"In my opinion, this test shows that the LLM agent is simply picking up whatever you are listing in the HTML. It does not matter if it is valid schema. If the system interprets the text as relevant to the prompt, it is included. It would therefore indicate that schema is *not* being used in the explicit sense it was designed for with those systems," he wrote."
"On the page about DUCKYEA t-shirts, he did not post the fake company's address. Instead, he put the address within made up JSON-LD schema markup. Then he waited and prompted both ChatGPT and Perplexity. Both ChatGPT and Perplexity read the fake and made up schema to find the address. Since the schema was not valid, he figured that it was just being read by these AI engines like any other page of text on the web."
A test created a fake company, DUCKYEA t-shirts, and placed the company address only inside deliberately invalid JSON-LD schema markup while leaving no address in visible page content. Both ChatGPT and Perplexity returned the address when prompted, indicating those LLM-driven systems read schema text as plain webpage content and used it to answer queries. OpenAI notes use of structured data feeds for shopping results, Google says the effect depends, and Microsoft describes schema's role in Copilot. The result suggests schema remains useful but not always treated as exclusive machine-readable signals by these LLMs.
Read at Search Engine Roundtable
Unable to calculate read time
Collection
[
|
...
]