Google’s John Mueller and Martin Splitt suggested caution about using markdown as a solution for optimizing for AI search, saying that it needlessly complicates something that is essentially simple.

Markdown Is A Poor User Experience

The first point that Google’s Martin Splitt touched on is that markdown by itself may not be a good user experience. He referenced that HTML layouts offer the opportunity to create positive user experiences with visually pleasing layouts and colors, something that markdown does not support.

Martin explained:

“And I mean, the other thing is also for users, you can’t just publish a set of Markdown documents because A, we like colors and images and stuff to kind of like flow in a nice layout and Markdown by definition, unless you put a layout on it, doesn’t. And Markdown doesn’t support layouts directly.

So you would have to have some sort of mechanism to… You’re basically recreating the browser. You’re recreating HTML parsing in the end. So might as well use HTML parsing because as you say, that has been around, that has been tried and tested for decades at this point.”

Markdown Creates Twice The Work

The other point Martin made was that using markdown for LLMs while also creating a separate HTML version for users doubles the amount of work and complicates the act of web publishing, which is the opposite of what SEOs and publishers should be doing: simplifying the act of web publishing.

Martin continued:

“The other thing is you would duplicate things if you were to acknowledge, like, users don’t want Markdown. They want the full-fledged website. And then I create a version just for LLMs, then you’re kind of making twice the work or having twice the work, no?”

John Mueller agreed and expanded on the topic by saying that he understood where people are coming from in terms of simplifying the process of publishing content because some HTML pages can actually be poorly presented.

Mueller added:

“Yeah, I think that’s always terrible on the web. And I understand where these ideas come from in that a lot of web pages are just terrible from a structural point of view and hard to use. And it’s tempting to say, well, users can see this complex, weird page, and automated systems, they should have it easy. You should just give them the information that they’re looking for.”

Something that they didn’t mention but is implied in what they were talking about is that humans evolved to primarily prioritize visual information; it’s the dominant way humans perceive the world.

According to scientists:

“…half of the human brain is devoted directly or indirectly to vision…”

That means communicating with images and attractive layouts can be beneficial for getting a message across.

Parallel Versions Of Content

Lastly, both Mueller and Splitt cautioned against having parallel versions of content because it needlessly complicates the act of publishing. Moreover, because an AI won’t email you to tell you that the markdown version of a web page is broken (the way a user might if your HTML is broken), it’s possible for the machine-facing version of the content to linger in a broken state for weeks or longer without the site owner catching on.

Mueller began this part of the discussion:

“Fundamentally, as soon as you have these parallel versions of your content, then everything becomes so much more complex. You have to maintain those multiple versions. You have to make sure that nothing breaks on a version that a user doesn’t see, because users might complain to you if your page doesn’t load properly. But if the LLM version of a page doesn’t load properly, then no user is going to tell you that something is broken.

And a lot of these automated systems, might not even recognize that something is broken because they see, it’s like, there’s some text here, must be what they want us to index.”

Martin Splitt agreed:

“Yeah, I think we learned that lesson with dynamic rendering, which was a nice stopgap solution for a while. But we found out in practice it oftentimes caused more problems and was really hard to debug because of this duality of the two different separate versions. Yeah, that’s not great.”

Takeaways

Google’s John Mueller and Martin Splitt cautioned against using markdown as a separate AI-optimized version of a website, making the point that publishers are better off improving their existing HTML pages rather than building parallel AI-specific versions of content.

• Google says markdown for AI SEO may not be optimal because it may lead to complications related to publishing parallel sets of content, adding complexity without concomitant benefits.
• Parallel content development is difficult to debug because failures in AI-facing versions can go unnoticed for long periods, unlike broken user-facing pages.
• Markdown content for both users and AI may not present the best experience for users. Although they didn’t mention it, the user experience is a real ranking-related factor, both directly and indirectly.
• HTML provides important advantages for human usability through layout, navigation, colors, and images, which help users consume information more effectively than raw markdown.
• Google compares parallel content publishing to dynamic rendering, suggesting that past attempts to maintain separate machine-optimized versions often created more problems than they solved.

Listen to search off the record here, starting at about the 14 minute mark:

Featured Image by Shutterstock/Krakenimages.com