Experiments by SEO experts reveal that structured data, long trusted for better rankings in traditional search, does not affect AI-driven search results. Large language models strip schema markup of its meaning, leaving only visible text as the deciding factor.
Search engine optimization has always depended on signals that help machines understand content. Structured data, also known as schema markup, has become a favored tactic because it provides Google with context about products, organizations, and reviews. The payoff was visible in rich snippets and featured listings.
The rapid rise of AI search is transforming the way we find information. Recent tests by SEO professionals Mark Williams-Cook and Julio C. Guevara, show that structured data offers no advantage in this new environment. Instead, the only information that matters is the content users can actually see.
The Williams-Cook Experiment
Mark Williams-Cook, an SEO consultant known for testing assumptions in the field, ran one of the first controlled experiments.
He presented a clear illustration on LinkedIn, showing what happens when a large language model processes schema markup.
LLMs break down text into tokens. A label like “@type”: “Organization” does not survive intact during this process. Instead, the words “type” and “Organization” are treated as ordinary text, divorced from their context as schema.
Williams-Cook explained that this makes the markup effectively useless for models trying to generate responses.
In other words, schema markup might exist on the page, but by the time it reaches the training process, it has been fragmented and reduced to meaningless tokens. The AI does not treat it as a structured signal at all.
Here’s the screenshot he shared on LinkedIn:
Putting Theory to the Test
Julio C Guevara ran a complementary experiment to see how this plays out in real-world conditions. He created two fake product pages for an invented item.
Page one: contained both visible text and structured data describing the product.
Page two: contained only structured data, with no visible text at all.
He then asked AI systems like Gemini and ChatGPT to extract details such as price, colors, and SKU numbers from both pages.
The outcome was that only the page with visible text produced correct answers. The page that relied solely on schema markup failed completely, as if the structured data were invisible.
“We tried hundreds of prompts,” Guevara wrote, “and the LLMs simply couldn’t see the text within the structured data.”
These results echo questions I have already explored in one of my recent news articles, where I examined whether structured data could support AI search indirectly through knowledge graphs and metadata. The new experiments add clarity by proving that large language models overlook schema markup entirely and depend only on visible text.
What This Means for SEO Strategy
The study shows that structured data continues to play a role in Google’s traditional search engine results. Rich snippets, carousels, and knowledge panels often depend on schema markup. That makes it valuable for click-through rates and visual presentation.
AI-driven search works differently. Tools powered by LLMs are not designed to read structured markup. They only learn from the text that appears as part of the page itself.
If businesses want their content to show up in AI responses, the writing must be clear, descriptive, and easy to parse directly from the page.
Why Models Ignore Schema
Large language models are not semantic databases. They learn by analyzing patterns in billions of pieces of text.
Tokenization, the process that breaks sentences into smaller parts, strips structured data of its labels. A model then treats those pieces as random characters rather than as connected signals.
That is why schema markup becomes background noise rather than a helpful clue. Even when present in training data, it has no special status in shaping how the model interprets meaning.
Could the Situation Change
SEO professionals point out that these are early days.
AI search is evolving quickly, and developers may find ways to incorporate structured data in future versions.
Schema is still a powerful way to express relationships between entities, and ignoring it entirely could limit accuracy in areas like product catalogs, recipes, or event listings.
Currently, however, the evidence is conclusive. If site owners want language models to understand their content, visible text is the only reliable pathway.
Practical Advice for Site Owners
If you’re managing a website and want to prepare for AI-driven search, here are the steps you can take right now:
- Prioritize clear, visible content. Write product details, specifications, and descriptions in plain HTML text where users and crawlers can read them.
- Use structured data for Google, not AI. Keep your schema markup, since it can still help with rich results in traditional search. But don’t assume it influences AI outputs.
- Test your content. Run your pages through popular AI models and see what information they extract. If the model misses key details, consider rewriting your text more explicitly.
- Stay updated. AI search is moving quickly. What doesn’t matter today may become relevant tomorrow if model developers start incorporating schema in new ways.
- Think beyond markup. Build authority through high-quality writing, backlinks, and strong topical coverage, which are far more likely to influence AI-driven visibility.
Key Takeaways
- Structured data has no effect on AI search visibility based on current tests.
- Tokenization breaks schema into meaningless parts, erasing its structure.
- Experiments confirm that models only use visible text when generating answers.
- Schema remains useful for traditional Google search features.
- Site owners should focus on clear, accessible writing to prepare for AI search.
Dileep Thekkethil
AuthorDileep Thekkethil is the Director of Marketing at Stan Ventures, where he applies over 15 years of SEO and digital marketing expertise to drive growth and authority. A former journalist with six years of experience, he combines strategic storytelling with technical know-how to help brands navigate the shift toward AI-driven search and generative engines. Dileep is a strong advocate for Google’s EEAT standards, regularly sharing real-world use cases and scenarios to demystify complex marketing trends. He is an avid gardener of tropical fruits, a motor enthusiast, and a dedicated caretaker of his pair of cockatiels.