The Schema Crisis: Why Google’s Latest Moves and New Data Are Upending SEO Assumptions

For over a decade, structured data—or "schema markup"—has been the cornerstone of technical SEO. It was the digital equivalent of a Rosetta Stone, providing search engines with a machine-readable roadmap to the content on a web page. However, the last week has served as a sobering "reality check" for the SEO industry. Following Google’s decision to sunset FAQ rich results and a damning new report from Ahrefs questioning the efficacy of JSON-LD for AI citations, the industry is facing an identity crisis regarding the role of structured data in the era of Generative AI.

The Shrinking Landscape: A Chronology of Declining Rewards

To understand the current tension, one must look at the narrowing window of visibility Google has provided to those who implement structured data. The narrative that schema is a "golden ticket" to SERP real estate has been eroding for years.

The Gradual Rollback

• 2023: Google began restricting FAQ rich results, limiting them primarily to authoritative government and health-related websites. Simultaneously, "HowTo" rich results were confined to desktop before being deprecated entirely.
• 2025: A massive cull occurred. Google announced the retirement of several structured data features, including Course Info, Claim Review, and Estimated Salary, citing low usage and lack of value for the end user.
• 2026: The trend continued with the deprecation of "Practice Problem" structured data. While John Mueller, Google’s Search Advocate, attempted to temper fears by stating that "markup types come and go," the cumulative effect has been a loss of faith among practitioners who view these rich results as the primary ROI for their schema efforts.

The recurring pattern is clear: Google adopts a markup type, the SEO industry treats it as a tactical advantage to "game" the SERP, and eventually, Google removes the visual reward. While the markup remains technically valid, the absence of a rich result strips it of its primary appeal to webmasters.

The Ahrefs Investigation: Deconstructing the "GEO" Myth

Perhaps the most significant blow to the status quo came four days after the FAQ announcement, when Ahrefs published a comprehensive study investigating the relationship between JSON-LD schema and AI citations.

The rise of Generative Engine Optimization (GEO) has been predicated on the theory that structured data is the "secret sauce" for appearing in AI Overviews, Google’s AI Mode, and ChatGPT. Ahrefs put this theory to the test by tracking 1,885 web pages that implemented JSON-LD, comparing them against control pages that did not.

The Statistical Reality

The findings were, by any measure, underwhelming:

• Google AI Mode: +2.4% change (statistically indistinguishable from noise).
• ChatGPT: +2.2% change (statistically indistinguishable from noise).
• Google AI Overviews: -4.6% change (statistically significant, though Ahrefs cautioned against direct attribution).

The takeaway was stark: For pages that were already well-indexed and frequently cited, adding JSON-LD provided no tangible uplift in AI visibility. Ryan Law, Ahrefs’ director of content marketing, bluntly stated on LinkedIn: "Does adding schema markup help your pages get cited in AI search? Probably not."

The Expert Consensus: "Snake Oil" vs. "Plumbing"

The Ahrefs report sparked a polarized debate among industry leaders, ranging from calls for a complete abandonment of schema-heavy strategies to more nuanced defenses of the technology.

The Case for "Snake Oil"

Critics, such as Mark Williams-Cook, director at Candour, have been vocal about the "GEO bros" who have been selling schema as a magic bullet for AI rankings. The sentiment is that agencies have been over-promising the benefits of structured data to clients who don’t understand the technical nuances of how LLMs consume data.

Lily Ray, VP of SEO and AI Search at Amsive, observed that the cycle of "spamming" new schema types to force visibility is repeating itself. She notes that as soon as a new structured data type is introduced, it is weaponized for rankings, forcing Google’s hand to deprecate the display feature.

The Structural Argument

Joost de Valk, founder of Yoast, believes the problem is architectural. He has proposed a new FAQSection type to the Schema.org community, arguing that we need to better distinguish between pages that are FAQs and pages that contain FAQ sections. This structural clarity, he argues, is what search engines actually need, rather than the current, cluttered implementation methods.

The "Training Data" Theory

Marie Haynes, a prominent SEO consultant, offered a provocative, albeit unconfirmed, theory: Google may have incentivized the adoption of FAQ schema primarily to harvest high-quality, structured training data for its LLMs. Now that the models have ingested this data, the "reward" of rich results is no longer necessary for Google to maintain its dominance.

What the Data Can’t Tell Us (Yet)

While the Ahrefs report is a critical piece of research, it is not the final word on schema. Several variables remain unaddressed:

1. The "New Page" Variable: Every page in the Ahrefs study already had more than 100 AI citations. It remains possible that schema is beneficial for smaller, newer, or less-indexed websites that are struggling to gain traction in the initial crawling and parsing stages.
2. Entity Understanding: There is a distinction between direct retrieval (what the searchVIU experiment tested) and entity understanding. While LLMs may prioritize visible HTML for answering a prompt, structured data may still be vital for Google’s internal Knowledge Graph and the way it classifies the authority of a site.
3. Type-Specific Effects: The Ahrefs study pooled all schema types (Product, Article, Organization, etc.) into one bucket. It is possible that specific types, such as Product schema, have a different impact on e-commerce AI responses than generic Article schema.
4. Beyond Google: The test focused on Google and ChatGPT. Whether other systems—such as Perplexity, Claude, or Bing—parse schema differently is an open question that warrants further investigation.

Implications for the Future of SEO

Where does this leave the average SEO practitioner? The narrative that schema is a "quick fix" for ranking or citation growth is effectively dead. However, labeling schema as "useless" would be a tactical error.

1. Shift from "Rankings" to "Machine Readability"

We must return to the original purpose of structured data: helping machines understand content. While it may not move the needle on AI citations, it remains a fundamental best practice for site architecture. If you are building a product-heavy site, Product schema is not for the "gain"; it is for the "context."

2. Prioritize Prose Over Markup

The searchVIU experiment confirms that AI systems rely heavily on visible, well-structured HTML. Instead of obsessing over hidden JSON-LD, SEOs should double down on:

• Clear Headings (H1-H6): Creating a logical document outline.
• Direct Answers: Providing concise, prose-based answers to user questions within the body copy.
• Entity Mapping: Ensuring that content is clearly associated with recognized entities through natural language.

3. Adopt a "Minimalist" Schema Strategy

Practitioners should stop adding schema "just in case." Instead, implement only the types that are explicitly supported and provide genuine descriptive value for the entities present on the page. If a markup type is not providing a direct rich result or a clear, documented benefit, the development time might be better spent on content quality.

Final Thoughts: The Maturation of GEO

The "Schema Crisis" is not a sign that structured data is failing; it is a sign that the industry is maturing. The initial "gold rush" phase of AI search, where practitioners tried to trick algorithms with fancy code, is coming to an end.

The future of SEO lies in understanding that LLMs are not just "super-powered search engines"—they are sophisticated information processors. They prioritize the clarity and structure of the information they consume. If that information is buried in complex JSON-LD but missing from the readable text, the AI will likely ignore it.

As the industry moves forward, schema should be viewed as "plumbing"—essential for the infrastructure of the web, but not the primary lever for growth. The real competitive advantage in the age of AI won’t be found in the code that the user never sees, but in the authority, clarity, and depth of the content that they do.