Revolutionary Insights: Why Schema Markup Doesn’t Influence AI Search Understanding – A Deep Dive by Mark Williams-Cook
In the evolving landscape of search engine optimization (SEO), many practitioners have held the belief that implementing structured data or schema markup directly enhances a webpage’s visibility in AI-driven search results. However, recent insights from leading SEO expert Mark Williams-Cook challenge this assumption, revealing that current large language models (LLMs) do not utilize schema markup during their processing of web content.
The Role of Large Language Models and Tokenization
Understanding this paradigm shift begins with exploring how LLMs operate. As Mark Williams-Cook elucidates, LLMs like GPT-4 work through a process called “tokenization.” This involves breaking down text into smaller units called tokens—common sequences of characters that the model recognizes as distinct elements.
For example, the schema snippet:
json
{
"@type": "Organization"
}
When processed, is not understood as a cohesive block. Instead, tokenization dissects it into multiple parts—such as "@type", "Type", and "Organization"—these individual tokens are then used by the model to predict subsequent text or extract meaning. Consequently, the structure of schema markup becomes fragmented, effectively “destroyed” during tokenization, and does not serve as a meaningful signal to the AI when interpreting content.
The Experiment That Challenges Prevailing Assumptions
Mark Williams-Cook’s experiment, which garnered attention across platforms like LinkedIn and X (formerly Twitter), involved analyzing whether schema markup influences an AI model’s understanding of webpage content. His findings are striking: adding structured data or schema markup does not enhance a page’s visibility in AI-powered search results.
This conclusion aligns with observations from the SE Roundtable community, which notes that despite widespread beliefs, structured data currently does not appear to impact AI search rankings or visibility in a significant way. The key insight is that because LLMs process content based on token patterns rather than markup semantics, schema does not confer the benefits traditionally associated with SEO in the context of AI search.
Implications for SEO Strategy
For digital marketers and SEO professionals, these findings prompt a reassessment of the value placed on schema markup solely for AI-driven search visibility. While structured data remains useful for traditional search features—such as rich snippets, knowledge panels, and enhanced listings—it does not inherently influence how AI models interpret page content in terms of ranking or relevance.
Instead