A web developer pitches a Thai business owner a package: "Add schema markup to every page, and AI search tools like ChatGPT and Google's AI Overviews will start citing you." It's a clean, technical-sounding promise—a few kilobytes of structured code, tucked invisibly into the page, unlocking visibility in the newest search surface. It's also one of the most repeated claims in AI-search content right now, echoed across marketing blogs with barely a source in sight.
It sounds like a simple technical fix with an outsized reward. A rigorous, matched study of 1,885 pages that actually added schema markup between August 2025 and March 2026, run by the SEO analytics company Ahrefs, found the opposite: no meaningful citation gain on Google AI Mode or ChatGPT, and an actual decline on Google AI Overviews. This article walks through exactly what that study measured, why a real and separately-confirmed number—that AI-cited pages are three times more likely to carry schema than uncited ones—doesn't contradict it, and what a Thai business should actually do instead of chasing a markup checkbox.
Schema markup doesn't earn an AI citation on its own—it's a marker of the kind of well-structured, well-maintained site that already tends to get cited, which is a different thing to fix than adding a code snippet.
The 1,885-Page Test
Ahrefs tracked 1,885 pages that added JSON-LD schema markup between August 2025 and March 2026, and compared their citation performance against 4,000 matched control pages that didn't make the change, using a matched difference-in-differences design—a method built specifically to separate the effect of one change from everything else happening to a page at the same time.
The results, by AI surface: Google AI Mode citations moved +2.4%, and ChatGPT citations moved +2.2%—both changes the authors describe as statistically indistinguishable from zero. Google AI Overviews citations actually fell -4.6%, a result the study calls statistically significant, though in practical terms it worked out to roughly 12 fewer daily citations per page. Across the board, adding schema markup to an existing page did not produce the citation boost the "just add schema" advice promises.
The Correlation Trap: Why a Real Number Doesn't Prove Causation
Here's the part that makes the myth so persistent: it isn't built on nothing. Ahrefs' broader analysis of 6 million URLs found that 53% of AI-cited pages carry schema markup, compared to a far lower rate among uncited pages—AI-cited pages are roughly three times more likely to have it. That's a real, verified number. It's also exactly the kind of correlation that gets mistaken for causation, which is the Correlation Trap this section is named for.
The difference-in-differences design exists precisely to catch that mistake. If schema itself caused citations, adding it to a page that didn't have it before should move that page's citation rate—and largely, it didn't. What the correlation more likely reflects is that the same businesses disciplined enough to implement structured data are also the ones maintaining current information, clear page organization, and consistent content elsewhere on the site—the actual traits AI systems seem to reward. Schema is a symptom of a well-run site, not the cause of its visibility.
What the Study Doesn't Rule Out
Fair to the data: this doesn't mean schema markup is worthless, and the study's own authors flag real limits on how far the finding travels. The pages tracked were already heavily cited—each had 100 or more monthly AI citations before the test began—so the result speaks to established pages, not brand-new ones starting from zero visibility. The measurement window was only 30 days after the change, other simultaneous edits to the page weren't isolated, and different schema types (Product, FAQ, LocalBusiness, and so on) weren't tested separately. None of that overturns the headline finding, but it means "schema markup categorically does nothing" would be overreaching the data just as much as the original myth overreached it in the other direction.
Schema also still does real, separate jobs that have nothing to do with AI citation—rich results in traditional Google Search, accurate business information in Knowledge Panels, and machine-readable prices, hours, and availability that other systems can reliably parse. Those are legitimate reasons to implement it, and structured pages remain one part of a broader search foundation, alongside the answer-first, sourced writing that actually earns a citation. "It will get you cited by AI" just isn't a reason schema belongs on that list, on the evidence available now.
