity disambiguation regardless of AI citation effects
Claim / ClaimReviewNot covered by the Ahrefs dataset; narrow, fact-check-specific use caseOnly if you publish actual fact-checks — see schema.org/ClaimReview and Google's fact-check markup docs

Note what's missing from the "proven" column: nothing here has evidence of causing AI citation gains. That's the honest state of the research, not an oversight in this table.

Why "structured data helps AI understand your content" oversimplifies

A common vendor framing is that structured data is how AI systems "understand" a page, as though it's a required parsing layer. It's worth being precise about what's actually happening: general-purpose language models and retrieval systems primarily process the rendered, visible content of a page, not a separate structured-data channel. Google's own general structured data guidelines are explicit that structured data has to accurately reflect visible page content — it doesn't grant access to information that isn't otherwise on the page, and it isn't a substitute for that content being well-written and complete. Schema is metadata about content, not a shortcut around producing the content itself.

When schema is still worth your engineering time

Direct answer: The case for structured data hasn't disappeared — it's just a different case than "citation hack":

  • Rich-result and SERP eligibility that still exists. Product, Recipe, Event, and several other types still drive real rich results in classic Search. That's still traffic.
  • Entity clarity. Organization and sameAs linking help disambiguate your brand across knowledge graphs, which is a real (if hard to isolate) input to how systems represent your business.
  • Correctness and machine-readability as hygiene. Accurate Product availability/price data, Article publish dates, and author markup reduce the chance of stale or wrong information surfacing anywhere it's consumed — a data-quality argument, not a ranking one.
  • Low marginal cost. For most CMS setups, adding well-formed JSON-LD for types that genuinely describe the page is cheap. The argument against it isn't "don't bother" — it's "don't expect it to move AI citations, and don't build a strategy around it."

Where structured data is oversold: as a lever to get cited more by AI Overviews, AI Mode, or chat assistants. Based on the best available controlled evidence, it isn't one. If a vendor pitch leans on schema markup as a primary AI-visibility strategy, ask what they'd point to as evidence — at the time of writing, the most rigorous public study on the question found no meaningful positive effect, and a small negative signal on Google AI Overviews specifically.

The practical takeaway

Implement schema for the reasons that are actually supported: indexation correctness, rich-result eligibility where those results still exist, and entity clarity. Don't reallocate content or engineering budget away from things with real evidence behind them — page-level topical depth, actual citation-worthy answers, and the kind of content that gets referenced by other sites — in favor of chasing an AI-citation effect that the current data doesn't show. Tools that track your AI citation presence over time (nqzai's monitoring surfaces this by function, without requiring you to guess) are far more useful for knowing whether anything you're doing is moving citations than any single markup checklist is.

Sources: