Despite how simple it seems, Markdown is one of the most talked-about languages these days. This lightweight markup language has seen a rise in popularity, especially after AI tools like Claude became popular. Though Markdown isn't the only lightweight markup language available today, it enjoys a special status that other options don't.
However, despite the success Markdown has enjoyed, it has had a strange history, including a near-controversial origin story. Here's a quick look at how Markdown went from a small project to a pillar of the AI world as we know it. It should also help you understand why many AI models are, in effect, fixated on this decades-old markup language.
Markdown had two authors, but one rarely gets mentioned.
Gruber got the byline, though Swartz wrote a lot of the code.
John Gruber is credited with inventing Markdown, but this narrative isn't without controversy. Aaron Swartz, the computer programmer/Internet hacktivist, also contributed to the project that later became Markdown when he joined as a beta tester. In addition to helping with the core code, Swartz is said to have created an HTML-to-Markdown converter tool. However, Gruber later said he was the sole creator of Markdown, not a co-creator, adding that "Aaron was my sounding board, my muse."
This way, even though Markdown had two authors, one is barely mentioned. For Gruber, however, the motivation for creating Markdown was simple, and something we can relate to: he was tired of following the HTML syntax to write blog posts. According to Gruber, he created Markdown to allow web writers to create text content using simple, readable, plain-text formatting syntax instead of worrying about HTML. The Markdown system will then convert that text into HTML or XHTML.
The blog post in which Gruber introduced Markdown as a standalone Perl script or a plug-in is still available. You can still check out the official blog post and the online demo page, which includes a Markdown/HTML/Preview interface. However, this simple tool that Gruber started was in for a surprise in the coming years.
No spec for a decade, then a naming fight
Every platform diverged, and the fix nobody endorsed is called CommonMark
As we said earlier, Gruber posted a brief blog post and a couple of pages to let people explore Markdown. The HTML-to-Markdown converter also included specifications for this lightweight markup language. This Perl script and the description page remained the only reference point for a decade (until 2014). This was not standardized documentation, and there were ambiguities and confusion. This is where players like GitHub, Reddit, and Stack Overflow came into the scene.
The Markdown that Gruber built was simple and lacked many of the features we are familiar with today. Things like code fences and nested lists were created as a result of various implementations by entities like GitHub and Stack Overflow. Meanwhile, there were calls to standardize Markdown, and the most important one was first called Standard Markdown but later renamed CommonMark. Jeff Atwood and John MacFarlane, the people behind CommonMark, wanted it to be "a standard, interoperable and testable version of Markdown."
Gruber, however, wasn't a fan of what CommonMark and never endorsed it. Because Gruber owns the copyright for Markdown, CommonMark remains one of many implementations, not an official thing. By this time, there were many other implementations from different parts of the developer community as well.
When Markdown stopped being a developer thing
Thanks to Notion, Obsidian, and Discord, it quietly stopped being niche.
In earlier years, Markdown was primarily used by developers. You were likely to find Markdown used for GitHub README files and other documentation. Some static website generators, such as Jekyll, also use Markdown. However, over the past 10 years, Markdown has gone from a developer thing to more mainstream. And we may have to thank note-taking and second-brain apps like Notion and Obsidian for this change.
Though Notion had been using a version of Markdown as part of its default editing experience, Obsidian played a larger role in making Markdown a go-to option for non-technical teams and individuals interested in writing and note-taking. Other platforms, such as Discord, were also instrumental in popularizing Markdown as a markup language. You may also have noticed that Markdown is now part of many tools and apps, including writing apps like Ulysses and Bear.
Now, something interesting here is that this wasn't a coordinated adoption. Instead, these platforms chose Markdown as the go-to option because it was simple and accessible. This simplicity and broader adoption rate would also put Markdown in the position it is now.
Why AI models default to a 2004 syntax
The whole internet was already written in it
You may have noticed that AI platforms and models have an affinity towards Markdown! Platforms like Claude and ChatGPT tend to provide responses in Markdown, and many of Claude's customization options also rely on this simplified markup language. However, it is not because Anthropic or OpenAI chose Markdown, but rather because the AI training data had a disproportionate amount of Markdown content.
Because the training data used organized text in Markdown, AI models also learned to use this format, making it a default option for many. It should be noted that platforms like Claude or ChatGPT do not explicitly follow Markdown as a written-down syntax. Instead, they simply reproduce text in the way the training exposed them to. Therefore, the reliance of AI models on Markdown is another testament to Markdown's popularity.
While at this, we must also appreciate the novelty of a simple project becoming the definitive characteristic of some of the most used tools of the decade.
The future of Markdown
Markdown has undoubtedly been through a happening journey, filled with highs, lows, and controversies. However, what happens to a markup language as AI-generated content becomes the norm on the internet is a different question. Sure, we are at a point where Markdown has some privileges, but it may not last much longer. It is also possible that, given how Markdown has survived for a couple of decades without a norm, a standardized set of norms could help it remain for a long time.