Picture this: Silom Silk Tailors chases the wrong fix
Picture a hypothetical business called Silom Silk Tailors, a bespoke tailoring shop in Bangkok. After reading a vendor's pitch, the owner paid to have full JSON-LD schema added across every service page—Product markup for suit categories, LocalBusiness markup for the shop's details—expecting AI search tools to start recommending them for "custom suit Bangkok" queries. Three months later, mentions hadn't moved. What had never actually been fixed was the content itself: the fabric-sourcing and turnaround-time page, the single most-asked question according to the shop's own enquiry log, still buried its real answer—two to three weeks for a first fitting, faster with a rush fee—inside a paragraph about "our commitment to craftsmanship." Rewriting that one page with the direct answer up front did more for citations in a month than the schema project had done in a quarter.
The Structural Substance Playbook
- Content-First Diagnosis — before spending on markup, check whether your pages actually state clear, current, specific answers. A citation gap is usually a content gap wearing a technical disguise.
- Schema as Hygiene, Not Magic — implement structured data for the jobs it's proven to do (rich results, accurate business listings, machine-readable prices and hours), not as a promised shortcut to AI visibility.
- Ask the Confound Question — if a competitor with schema also gets cited more, ask what else is true about their site before crediting the markup. Usually it's the underlying content quality doing the work.
- Watch the Overviews Surface Separately — Google AI Overviews was the one surface with a measured decline in this study; if that's a meaningful traffic source for your business, monitor it specifically rather than assuming what works on ChatGPT works there too.
A hospital's outpatient department page faces the identical trap: adding MedicalClinic schema without ever stating wait times, required documents, or walk-in hours in plain language fixes nothing an AI system—or a worried patient—actually needed answered.
Substance: what it all comes down to
If there's one word for what this study actually proves matters, it's Substance—not the markup wrapped around a page, but whether the page underneath says something clear, current, and specific enough to be worth repeating.
So here's the question worth sitting with: if you stripped every line of schema code off your most important page today, would an AI system—or a human reading it for the first time—still get a clear, complete answer from what's left?
FAQ
Does adding schema markup help get my site cited by AI search tools?
The best controlled evidence available—a matched study of 1,885 pages—found no statistically meaningful citation gain from adding schema on Google AI Mode or ChatGPT, and an actual decline on Google AI Overviews. It's not the lever the "add schema, get cited" advice claims it is.
If schema doesn't cause citations, why do cited pages have it so often?
Because the businesses careful enough to implement structured data tend to also keep their content current, clear, and well-organized—the traits that actually earn a citation. Schema correlates with good sites; it doesn't create them.
Should I remove schema markup from my website?
No. It still serves real, separate purposes—rich results in Google Search, accurate Knowledge Panel information, and machine-readable business details. Just don't implement it expecting an AI-citation boost as the payoff.
Does this study mean structured data is completely worthless for AI search?
Not quite that absolute. The study tracked already well-established pages over a 30-day window, so it speaks most directly to businesses with existing visibility, not brand-new pages. What it does rule out is schema as a standalone fix.
What should I actually prioritize instead of schema markup?
Clear, specific, current content that answers the exact questions your customers ask, in plain language, near the top of the page—the same content quality that this study suggests is driving the citation correlation with schema in the first place.
Is this the same as the debate over keyword-stuffing and other old SEO tricks?
It rhymes. A separate, controlled academic study of AI citation tactics found that superficial technical tricks like keyword stuffing produced no benefit either, while genuinely improving the substance of a page's content did. We cover that study's full findings here.
What this looks like with NAWA AI Websites
NAWA AI Websites treats schema markup as one part of a search foundation, not a standalone fix—structured pages, metadata, schema, and internal links are built to work together, on top of content that already states a clear, direct answer. We don't sell schema implementation as a citation guarantee, because the evidence doesn't support that claim, and neither do we.
If you're not sure whether your site's technical setup or its actual content is the real gap holding back AI visibility, talk to us about an AI website and search visibility review.
Sources: Ahrefs, "We Tracked 1,885 Pages Adding Schema. AI Citations…"