Schema markup doesn't increase AI citations. That finding cuts against a claim repeated constantly in agency decks and LinkedIn posts: add schema markup, and your content gets picked up by AI Overviews, ChatGPT, and every other answer engine racing to summarize the web.
That doesn't mean schema is optional. It means the reason it's not optional has been misstated, and getting that reason right matters for how we prioritize the work.
What the Evidence Actually Shows
Ahrefs ran the controlled test: 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched against 4,000 control pages with similar citation histories, tracked across Google AI Overviews, AI Mode, and ChatGPT.
Google AI Overviews: -4.6% (small decline, statistically significant)
Google AI Mode: +2.4% (indistinguishable from zero)
ChatGPT: +2.2% (indistinguishable from zero)
The only movement that cleared the bar for statistical significance ran in the wrong direction. A separate searchVIU experiment explains why: when ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode retrieve a page in real time, all five extract only the visible HTML. JSON-LD, hidden Microdata, and hidden RDFa go unread. A system that isn't reading the markup when it fetches the page isn't citing pages because of that markup.
Every page in the dataset already had more than 100 AI Overview citations before the schema was added, so the study measured whether schema pushes an already-cited page higher, not whether it helps a page get noticed for the first time. That leaves room for schema to matter earlier in a page's life in ways this dataset couldn't capture.
Google and Bing are the only two platforms that have gone on record saying structured data helps them understand a page. That's confirmed infrastructure for two named platforms, not a universal citation lever, and it lines up with what the retrieval data shows. Nobody at OpenAI, Anthropic, or Perplexity has said the same about their own systems.
So Why Do We Still Treat Schema as Required, Not Optional
Because the AI-citation argument was never the strongest case for schema. It was just the newest one, and it arrived at a moment when everyone building an SEO strategy wanted a clean story connecting technical work to the AI search shift. The older, better-supported reasons for implementing structured data didn't go anywhere.
Schema is how a page tells Google explicitly what it is, rather than asking Google to infer it. A recipe page marked up with Recipe schema doesn't leave the ingredient list, cook time, and rating to guesswork. A service page marked up with Service schema doesn't leave the question of what the page offers open to interpretation. That explicitness earns rich results: the star ratings, breadcrumb trails, and other search features that take up more space and attract more clicks in a standard search result. None of that depends on what an AI system decides to do with the page later. It's a direct mechanical relationship between marked-up content and how Google's own results page renders.
Schema also builds entity clarity across a site's content, which matters independently of any single search result. When every service page, every case study, and every author bio is consistently marked up, Google's knowledge graph has a coherent, unambiguous picture of the organization, the people in it, and how the content connects. That coherence compounds. Authoritative, well-structured sites tend to perform well across the board, and schema shows up as a marker of that, not a cause of it: sites that invest in clean structured data tend to also invest in the technical rigor and content quality that actually drives performance.
That's also the honest explanation for why so many AI-cited pages happen to have schema in the first place. Sites that implement structured data also tend to run clean technical SEO, publish authoritative content, and maintain their pages actively, which means the schema and the citation are correlated because they share a common cause rather than because one produces the other.
Where This Leaves Actual Implementation
We treat structured data as part of the technical foundation on every build, the same way we'd treat clean URLs or a working sitemap, rather than an AI optimization tactic bolted on separately. That's an important distinction, because it changes what we're actually optimizing for.
In practice, that means a two-layer approach: global schema and page-specific schema. On one retail security client's site, Organization and WebSite schema run sitewide, establishing one canonical entity with a stable ID. Each page then carries its own schema for whatever's actually on it: a Service node whose provider points back to that same Organization ID, a FAQ Page for genuine Q&A content, a Breadcrumb List for site structure, and Product entries under an offer catalog, each with its own name, image, and brand. The content is unique to the page; the core business entity isn't redefined on every one of them. A clean implementation like this is also easier to validate and maintain, and less prone to the conflicting or duplicate markup that creates errors in Search Console.
We prioritize schema types based on what the page actually is and what Google can do with that information today, rather than speculating about what an AI system might value next quarter. Service pages get Service schema referencing the sitewide Organization entity. Case studies and articles get Article schema with the author, publication date, and organization clearly attributed. That attribution work matters for a reason that has nothing to do with AI citation odds. It's the same experience, expertise, authority, and trust signal work that supports how Google evaluates a page's credibility and the site behind it, and that evaluation happens whether or not an AI system ever touches the content downstream.
FAQ Page schema is worth a specific note, because the ground shifted under it in 2026. Google discontinued FAQ rich results in Search that May, which means the dropdown-in-the-SERP payoff that made FAQ schema an easy sell for years is gone. We still use it where a page genuinely answers a series of discrete questions, because it remains a clear, structured signal of what the content covers. The justification changed: it's a clarity play now. That's a smaller win than a rich result, which is why it shouldn't be treated as a default addition to every page the way it once was.
We validate rather than assume. Schema that's malformed or inconsistent with the visible content on the page isn't neutral. It's worse than no schema at all, because it creates a mismatch between what the markup claims and what a visitor or a crawler actually finds, and that mismatch is the kind of signal that erodes trust rather than building it. Every implementation gets checked against Google's structured data testing tools before it ships. The code being written is the start of that process, not the end of it.
The Honest Version of the Pitch
Schema markup is not going to be the thing that gets a page cited in an AI Overview it wasn't already positioned to earn. If that's the promise driving the budget conversation, it's worth resetting the expectation before the work starts, because the data available right now doesn't support it, and setting an expectation the evidence can't back up is the kind of thing that erodes trust in the whole engagement once someone checks.
What schema reliably does is give Google explicit, structured information about a page instead of leaving Google to infer it, and that explicitness earns rich results, strengthens entity clarity across a site, and builds the kind of technical foundation that correlates with the broader authority signals AI systems and traditional search both reward. That's a real, defensible reason to treat structured data as required infrastructure. It's just a different reason than the one making the rounds in most AI-search content this year, and the difference matters for anyone deciding where their technical SEO budget actually needs to go.
If your site's schema implementation hasn't been reviewed since this shift in how AI search actually works, that's a reasonable place to start.
