Add JSON-LD to a page and an AI engine can parse its facts, author, dates and structure more reliably. What that markup will not do, on the current evidence, is make an assistant decide to quote you. When Ahrefs tracked 1,885 pages that added schema between August 2025 and March 2026 against 4,000 control pages, AI citations barely moved: about plus 2.4% on Google AI Mode and plus 2.2% on ChatGPT, both statistically indistinguishable from zero, and minus 4.6% on Google AI Overviews. So the honest answer for publishers is that schema is worth doing for comprehension, eligibility and tidiness, but it is not the lever that wins you a mention in an AI answer. The things that do that sit upstream of the markup.
What schema markup actually does
Schema markup, usually written as JSON-LD and drawn from the shared vocabulary at schema.org, is a structured description of what a page is about. It labels the article, the author, the publish and update dates, the organisation, a product, an FAQ or a how-to, in a format a machine can read without having to infer anything from the prose.
That has three concrete uses. It makes a page eligible for rich results in classic search. It feeds entity recognition, so a search engine or assistant can connect your brand, author or product to a known entity in its knowledge graph. And it gives a retrieval system a clean, unambiguous version of the facts on the page. None of those three is the same as being chosen as a citation in a generative answer, which is the distinction most "schema for AI" advice blurs.
What the 2026 evidence says about schema and AI citations
The most useful recent work separates correlation from cause, and the conclusion is consistent across studies: cited pages tend to carry schema, but adding schema does not appear to cause the citation.
The Ahrefs test: citations barely moved
Ahrefs measured citation changes on pages that added JSON-LD and compared them with matched controls across Google AI Overviews, AI Mode and ChatGPT. Adding markup produced no meaningful lift on any platform. Separately, across roughly six million URLs, AI-cited pages were nearly three times more likely to carry JSON-LD than uncited ones. The reconciliation is straightforward: pages that invest in schema are usually the same pages that invest in content quality, authority and maintenance, and those are the real citation drivers. The markup is a marker of a well-run site, not the reason the site gets quoted.
When thin schema can backfire
Implementation quality matters more than presence. A cross-platform empirical study by Kurt Fischman, published on SSRN, found that generic schema with only minimal attributes was associated with a lower citation rate than no schema at all, in the region of 41.6% against 59.8%, while only attribute-rich schema with fully populated fields beat the no-markup baseline. The effect was strongest for lower-authority domains, consistent with the idea that a complete, factual payload can partly compensate for weak authority signals. The practical reading: a sparse, boilerplate block is not a free win, and may be worse than nothing.
What still appears to help
The supporting case for schema has not disappeared. A controlled Search Engine Land experiment in 2025 found that a well-implemented schema page surfaced in an AI Overview while the no-schema version of the same content failed even to index. The signal across all of this is not "schema is useless" but "schema is comprehension and eligibility infrastructure, and only complete, accurate markup on content that is already strong does any work."
What Google says about structured data and AI features
Google's own position, set out in its Search Central guidance on AI features, is that structured data is not a direct ranking factor and is not required for a page to appear in generative AI search. There is no special schema.org type that triggers an AI Overview or an AI Mode citation. What Google does say is that valid structured data helps its systems understand a page, supports rich results, and feeds the entity recognition that AI answers draw on. Google has also framed answer engine optimisation and generative engine optimisation as, in its words, still SEO: the same fundamentals of crawlable, useful, trustworthy content, with structured data as a supporting signal rather than a shortcut.
The FAQ rich result deprecation and what it signals
A timely data point sharpens the same conclusion. On 7 May 2026 Google announced that FAQ rich results would no longer appear in Search, with the search appearance, the rich result report and Rich Results Test support removed by June 2026. FAQPage markup was for years the standard "do this for visibility" recommendation. Its retirement, landing in the same week as the Ahrefs findings, underlines that bolting on markup as a standalone visibility tactic is past its usefulness. FAQ structured data can still help a machine parse a question-and-answer block, but it no longer earns a search feature, and it never reliably earned an AI citation.
So should publishers still use schema markup?
Yes, for what it is actually good at. Use Article and Organization schema so assistants and search engines can attribute your content, dates and brand correctly and connect them to the right entity. Use Product, Recipe or HowTo markup where it matches the page and where rich results still exist. Populate fields fully and accurately rather than shipping a thin generic block, because incomplete markup can underperform no markup. Keep it consistent with the visible content, because mismatched or misleading markup is a liability, not a help.
What schema will not do is substitute for the work that wins citations: clean, retrievable text served without a heavy client-side rendering barrier, a direct answer near the top of the page, specific and verifiable claims, clear authorship, and genuine authority earned across the web. Structured data describes content that is already good. It does not make weak content quotable.
Comprehension is not compensation
There is a further gap that structured data does nothing about, and it matters most to publishers. Markup helps an AI system understand and, occasionally, cite your page. Neither understanding nor citation puts anything in your accounts. When a Live Search Agent retrieves your content to assemble an answer, that retrieval is a read of your material whether or not it ends in a visible citation, and increasingly it does not end in a click back to your site. Schema improves how machines comprehend the read. It does nothing to monetise it. That is the layer blankspace works on: detecting AI and Live Search Agent retrieval at the CDN edge and turning those reads into revenue, regardless of whether the assistant chooses to cite the source. Structured data is comprehension hygiene worth doing. Capturing value from the read is a separate problem, and it is the one that decides whether AI traffic is a cost or an income line.
Frequently asked questions
Does adding schema markup increase AI citations?
On the best current evidence, no, not directly. A 2026 Ahrefs study of 1,885 pages that added JSON-LD found no meaningful citation lift across Google AI Overviews, AI Mode or ChatGPT. Cited pages do tend to carry schema, but that reflects the kind of well-run, authoritative sites that get cited rather than a causal effect of the markup itself.
Can bad schema markup hurt my visibility?
It can. A cross-platform study found that generic schema with minimal attributes was associated with a lower citation rate than having no schema at all, while only fully populated, attribute-rich markup beat the no-schema baseline. Mismatched or misleading markup is also a liability. Incomplete structured data is not a safe default, so populate fields properly or leave them out.
Is structured data a Google ranking factor?
No. Google states that structured data is not a direct ranking factor and is not required to appear in AI features, and that there is no special schema for generative search. It does help Google understand a page, supports rich results, and feeds the entity recognition that AI answers rely on, which is a supporting role rather than a ranking lever.
Why did Google remove FAQ rich results?
Google announced on 7 May 2026 that FAQ rich results would be retired from Search, with the feature and its reporting removed by June 2026. FAQPage markup no longer earns a dedicated search appearance. The markup can still help a machine parse a question-and-answer section, but it is no longer a visibility tactic in its own right.
If schema does not drive citations, what does?
The drivers sit upstream of markup: content that is retrievable and served as clean readable text, a direct answer to the question high on the page, specific and verifiable claims, clear authorship, freshness, and authority earned across the web. Schema describes and tidies that content for machines, but it cannot manufacture the quality or trust that wins the citation, and it cannot capture revenue from the AI read at all.