Schema markup endured one of its most consequential weeks in recent memory. Within a span of four days, Google officially ended FAQ rich results for the vast majority of sites, and Ahrefs released a controlled study indicating that adding JSON-LD schema did not produce a clear citation lift across Google AI Overviews, AI Mode, or ChatGPT. These two developments strike at the heart of the most common arguments for implementing structured data: enhanced visibility in search engine results pages and, more recently, improved citation rates in generative AI outputs. The data and the policy changes together force a reassessment of what schema actually delivers in the current search and AI landscape.
Google has been steadily narrowing the visible rewards tied to specific structured data types since 2023. FAQ rich results were restricted to authoritative government and health websites. HowTo rich results were limited to desktop and later deprecated entirely. In 2025, the company announced the retirement of several structured data features, including Course Info, Claim Review, and Estimated Salary. Book Actions was initially included in that retirement list but was later carved out after Google removed its deprecation banner. The company described the remaining retirements as features that were not commonly used in Search and no longer providing value to users. In 2026, Practice Problem structured data was deprecated. John Mueller noted on Reddit that markup types come and go, but a precious few should be held on to. The pattern is unmistakable: visible structured data rewards tend to disappear after they become familiar SEO tactics. The markup itself remains technically valid, but the rich result does not. Google does not always frame these removals as responses to overuse, but the recurring cycle offers diminishing reasons to treat any single markup type as a durable strategic asset.
What makes the recent updates different is that the evidence for one proposed replacement value has also weakened. The GEO advisory space has been claiming that schema markup boosts AI citations. The Ahrefs report tested part of that claim directly.
The Ahrefs Study: What the Data Actually Showed
Ahrefs tracked 1,885 web pages that added JSON-LD schema. Each page was matched against control pages that never added schema. Citation changes were measured across Google AI Overviews, Google AI Mode, and ChatGPT. The results were essentially flat. Google AI Mode showed a 2.4 percent increase, ChatGPT showed a 2.2 percent increase, and Google AI Overviews showed a 4.6 percent decline. The first two figures are too small to distinguish from random variation. The AI Overviews decline was statistically significant, but Ahrefs stated that it cannot confidently attribute that drop to schema. Every page in the dataset already had more than 100 AI Overview citations before any schema was added. These pages were already being crawled, indexed, and cited by AI systems before the markup was introduced.
Ahrefs acknowledged that for pages not yet visible to AI, schema might still help with crawling, parsing, or indexing. But the study's data cannot confirm that possibility. Gianluca Fiorelli, a strategic SEO consultant, called the study one of the more honest pieces of research to come out of the AI Search space in 2026. He argued that the scope was narrower than the headline suggested, comparing it to testing whether adding a label to a bottle already on the supermarket shelf makes customers pick it up more often. The distinction is important: the study measured the effect of schema on pages that were already performing well, not on pages struggling to gain AI visibility.
Ahrefs also cited a separate experiment from searchVIU, which found that five AI systems relied on visible HTML during direct page retrieval and did not use hidden JSON-LD, Microdata, or RDFa. That finding covers one stage of the AI pipeline, specifically the retrieval phase. It does not rule out schema playing a role earlier in indexing or in entity understanding. But it does suggest that when an AI system is directly reading a page to extract content, the visible structure matters more than the hidden markup.
Ryan Law, Ahrefs' director of content marketing, summarized the findings on LinkedIn with characteristic directness. He stated that adding schema markup probably does not help pages get cited in AI search and that it is probably not some magic fix for improving AI citations.
The Practitioner Debate Heats Up
Both updates land in the middle of an active and increasingly polarized argument about schema and GEO. Roughly 168,000 pages use the phrase "FAQ schema is critical for GEO," according to search results flagged on LinkedIn by Lily Ray, vice president of SEO and AI Search at Amsive. She called the trend familiar, noting that anything that can be spammed in SEO will be spammed. Ray had warned about this dynamic in a 2019 Moz article when FAQ schema first launched. She described Google's FAQ removal as the same cycle repeating itself. A useful markup type gets scaled aggressively as a tactic, Google eventually pulls the reward, and the industry moves on to the next opportunity.
Joost de Valk, founder of Yoast, made the connection explicit in a blog post. He said the GEO industry is replaying early SEO, only faster, and that the FAQ schema deprecation is the first concrete proof point that the cycle is back on. De Valk has also filed a proposal with Schema.org for a new FAQSection type, aiming to address what he sees as a structural problem: separating the concept of "this page has an FAQ section" from "this page IS an FAQ."
