SERP FAQ Removal Weakens Schema Value

Schema markup hasn’t disappeared, but its role as a shortcut to AI citations looks much weaker following Google’s removal of FAQ rich results and new research from Ahrefs.

Google has steadily reduced visible rewards for structured data over the past few years, removing or limiting features such as FAQ, HowTo, Course Info, Claim Review, and other rich result types. More recently, Practice Problem markup was also deprecated, reinforcing a clear trend: structured data types often remain valid in code, but lose their visibility in search results once they become widely used.

Against this backdrop, Ahrefs analysed nearly 1,900 webpages that implemented JSON-LD schema and compared them with similar pages that did not. The aim was to test whether schema improves citation rates in AI systems such as Google AI Overviews, AI Mode, and ChatGPT.

The results showed little meaningful difference. AI Mode saw a very small increase of around 2.4%, ChatGPT about 2.2%, while Google AI Overviews actually showed a decline of roughly 4.6%. None of these changes were large enough to suggest a reliable uplift from adding schema. Most of the pages tested were already being cited in AI Overviews before schema was added, which further limits the strength of the conclusions.

Ahrefs noted that schema could still help newer or less visible pages get discovered or understood, but the data does not confirm any direct citation advantage for pages already indexed by AI systems. Some researchers also pointed out that the test only measured one stage of the pipeline and did not account for earlier indexing or entity recognition benefits.

Within the SEO community, the findings have sparked debate. Some argue that schema has been over-promoted in “GEO” (Generative Engine Optimisation) discussions as a way to influence AI visibility. Others see the study as evidence that structured data is being misunderstood when applied to AI citation strategies. There is also a recurring concern that tactics which become widely adopted tend to lose effectiveness over time, mirroring what has happened with other SEO features in the past.

It is also important to note that the study combined multiple schema types together, including FAQ, Article, Product and Organisation markup. This makes it difficult to isolate whether any specific type performs differently. In addition, the research window was relatively short, and it did not examine how schema might affect longer-term indexing, eligibility, or entity understanding, which remain possible indirect benefits.

For now, structured data still plays a role in helping search engines interpret content and supporting certain rich results, particularly for Product, Review, Event and Video content. However, the idea that simply adding schema will improve AI citations is not supported by the current data.

The broader takeaway is that AI systems appear to rely more heavily on visible, well-structured HTML content during retrieval, rather than hidden JSON-LD markup alone. This shifts the emphasis back towards clear page structure, direct answers, and readable content as the primary drivers of AI visibility.

Overall, the evidence suggests schema should still be used where it is relevant and required, but expectations around AI citation gains need to be more cautious. Rather than acting as a growth lever for AI search visibility, schema is increasingly behaving more like background infrastructure that supports interpretation rather than directly influencing outcomes.