Your Notes Are Already AI-Ready

In February 2026, Cloudflare shipped a feature that converts web pages from HTML into Markdown so AI agents can read them. One of its own posts dropped from 16,180 tokens to 3,150 — an 80% cut ¹. The surprise is not that the web is converting. It is that plain-Markdown notes skipped the step the web is now paying for.

Cloudflare named the trend plainly: "Markdown has quickly become the lingua franca for agents and AI systems as a whole" ². The pattern predates them. In September 2024, Jeremy Howard proposed llms.txt — a Markdown file websites publish for language models — because "context windows are too small to handle most websites in their entirety" ³. Two independent efforts, eighteen months apart, reaching for the same plain-text format. That is not a fad. It is a format settling into place.

What Cloudflare actually shipped

Cloudflare's "Markdown for Agents," launched February 12, 2026, lets a client request a page as Markdown instead of HTML ¹. The client sends an Accept: text/markdown header; the converted response carries an x-markdown-tokens count ¹. The point is cost. Tokens are what AI reads, and Markdown carries the same meaning in far fewer of them.

The arithmetic is concrete. A simple ## About Us heading costs roughly 3 tokens in Markdown; the HTML equivalent — <h2 class="section-title" id="about">About Us</h2> — burns 12 to 15 ¹. Multiply that across a page of nested tags and you reach the 16,180-versus-3,150 gap ¹. Cloudflare's own metaphor is the sharpest line on the page: "Feeding raw HTML to an AI is like paying by the word to read packaging instead of the letter inside" ⁴.

The number that matters

Read the 80% figure honestly — it is a vendor measuring its own feature, on one page ¹. Take it as "Cloudflare measured this," not "studies prove it." But it is reproducible: the x-markdown-tokens header lets anyone check the count ¹. And the reason does not depend on Cloudflare. HTML mixes content with presentation; Markdown carries structure as plain text.

That distinction is the whole argument. The web's content was written for browsers, then dressed in markup. To feed it to a model cheaply, someone has to undress it again. Your notes were never dressed. If you write in Markdown, a heading is ## and a list is a dash — structure and text are the same thing, already in the format the conversion step is reaching for.

The skeptic has a point, and it is not your point

The honest objection comes from inside Google. In February 2026, Google Search Advocate John Mueller posted on Bluesky: "Converting pages to markdown is such a stupid idea. Did you know LLMs can read images? WHY NOT TURN YOUR WHOLE SITE INTO AN IMAGE?" ⁵. He was sarcastic, agreeing with Jono Alderson that flattening a web page into Markdown strips structure it relied on ⁵.

He is right about the web's conversion problem. A modern web page is layout, scripts, navigation, and interaction; squeeze it into Markdown and some of that genuinely dies. But that critique is about conversion — and conversion is the step you never took.

Your notes were born in Markdown, written by a human in the format itself. There was no HTML, so there is no lossy flattening. The structure is native, not reverse-engineered. Mueller's warning lands on the web's mess. It misses a note that was clean from the first keystroke.

This is also the line not to overclaim. Markdown does not win every benchmark — table-heavy extraction is a real edge case, and HTML keeps some structure Markdown discards. The point is narrower and harder to argue with: Markdown is the format the ecosystem is standardizing on, and your notes are already in it. For the same conclusion from the lock-in angle, see Markdown vs Proprietary Formats.

Already AI-ready, still readable in 2040

A common fear runs the other way: that Markdown itself could vanish, leaving .md files no program can open. The format answers this by being almost nothing. Markdown is "a plain text format for writing structured documents, based on formatting conventions from email and usenet" ⁶. There is no proprietary container, no vendor whose collapse takes your files down.

It is also too widespread to revoke. CommonMark lists Reddit, GitHub, GitLab, Stack Overflow, Discourse, and more among its adopters ⁶; no single owner can switch it off. And it has already survived. John Gruber released Markdown 1.0.1 on December 17, 2004, with one stated goal: "The overriding design goal for Markdown's formatting syntax is to make it as readable as possible" ⁷.

A file written that day opens unchanged in any text editor now. The property that makes a note cheap for a model to read in 2026 is what lets a human read it in 2040 — our plain-text love letter makes the longer case.

Frequently Asked Questions

Why does AI prefer Markdown over HTML?

Markdown carries the same structure in far fewer tokens, and tokens are what AI reads. Cloudflare measured one of its own pages at 16,180 tokens in HTML versus 3,150 in Markdown — an 80% reduction ¹. A ## About Us heading costs about 3 tokens; its HTML equivalent burns 12 to 15 ¹. Less packaging, same meaning.

Is Markdown actually better for AI, or is that hype?

It is a real, measured advantage, with a caveat. Cloudflare's 80% figure is a vendor measuring its own feature, so treat it as "Cloudflare measured" rather than proof ¹. Markdown does not win every benchmark; table extraction is a known edge case. But the ecosystem is standardizing on it: Cloudflare calls it the "lingua franca for agents" ².

Are my Markdown notes future-proof for the next 20 years?

Markdown is plain text, "a plain text format for writing structured documents" ⁶, so a .md file opens in any text editor with no special software. A Markdown 1.0.1 file from December 17, 2004 still opens unchanged today ⁷. CommonMark is an open spec adopted by Reddit, GitHub, Stack Overflow, and others ⁶, so no single owner can discontinue it.

What is Markdown, and why do LLMs read it well?

Markdown is a plain-text writing format that uses light conventions (like # for headings and - for lists) drawn from email and usenet habits ⁶. Models read it well because structure and content are the same plain text, with no tags to wade through. That is why Cloudflare converts HTML pages into it before feeding them to agents ¹.

Do I need an AI feature to make my notes AI-ready?

No. AI-readiness is a property of the format, not a feature you switch on. If your notes are plain Markdown, they are already in the format conversion tools are reaching for. If you later choose to point an AI at them, that is a bring-your-own-key choice you make, not a precondition for your notes being ready. For the workflow that turns that format into a working AI memory, see Markdown Notes as AI Memory.

Is Markdown a good format for notes?

Yes, for two reasons that now reinforce each other. It is durable: plain text, an open spec, readable in any editor for decades ^{6 7}. And it is AI-ready: the format the web is converting toward, with no conversion needed on your side ¹. For non-coders, Markdown for People Who Don't Code shows how little there is to learn.

The web is spending engineering hours to become what your notes already are. Write in plain Markdown and you are not preparing for the AI era — you are already in the format it reads. In MNMNOTE, notes are stored locally on your own device, in open Markdown, readable in any editor and ready for any model you point at them.

mnmnote.com

References