The frustration was sharpest from practitioners who had watched the GEO playbook harden around schema as its most concrete recommendation. Mark Williams-Cook, director at Candour and founder of AlsoAsked, shared the Ahrefs report on LinkedIn with a pointed observation. He wrote that GEO proponents are selling snake oil with schema to boost citations, and that voices like Gianluca Fiorelli are talking sense.
Marie Haynes, founder of Marie Haynes Consulting, offered a different theory entirely. She suggested that Google needed the web's FAQs to train its AI models and therefore gave publishers an incentive to add them in the form of rich results. Now that the training data has been collected, the incentive is no longer necessary. Her theory is unconfirmed by any primary source, but it illustrates how far speculation has traveled in the absence of clear official explanations.
Not all practitioners accepted the gloomy readings. Google's broader guidance still presents structured data as a way to make page information machine-readable. At a 2025 Search Central Live event in Madrid, the Search Relations team told attendees that supported structured data types are still worth using. The official position has not changed, even as specific visible rewards have been removed.
What the Data Cannot Answer Yet
Several important questions remain outside the reach of current research. Whether schema helps pages that are not yet being cited by AI is a separate question that the Ahrefs data cannot answer, because every page in the study already had more than 100 AI Overview citations before the markup was added. The test also pooled all schema types together. Article, FAQ, Product, HowTo, and Organization were all treated as a single category. Type-specific effects have not been isolated, and they could look very different from the aggregate result.
The 30-day measurement window may miss slower effects. On live websites, schema changes can overlap with other page modifications, making it difficult to separate what the markup did from what changed around it. The report only examined schema placed directly in the page's HTML, not schema injected via JavaScript, which AI crawlers are known to treat differently. Ahrefs measured Google AI Overviews, AI Mode, and ChatGPT. Whether Bing, Copilot, Perplexity, Claude, or other answer systems treat schema differently from the systems studied is an open question.
Google's FAQ deprecation notice states that the company will continue using FAQ structured data to better understand pages. What that produces in measurable terms is unclear. The same uncertainty applies to whether schema affects citations indirectly through eligibility signals, entity understanding, or source selection, rather than during the direct retrieval phase that the searchVIU experiment tested. Nobody has published data that isolates that indirect path.
Why This Matters for SEO and Content Strategy
The Ahrefs data provides no measured reason to add JSON-LD with the expectation of short-term AI citation gains for pages already visible in AI Overviews. The trickier question is what to do with schema strategies more broadly. Product, Review, Event, Video, and several other structured data types still support active rich result features. Organization, Person, and Article markup can still help describe entities and content, even when the payoff is less directly visible than a rich result or a citation count.
A blanket "schema doesn't work" reading overstates what the data actually showed, because the test pooled all types and measured only one specific outcome. What the data does challenge is a particular sales pitch. The claim that adding schema boosts AI citations has been one of the more concrete recommendations in GEO guides. Frase.io, for example, called schema markup critically important for AI search, GEO, and AEO. Without data support for that claim, it becomes harder to justify the implementation effort and maintenance cost.
The searchVIU finding that AI systems rely on visible HTML during direct page retrieval rather than on hidden JSON-LD points to a practical implication. Content structure, clear headings, and direct answers written in natural prose may matter more for AI citation than the presence of markup. That does not mean schema is useless, but it shifts the emphasis toward the quality and clarity of the visible content itself.
Looking Ahead: Schema as Plumbing, Not Leverage
The question hanging over the SEO industry is where schema creates measurable value. Adding JSON-LD did not measurably increase AI citations for pages already visible in AI Overviews. For those pages, schema looks more like plumbing that serves other systems than a lever that moves citation counts. That is still real value, but it is a fundamentally different pitch from the one that has been driving much of the recent GEO conversation.
For pages that are not yet visible to AI systems, the data simply does not exist to say whether schema helps or not. That gap leaves room for continued debate, but it also means that practitioners who are making decisions today must weigh the known evidence against the unknown possibilities. The known evidence, at least from the Ahrefs study, does not support the claim that schema boosts AI citations for pages that already have a track record of being cited.
The cycle that Lily Ray and Joost de Valk described is not new, but it is accelerating. The GEO industry, in its rush to offer concrete tactics, may have latched onto schema as a safe recommendation. The data now suggests that recommendation needs to be reexamined. Structured data still has legitimate uses, but the case for implementing it specifically to improve AI citation rates has weakened considerably. The wise path forward involves treating schema as infrastructure, not as a growth lever, and focusing content strategy on the visible signals that AI systems demonstrably rely on during retrieval.
Featured Image: BEST-BACKGROUNDS/Shutterstock
