Nana Wu
For years, the SEO community has operated under a widely accepted assumption: if you want your content to be favored by AI search engines—such as Google’s AI Overviews, Perplexity, or ChatGPT—you must implement robust schema markup. The logic seemed sound. By providing structured data (JSON-LD) to search engines, you are essentially handing them a map to your content, making it easier for Large Language Models (LLMs) to parse, categorize, and verify your information.
However, a groundbreaking new report from Ahrefs has cast a long shadow over this "best practice." In a comprehensive study analyzing the relationship between schema implementation and AI citations, Ahrefs found that simply adding JSON-LD did not result in a clear, measurable increase in AI citations. While schema remains a cornerstone of traditional SEO for rich snippets and knowledge graph optimization, its utility as a direct "ranking factor" for AI visibility appears to be significantly overstated.
The Core Findings: Correlation vs. Causation
To understand the scope of the investigation, it is necessary to look at the numbers. Ahrefs analyzed 6 million URLs and observed a striking correlation: pages cited by AI were approximately three times more likely to include JSON-LD than those that were not.
In the SEO world, this is often interpreted as a "smoking gun" suggesting that schema improves AI visibility. Yet, Ahrefs recognized that this observation was plagued by a classic data fallacy: confusing correlation with causation. Websites that invest in implementing complex schema markup are often the same websites that invest in high-quality content, robust technical SEO, and strong link-building strategies.
To isolate the impact of schema, Ahrefs conducted a controlled experiment. They tracked 1,885 web pages that added JSON-LD schema for the first time. To ensure scientific rigor, each "treated" page was matched against three "control" pages from different domains. These control pages were selected based on similar pre-existing citation levels and were verified to have never implemented JSON-LD. Ahrefs then measured citation changes across three major platforms—Google AI Overviews, AI Mode, and ChatGPT—for a 30-day window before and after the implementation of the schema.
The result? None of the platforms showed a meaningful, statistically significant increase in citations after the schema was added.
Chronology of the Investigation
The study was not a simple "before and after" snapshot; it was a multi-phased longitudinal analysis designed to account for external noise.
Phase 1: The Baseline Observation (Pre-Study)
Before the experiment began, Ahrefs identified a massive disparity in the digital landscape. Pages that were already performing well in AI environments were heavily populated with JSON-LD. This established the "myth" that schema was the driver of this success.
Phase 2: The Implementation Window
Ahrefs monitored the 1,885 target pages during their transition. By utilizing a "matched difference-in-differences" analysis, researchers were able to filter out platform-wide fluctuations. For example, if AI Overviews generally became more or less citation-heavy during the month of the test, the researchers adjusted the baseline to ensure those trends weren’t falsely attributed to the schema update.
Phase 3: The Post-Implementation Monitoring
For 30 days following the update, the team tracked citation velocity. They utilized their "Brand Radar" tool and "Agent A" to crawl and log AI citations across the web. The study included four distinct tests to verify the results, and in every single instance, the outcome remained the same: the addition of schema resulted in a "null" effect.
Supporting Data and the "AI Overview Decline"
One of the more nuanced findings in the report involves the AI Overview section specifically. During the study, the team observed a 4.6% decline in citations for the treated pages. While this might initially look like a negative impact of schema, the data provides a more sobering context.
Both the treated pages and the control pages were already experiencing a downward trend in citation frequency before the schema was ever added. While the treated pages did decline slightly faster than the control group, the difference was marginal—roughly 12 fewer citations per page in a pool where these sites were already receiving hundreds.
Ahrefs remains cautious, noting that while this could suggest a minor negative effect, it is just as likely to be a statistical coincidence or an artifact of broader algorithm changes. The report explicitly avoids declaring that schema harms performance, favoring the conclusion that it simply fails to provide the anticipated boost.
The Technical Reality: How AIs Actually "See" Pages
Perhaps the most damaging evidence against the importance of schema for AI retrieval comes from an experiment conducted by searchVIU, which Ahrefs cites as a vital piece of the puzzle.
In this experiment, researchers tested five different AI systems to see how they interacted with pages in real-time. The test asked a fundamental question: Do these models look at the underlying JSON-LD/Microdata, or do they look at the rendered HTML?
The results were unanimous: none of the AI systems utilized the schema markup during the real-time fetch. They consistently ignored JSON-LD, Microdata, and RDFa, choosing instead to parse the visible HTML content of the page. This suggests a significant disconnect between what developers think AI needs (structured metadata) and what AI actually consumes (the rendered user experience).
Limitations of the Study
It is important to address what the report doesn’t cover.
- The "Already Visible" Bias: Every page in the study was already receiving 100+ citations from AI. These pages were already "in the system." The study does not answer whether schema helps a "zero-visibility" page get indexed or crawled for the first time.
- The Speed of Change: The 30-day window might be too short to observe long-term indexing shifts.
- The "Everything Else" Factor: When webmasters add schema, they often clean up their site code, improve page speed, or update content. Separating the "clean-up" effect from the "schema" effect is nearly impossible in a real-world environment.
Implications for SEO Strategy
So, should you stop using schema? Absolutely not.
The report focuses exclusively on AI citations. It does not address the vital role schema plays in Google’s traditional SERPs, such as:
- Rich Snippets: Star ratings, event dates, and recipe cards that increase click-through rates (CTR).
- Knowledge Graphs: Providing structured entity information that helps Google understand who you are and what your brand represents.
- Semantic Clarity: Even if AI models aren’t "reading" the schema during a retrieval event, schema remains a best practice for search engine crawlers to map relationships between entities.
The takeaway is not that schema is useless; the takeaway is that schema is not a "magic bullet" for AI traffic. If your goal is to be cited by ChatGPT or Google’s AI Overviews, this study suggests you should stop obsessing over JSON-LD and return your focus to the fundamentals: originality, authority, and content quality.
Conclusion: A Shift in Perspective
The Ahrefs report serves as a reality check for the SEO industry. In the race to appease the algorithms of the future, we have often relied on technical "hacks" in the hope of gaining an edge. However, the data suggests that AI models are becoming increasingly sophisticated at interpreting the content of a page directly, rather than relying on the structured tags we leave for them.
As we look toward 2025 and beyond, the focus of SEO will likely shift further toward "human-centric" metrics. If the machines are indeed ignoring our carefully crafted JSON-LD in favor of the raw, visible information on our pages, then the most effective way to earn an AI citation remains the same way we have always earned a reader’s trust: by providing the most accurate, well-cited, and insightful information available.
Schema will continue to have its place in the developer’s toolkit, but perhaps it is time to stop viewing it as a shortcut to AI authority. The data is clear: if you want to be cited, be worth citing.
Dwi Wanna
Evan Lee Salim
