# Markdown versions

Markdown versions of linked pages (.md twins)

Also called markdown versions of pages or .md URLs.

**In short:** Serve a markdown version of each page your llms.txt links (page.md or page.html.md), or link markdown URLs directly.

- Category: Ecosystem Integration · weight 18 of 100 within the category
- Ecosystem Integration is worth 20 in technical, 15 in generic, regulated and commerce (points of 100, by scoring profile)
- Audit ID E3 (as shown on reports) · rubric v2.1

## What it checks

Whether the pages your llms.txt links to have clean markdown versions.

## Why it matters

A markdown version gives an agent the page’s content without navigation, scripts and ads — cheaper to read and easier to quote accurately. It is among the cheapest generative engine optimization wins available, because the content already exists.

## When it applies

Applies when at least one sampled page gave a definite answer. Unverified when none did (every check blocked, rate-limited or timed out) or when we couldn’t read any same-site page links in the file; Not applicable when the file has no URLs at all.

An audit that doesn’t apply, or that we couldn’t verify, is removed from your score and its weight shared among the Ecosystem Integration audits that do apply. It is never scored as a zero.

## How it’s scored

The share of sampled pages with a working markdown version: the first three distinct same-site pages linked from your llms.txt, each checked through its declared text/markdown alternate first, then the .md-appended form, then the extension-replaced form. A linked URL that is itself served as markdown or plain text counts as its own markdown version.

> **Measurement.** Every scan checks up to three pages, at most four candidate URLs each, with HEAD requests (a short ranged GET where HEAD is refused) and a 10-second budget for the whole sample. A 404 or 410 on every candidate is a miss; a 403, 429, timeout or network error leaves that page out of the count rather than holding it against you. The research corpus is sampled the same way as LLMSRankBot, which obeys robots.txt.

> **llms.txt v2.** llms.txt v2 (Aug 2026) accepts two forms: .md appended (page.html.md) or the extension replaced (page.md), with index.html.md or index.md for URLs without a file name. Both are correct under the spec, and rubric v2.0 counts either, trying a page’s declared markdown alternate first.

## How to fix

1. Serve a markdown version of each linked page, with .md appended (page.html.md) or the extension replaced (page.md).
2. For URLs without a file name, serve index.html.md or index.md.
3. Declare it on the page with rel="alternate" type="text/markdown" (see E8), and link the .md URLs from your llms.txt directly (see C8).

## Example

```
- [Refunds](https://acme.com/docs/refunds.md): How refunds work
```

## Frequently asked questions

### What URL should the markdown version of a page have?

llms.txt v2 accepts two forms at the same URL as the page: `.md` appended (`/docs/refunds.html.md`) or the extension replaced (`/docs/refunds.md`); for a URL without a file name, `index.html.md` or `index.md`. We try a page’s declared `rel="alternate" type="text/markdown"` link first, then the appended form, then the replaced form, and the first that works counts, so either convention scores the same. A linked URL that is itself served as `text/markdown` or `text/plain` counts as its own markdown version.

Sources: [llmstxt.org — The /llms.txt file, v2 (modified 2026-08-10)](https://llmstxt.org/) · [LLMSRank — Markdown pages and link relations](https://llmsrank.dev/learn/markdown-pages.md)

### What response counts as a working markdown version?

HTTP 200 with a content type other than `text/html`, so serve it as `text/markdown` (RFC 7763) or `text/plain`. A single-page app that returns its HTML shell for every path, `.md` included, earns nothing. A 404 or 410 on every candidate is a miss, while a 403, 429, network error or robots.txt block is left out of the count as unverified rather than held against you.

Sources: [RFC 7763 — The text/markdown media type](https://datatracker.ietf.org/doc/html/rfc7763) · [LLMSRank — Markdown versions audit](https://llmsrank.dev/docs/audits/markdown-versions.md)

### Which pages does LLMSRank sample for markdown versions?

The first three distinct same-site pages linked from your llms.txt, skipping assets such as images, PDFs, JSON, XML and `.txt` files; a link that already ends in `.md` is checked as is. The score is the share of sampled pages with a working markdown version, so the pages you list first matter most.

Sources: [LLMSRank — Markdown versions audit](https://llmsrank.dev/docs/audits/markdown-versions.md) · [llmstxt.org — The /llms.txt file, v2 (modified 2026-08-10)](https://llmstxt.org/)

### Should my llms.txt link to the markdown URL or to the HTML page?

Linking the `.md` URL directly is simplest for agents: it counts here as is, and it earns [LLM-friendly links](https://llmsrank.dev/docs/audits/llm-friendly-links.md) credit. If you prefer to link the HTML page, declare its markdown version with `<link rel="alternate" type="text/markdown">` (see [Link relations](https://llmsrank.dev/docs/audits/link-relations.md)); we find the twin through that declaration and the other audit counts the link as LLM-friendly too.

Sources: [llmstxt.org — The /llms.txt file, v2 (modified 2026-08-10)](https://llmstxt.org/) · [LLMSRank — LLM-friendly links audit](https://llmsrank.dev/docs/audits/llm-friendly-links.md)

### How many sites publish markdown versions of their pages, in LLMSRank’s research?

About a third of those we could check. In our 187-site research corpus (rubric v2.1; markdown versions sampled 2026-09-29 by LLMSRankBot, which obeys robots.txt), 122 sites gave a definite answer for at least one sampled page. 42 of the 122 (34.4%) earned full marks, with a working markdown version for every sampled page, 69 (56.6%) had none, and the mean score was 0.391. 64 sites were Unverified: their checks were all blocked, rate-limited or timed out, we couldn’t read any same-site page links in the file, or they joined the corpus after the September sample. 1 was Not applicable. Good examples include netlify.com, you.com, cloudflare.com, developers.deepl.com and docs.expo.dev.

Sources: [LLMSRank — How we score (rubric v2.0)](https://llmsrank.dev/docs/method.md) · [LLMSRank — Markdown versions audit](https://llmsrank.dev/docs/audits/markdown-versions.md)

See also: [all audits](https://llmsrank.dev/docs/audits.md) · [how scoring works](https://llmsrank.dev/docs/method.md)
