AI systems, including ChatGPT and Perplexity, appear to treat structured data embedded in webpages as ordinary text rather than as a distinct machine-readable signal, according to a new test shared by an SEO researcher, adding to uncertainty about how such models interpret schema markup.
A recent experiment suggests that large language models may not interpret structured data in the way publishers and search professionals expect.
Mark Williams-Cook, an SEO consultant, said he tested how ChatGPT and Perplexity handle schema markup by creating a fictional clothing brand, DUCKYEA T-shirts, and publishing a webpage for it. The page did not display the company’s address in visible content. Instead, the address was placed only inside JSON-LD schema markup.
The schema markup was intentionally invalid.
When both AI systems were asked about the company’s address, they returned the fabricated information contained in the schema, despite its lack of validity. Williams-Cook said the results suggest the models read schema as part of the page’s text rather than processing it as structured data with defined rules.
Test Points to Text-Based Extraction
Williams-Cook said the outcome indicates that the models appear to extract information based on relevance to a user prompt, without checking whether the schema complies with formal standards.
“In my opinion, this test shows that the LLM agent is simply picking up whatever you are listing in the HTML,” he wrote in a LinkedIn post discussing the findings. “It does not matter if it is valid schema.”
The experiment builds on earlier testing he has shared publicly, which also questioned whether schema markup provides any special advantage when content is accessed by AI-driven systems rather than traditional search engines.
Mixed Messages From Major Platforms
The findings come amid growing discussion about how structured data fits into AI-powered search and answer systems.
OpenAI has previously said that it uses structured data feeds for some shopping-related features. Google’s search team has stated that schema can help in certain situations, depending on context. Microsoft has also outlined how structured data may support its Copilot product.
However, those statements generally refer to curated feeds or platform-specific integrations, not to how AI models interpret schema embedded directly in web pages.
Williams-Cook’s test focuses on that distinction, suggesting that when schema appears in page HTML, AI systems may not treat it differently from any other text.
Why This Matters for Publishers and SEOs
The test highlights a risk that many site owners may be overlooking. If AI systems read schema markup the same way they read page text, inaccurate, outdated, or experimental markup can appear in AI-generated answers, even when that information is not visible to users.
The findings also reinforce the point that clear, visible content continues to carry more weight than technical layers embedded in code. When AI systems generate responses, they appear to rely on information they can interpret as relevant and readable, rather than on how that information is formally marked up.
Practical Guidance for Site Owners
Here are several practical steps site owners can take in response to these findings:
- Schema markup should continue to be used to support established search features such as rich results and enhanced listings.
- Critical facts should not be placed only in schema markup if they are absent from visible page content.
- Schema should be treated as a supporting signal rather than a primary method for shaping AI-generated answers.
- Markup should be reviewed regularly to ensure it does not contain outdated, speculative, or test data.
- Key information should be presented clearly on the page itself, where both users and AI systems are most likely to rely on it.
Key Takeaways
- AI systems may read schema the same way they read page text.
- Invalid or hidden markup can still influence AI answers.
- Schema remains useful for search features, not AI shortcuts.
- Visible, well-written content continues to matter most.
- Claims about schema-driven AI optimization deserve scrutiny.
Zulekha
AuthorZulekha is an emerging leader in the content marketing industry from India. She began her career in 2019 as a freelancer and, with over five years of experience, has made a significant impact in content writing. Recognized for her innovative approaches, deep knowledge of SEO, and exceptional storytelling skills, she continues to set new standards in the field. Her keen interest in news and current events, which started during an internship with The New Indian Express, further enriches her content. As an author and continuous learner, she has transformed numerous websites and digital marketing companies with customized content writing and marketing strategies.
