ChatGPT & Perplexity Treat Structured Data As Text On A Page

02/06/26
Source: Search Engine Roundtable by barry@rustybrick.com (Barry Schwartz). Read the original article

TL;DR Summary of Structured Data’s Role in AI and LLMs: What Recent Tests Reveal

Recent experiments show that AI engines like ChatGPT and Perplexity do not utilize structured data (schema) in a uniquely effective way. These models appear to read schema markup as plain text rather than interpreting it with special significance. While companies like OpenAI, Google, and Microsoft acknowledge some benefits of schema, the practical impact on AI understanding remains limited. The key takeaway is that schema is helpful for SEO fundamentals but not a magical solution for AI comprehension.

Optimixed’s Overview: Understanding the Real Impact of Structured Data on AI Language Models

Background and Experimentation

Mark Williams-Cook conducted a test with a fictitious company named DUCKYEA t-shirts to explore if AI language models use structured data differently than regular webpage text. By embedding the company’s address only within a fabricated JSON-LD schema markup (not visible on the page itself), he tested how ChatGPT and Perplexity would respond when prompted for the address.

Key Findings

  • Both ChatGPT and Perplexity extracted the address from the fake schema, despite it being invalid.
  • This suggests that these AI models treat schema markup as part of the page text, without validating or interpreting it as structured data.
  • Therefore, the AI does not leverage schema in the explicit, intended manner designed for search engines or data feeds.

Industry Perspectives

While OpenAI has acknowledged using structured data feeds for shopping results, and Google’s John Mueller states the usefulness of schema depends on context, Microsoft has highlighted schema’s role in assisting Copilot functionality. Despite this, the consensus from tests like Williams-Cook’s is that schema markup is primarily valuable for traditional SEO rather than providing a distinct advantage in AI language model comprehension.

Conclusion

Structured data remains a good practice for SEO and content organization. However, claims that schema offers a magical improvement in how AI engines understand or utilize web content lack strong evidence. As Mark Williams-Cook summarized: schema is beneficial, but repackaging basic SEO principles as revolutionary AI tools is misleading.