Schema Is Not Dead — Stop Listening to Bad Takes
A viral experiment claimed schema markup is useless for AI. Here’s why that conclusion is dangerously wrong.
There’s a lot happening in the agentic commerce space right now — and I’ll get to the OpenAI news next week — but a post has been doing the rounds this week that I think is worth addressing directly, because the takeaway people seem to be drawing from it could do real damage to your website.
On the face of it, the post is saying that schema markup is useless. I don’t think that’s what the author actually intended, but that’s certainly how people are reading it.
What the experiment actually showed.
SEO Mark Williams Cook ran an experiment where he created a fictional business with a fictional website, took a real physical address, and embedded it inside fake schema markup on the page. He then went to Perplexity and ChatGPT, asked for the address of the fake business, and — sure enough — both LLMs returned the correct address.
The conclusion being drawn by some is that because the LLMs found the address even with fake schema present, schema doesn’t matter. Some people have gone as far as recommending you reduce your schema efforts entirely. I think that’s unbelievably bad advice.
What the experiment actually proved.
Here’s the thing: the LLMs found the content. Yes, the schema was fake — but the content itself was still structured on the page. It was readable and clear. What the experiment demonstrates isn’t that schema is irrelevant; it’s that well-structured, readable content can be understood by machines even without perfect markup.
Content is the story. Schema is the index.
Those are two different things and conflating them is where this conversation goes wrong. The experiment was really aimed at a subset of the AIO / GEO / EIAO crowd who treat schema as a magic substitute for thin content. And on that point, I agree — schema doesn’t replace great content. But that’s a very different claim from “schema doesn’t matter.”
Why schema still matters — especially at scale.
Think about a site with a thousand products. Weight, size, colour, compatibility with specific vehicle models — all of that detail might exist somewhere on each page, but if it’s written differently by five or six different people updating products, you end up with a completely inconsistent set of pages. Different standards, different formats, different terminology. Schema is what organises that information and enforces consistency across the whole catalogue.
We’ve been preaching good document structure for 27 years in SEO. Things like alt tags — a lot of SEOs ignore them, but we’ve always used them, because they’re good document structure. They’re good for screen readers. And now they’re useful for limiting the number of tokens an LLM needs to process a page, because a properly described image doesn’t need to be fully ingested.
The same principle applies to your entire site. Structured, machine-readable content reduces hallucinations and mistakes — which is exactly why all of the AI agents we run internally operate from structured skill files written in Markdown. Plain text, but organised. That structure is what stops the agent from guessing.
The piece most people are missing: entity connection.
Thinking of schema purely as something Google uses to create rich results in search is missing the bigger picture entirely. The place where schema is genuinely critical right now is entity connection — communicating to machines that this product is related to that product, that this part fits that model, that these items belong to the same category.
That relational layer is extraordinarily hard to communicate in plain unstructured text at scale. Schema is the tool that makes it possible.
When in doubt, go back to basics.
When you’re reading takes on this topic and getting confused by the noise, come back to a simple question: how is a machine going to understand what this product or page is about? Do I have consistency across similar pages? Does my schema communicate how this thing relates to everything else on my site?
Good content means content that solves the user’s problem in the shortest time possible, with the least cognitive load. Schema supports that goal — it doesn’t replace it, but it absolutely enables it at scale.
Pay attention to your schema. It matters — despite what you might be hearing out there.
Got thoughts on this? Join the discussion on LinkedIn, X, or Facebook. And if you’d like to beta test SKAW on your site, get in touch — I’m looking to walk people through how it works, which will also inform our tutorial videos. You can reach out to me at jim@stewartmedia.biz
The post Is Schema Dead? Why That Viral Experiment Got It Wrong. appeared first on StewArt Media.