A Brief History of Markdown: From Email Conventions to AI's Native Language
A Brief History of Markdown: From Email Conventions to AI's Native Language
In 2004, a blogger named John Gruber had a simple problem: he wanted to write for the web without typing HTML. He teamed up with programmer Aaron Swartz and created Markdown — a lightweight syntax that converted plain text into clean HTML. Neither of them imagined it would become the default language of the AI era.
The Problem Markdown Solved
Before Markdown, writing for the web meant either using a WYSIWYG editor that produced bloated HTML, or writing raw HTML by hand. Neither option felt natural. Writers wanted to focus on words, not tags.
Gruber's insight was that plain-text email already had informal conventions — people wrote **bold** to mean bold, used hyphens to make lists, and put asterisks around important words. Markdown simply formalized those habits into a consistent syntax and added a converter that turned them into HTML.
The original Markdown specification was published on Daring Fireball on March 19, 2004. The first converter was a Perl script.
Early Adoption: Blogging and Developer Tooling
Within a few years, Markdown spread through the developer community. It was clean, readable in its raw form, and easy to version-control with Git. Blogging platforms like Jekyll (2008) made Markdown the default writing format for static sites.
GitHub adopted Markdown in 2008 for READMEs, issues, pull requests, and comments. This was a turning point. Suddenly millions of developers were reading and writing Markdown every day without thinking of it as a "markup language" — it was just how you wrote on GitHub.
Stack Overflow, Reddit, and dozens of other high-traffic platforms added Markdown support. Each platform implemented it slightly differently, which created compatibility headaches.
The CommonMark Standard (2014)
By 2012, there were at least a dozen different Markdown implementations, each with different behavior for edge cases. Jeff Atwood (co-founder of Stack Overflow) wrote a post titled "The Future of Markdown" calling for a standardized specification.
In 2014, a group of researchers and developers published CommonMark — a rigorous, unambiguous specification for Markdown with a comprehensive test suite. John MacFarlane (author of Pandoc), Jeff Atwood, and others contributed. CommonMark became the foundation that most modern Markdown parsers are built on.
GitHub Flavored Markdown (GFM) followed in 2017, extending CommonMark with tables, task lists, and strikethrough — features the developer community needed most.
The Note-Taking Revolution
The 2010s brought a wave of note-taking applications that embraced Markdown as their native format: Obsidian, Notion, Bear, Typora, Logseq, and dozens more. Markdown appealed to these apps because:
- Notes are portable — plain
.mdfiles work everywhere - Notes are future-proof — no proprietary format lock-in
- Notes are readable raw — you can read a Markdown file in any text editor without rendering
Obsidian, released in 2020, built an entire knowledge graph system on top of plain Markdown files. It became one of the fastest-growing productivity tools ever made. For a comprehensive look at today's landscape, see our guide to the best Markdown apps in 2026.
Markdown Meets AI
The AI revolution of the early 2020s gave Markdown an unexpected new role. When users started feeding content to large language models like ChatGPT, Claude, and Gemini, a pattern emerged: Markdown inputs produced dramatically better outputs than HTML.
The reasons are structural. LLMs tokenize text, and Markdown carries almost no overhead compared to HTML. A page that takes 8,000 tokens in HTML might take 2,800 tokens in Markdown — a 65% reduction. More importantly, LLMs were trained on enormous amounts of Markdown (GitHub, documentation sites, Reddit) and understand it natively.
Today, Markdown is the de facto input format for AI workflows:
- AI tools like Claude and ChatGPT render Markdown in responses
- Developers use Markdown in system prompts and few-shot examples
- Web clippers like Web2MD convert pages to Markdown specifically for AI consumption
- The emerging
llm.txtspecification asks websites to publish Markdown summaries for AI crawlers (explored further in Will Markdown Become the Programming Language of the AI Era?)
A Timeline of Key Moments
| Year | Event |
|------|-------|
| 2004 | John Gruber and Aaron Swartz create Markdown |
| 2008 | GitHub adopts Markdown for READMEs |
| 2012 | Jeff Atwood calls for standardization |
| 2014 | CommonMark specification published |
| 2017 | GitHub Flavored Markdown (GFM) spec released |
| 2020 | Obsidian launches, Markdown becomes PKM standard |
| 2022 | ChatGPT launches; Markdown becomes AI input format |
| 2024 | llm.txt proposal emerges; Markdown as GEO standard |
Why Markdown Endures
Twenty years after its creation, Markdown remains dominant for a simple reason: it respects the writer. It adds just enough structure to be useful without getting in the way. You can write Markdown in Notepad. You can read it without rendering it. You can convert it to HTML, PDF, Word, or a dozen other formats.
In a world of increasingly complex software, Markdown's simplicity is a feature, not a limitation.
Want to convert any webpage to clean Markdown instantly? Try Web2MD — one click turns any article into AI-ready Markdown.