Same toolbox, different jobs.
Markdown and llms.txt keep getting lumped together. They're not the same tool. Grabbing the wrong one for the wrong job is a stripped thread waiting to happen.
There's a conversation running hot through every corner of the internet about how to make your content readable by AI. I keep seeing the same two tools get conflated — Markdown files and llms.txt. People treat them like they're interchangeable, or that one replaces the other, or that doing one means you've covered both.
They're not the same. They don't solve the same problem. Mixing them up is like grabbing a torque wrench when you need a thread tap — both live in the same roll cart, but try using one for the other's job and you'll strip something expensive.
Here's how to keep them straight.
Markdown: clean records that anyone can read.
Serving your posts or pages as plain Markdown — usually at the same URL with a .md extension, like yoursite.com/post.md — is the equivalent of keeping a standardized, legible inspection sheet instead of a handwritten ticket that only you can decode.
When an AI crawler hits your page as HTML, it gets the full document: nav bar, footer, cookie banner, sidebar ads, whatever JavaScript decided to inject that afternoon. It has to work through all of that noise to get to your actual content. When it hits the same page as Markdown, it gets the words. That's it.
This post is written in Markdown. Here's what that actually looks like under the hood:
Markdown source## My actual heading
This is the paragraph I want you to read.
- First point
- Second point
- Third point
Versus the HTML version of that same content, buried inside <div class="content-wrapper-inner-v2"> with a sticky header, three ad slots, and a related-posts widget fighting for the same DOM space.
Tools like ChatGPT, Claude, and Perplexity parse Markdown faster and more accurately than raw HTML. The content isn't changing — the readability is. Same information, cleaner transfer.
Think of it like digitizing your ROs into a consistent format instead of letting every tech develop their own shorthand. The job story is the same. One version is legible to everyone in the shop. One version requires you to have worked there for three years to interpret.
llms.txt: your shop's identity card, not your service history.
llms.txt is a different animal entirely.
It lives at your site root — yoursite.com/llms.txt — and it doesn't describe any one piece of content. It describes you. Think of it less like a service record and more like the placard on your bay door that tells a first-time customer what you specialize in before they ever walk through.
# Your Name
> One sentence about what you do and who you do it for.
## What to read
- [Your best post](/post-slug)
- [Your about page](/about)
- [Your services](/work)It's a curated map of your site — not every URL, just the stuff you actually want AI to know about you. Your beat. Your voice. Your best work. The context that helps a model understand who you are when someone asks it a question you'd want to show up in.
llms.txt is your reputation in the referral network. It's what makes someone say "go to that shop, they know their stuff" before anyone's looked at a single record.
Do both — but start with llms.txt.
Here's why the order matters.
AI assistants are increasingly the discovery layer. When someone asks Claude or ChatGPT "who writes clearly about AI for the auto service industry?" or "who does good independent diagnostic work in New England?", the model needs context about you — not just clean text from one post. It needs to know your beat, your location, what you're actually good at.
That's not in any individual post. That's in llms.txt.
Markdown versions of individual posts then make sure that when an AI does pull a specific article — because someone asked about your topic, or because you got cited somewhere — it gets the cleanest possible version of your thinking. No nav cruft. No ads. Just the argument.
The reputation that makes someone say "go to that shop." Drives discovery. Tells AI who you are before it even reads a word you've written.
The clean build thread that proves it when they go looking. Handles the read once discovery has already worked. No noise, just content.
One drives discovery. One handles the read after discovery worked.
A shop with spotless service records but no signage is invisible to new customers. A shop with great signage but sloppy records loses the customer once they're in the bay. You need both. Start with the sign.
Practical setup.
Neither of these takes long.
For llms.txt:
Create a plain text file and drop it at your site root. Write a one-sentence bio, link to your best five posts, link to your about and contact pages. That's it. Fifteen minutes, done, costs you nothing.
For Markdown versions of posts:
- Static site generators — 11ty, Astro, Hugo — can serve
.mdalongside HTML with one config change - Ghost has plugins for it
- Fully managed CMSs like Squarespace require a workaround; you may need to serve these separately
- At minimum, keep your HTML clean and semantic: real
H1tags, proper paragraph structure, alt text on images — crawlers still parse HTML, and clean structure still helps
llms.txt is a proposed convention, not an official standard yet. Not every AI system reads it. But the platforms that do are the ones driving discovery right now, and it costs you basically nothing to add. When a convention is free to implement and the downside is zero, you implement it.
The short version.
Markdown makes individual content cleaner to ingest. llms.txt tells AI who you are in the first place.
They're not competing. They're not interchangeable. They're two tools that solve two different problems, and you need both of them.
Get the sign up first.