SERP FAQ Removal & New Data Challenge Schema’s AI Search Value

There is a specific kind of frustration that comes with technical SEO. You spend hours—sometimes days—meticulously implementing a set of rules, checking your JSON-LD for errors, and deploying a strategy because the "experts" say it is the key to unlocking more visibility. Then, in a single update, the goalposts move. The feature disappears, or the promised lift simply never happens.

This is exactly where we find ourselves right now with schema markup. In a very short window, two major events have shaken the foundation of how we view structured data. First, Google effectively ended FAQ rich results. Shortly after, a comprehensive report from Ahrefs suggested that adding JSON-LD doesn't actually provide a clear boost in citations across the major AI search interfaces.

For those of us who treat SEO as a craft rather than a series of hacks, this is a necessary moment of reflection. It forces us to ask whether we are optimizing for actual user value or simply chasing a ghost in the machine. When the common pitches for schema—better SERP visibility and AI citation gains—start to crumble, we have to look at what is actually left.

Google's Visible Schema Rewards Have Been Narrowing For Years

If we step back, the recent removal of FAQ rich results isn't an isolated incident; it is the latest chapter in a long-term trend. Since 2023, Google has been systematically pulling back the visible rewards it once gave to specific types of structured data.

It started with a tightening of the reins on FAQ rich results, which were restricted to authoritative health and government sites. Around the same time, HowTo rich results were limited to desktop users before being deprecated entirely. The trend accelerated into 2025, when Google retired several other features, including Estimated Salary, Claim Review, and Course Info. While Book Actions was briefly on the chopping block, it was eventually spared after Google removed the deprecation notice.

Google's official reasoning is usually that these features are "not commonly used" or no longer provide meaningful value to the end user. By 2026, this pattern continued with the deprecation of Practice Problem structured data. John Mueller has pointed out that while many markup types come and go, there are a few essential ones that practitioners should keep.

The pattern here is clear: a specific markup type becomes a popular SEO tactic, it gets scaled across the web, and once it becomes a predictable "hack," Google removes the visible reward. The underlying markup might still be valid in the code, but the "rich result" that drove the clicks disappears. This suggests that treating any single markup type as a long-term growth strategy is a risky bet.

What The Ahrefs Report Found

While Google was removing visible rewards, the industry began pivoting toward "GEO" (Generative Engine Optimization), claiming that schema is the secret to getting cited by AI. Ahrefs decided to test this claim with a rigorous study.

They tracked 1,885 web pages that implemented JSON-LD schema and compared them against a control group of pages that did not. They measured whether this addition led to more citations in Google AI Overviews, Google's AI Mode, and ChatGPT. The results were, for the most part, flat.

The data showed a marginal increase of 2.4% for Google AI Mode and 2.2% for ChatGPT—numbers so small they are essentially indistinguishable from random noise. More surprisingly, Google AI Overviews actually showed a decline of 4.6%. While this decline was statistically significant, Ahrefs noted they couldn't definitively blame the schema itself for the drop.

There is a critical nuance here: every page in the study already had over 100 AI Overview citations before the schema was added. These pages were already being crawled and recognized. As consultant Gianluca Fiorelli noted, this is like testing if adding a label to a bottle already on a supermarket shelf makes people buy it more often. The bottle is already there; the label isn't what got it into the store.

Furthermore, Ahrefs referenced a searchVIU experiment which found that five different AI systems relied on the visible HTML of a page during direct retrieval, ignoring the hidden JSON-LD, RDFa, or Microdata. This suggests that for the final stage of AI retrieval, the "hidden" technical markup may be less important than the actual content the user sees.

The Practitioner Debate

These findings have landed right in the middle of a heated debate about the validity of GEO. If you search for it, you'll find roughly 168,000 pages claiming that FAQ schema is "critical" for AI search visibility. Lily Ray has pointed out that this is a classic SEO cycle: anything that can be spammed, will be spammed.

Ray noted that she saw this same pattern back in 2019 when FAQ schema first launched. The industry finds a lever, pushes it to the limit, and then Google removes the lever. The current obsession with schema for AI citations is simply the latest iteration of this cycle.

Joost de Valk, the founder of Yoast, has been even more direct, suggesting that the GEO industry is essentially replaying the early days of SEO, just at a much faster pace. He views the deprecation of FAQ rich results as concrete proof that the cycle is repeating. To address the structural issue, de Valk has even proposed a new FAQSection type to Schema.org, attempting to distinguish between a page that simply contains an FAQ section and a page that *is* an FAQ.

What The Data Can't Answer Yet

It is important not to over-read the Ahrefs data. While it challenges the "magic pill" narrative, there are several gaps in the research that leave room for nuance.

First, the study didn't look at pages that were not already being cited. It is entirely possible that for a page that is completely invisible to AI, schema could still help with the initial crawling, parsing, or indexing process. The data tells us that schema doesn't necessarily boost a page that is already winning, but it doesn't prove it's useless for a page that is losing.

Second, the study pooled all schema types together. Article, Product, Organization, and FAQ were all treated as one category. It is possible that some types of schema are far more effective than others, but those specific effects were not isolated.

Finally, the 30-day window used for measurement might be too short to capture slower, systemic changes. In a real-world environment, schema changes rarely happen in a vacuum; they usually overlap with content updates or site architecture changes, making it difficult to isolate the exact impact of the JSON-LD alone.

Why This Matters

So, where does this leave us? The immediate takeaway is that you should not add JSON-LD with the expectation of a short-term spike in AI citations, especially if your content is already performing well in AI Overviews.

However, this is not a signal to delete all your structured data. A blanket "schema doesn't work" conclusion is an oversimplification. Many schema types—such as Product, Review, Event, and Video—still power active rich results that drive significant organic traffic. Additionally, Organization, Person, and Article markup remain useful for defining entities and helping search engines understand the relationship between different pieces of content.

What the data actually challenges is the sales pitch. The idea that you can "trick" an AI into citing you simply by adding a few lines of code is likely a myth. The value of schema is shifting from a "growth lever" to a "foundational requirement."

Looking Ahead

The central question for the industry now is where schema creates measurable value. If it isn't a direct lever for increasing citation counts, what is it?

The most likely answer is that schema is "plumbing." Plumbing isn't exciting, and it doesn't usually result in a sudden surge of traffic, but it ensures that the systems connecting your site to the search engine are functioning correctly. It provides a clean, standardized way for machines to understand your data, even if that understanding doesn't always manifest as a visible reward in the SERPs.

Moving forward, the goal should be to implement schema for the sake of clarity and entity definition, rather than as a shortcut to visibility. When we stop treating technical SEO as a series of shortcuts, we can focus on the only thing that has ever consistently worked: creating content that is genuinely useful to the human being on the other side of the screen.

More Resources

For those looking to dive deeper into the intersection of AI and search behavior, I recommend exploring the Ahrefs test on AI citations and researching the shift toward human-centric behavior over LLM-centric optimization.

How I would turn this into action

For me, the useful part of "SERP FAQ Removal & New Data Challenge Schema’s AI Search Value" is not only the idea itself, but the operating habit behind it. I would use the article as a checklist for decisions: what deserves attention now, what should be monitored, what needs a stronger evidence base, and what can wait until the system has more scale.