LTLLMTXT.co
Markdown & Structure

Structuring Your llms.txt with Markdown

The choice of Markdown for llms.txtis not incidental. Markdown's few structural elements — headings, blockquotes, lists, links — map almost perfectly onto what the file needs to express. Used well, that structure is what turns a text file into something a language model can read as an outline.

https://
#>-

Headings carry the hierarchy

In llms.txt, headings are not decoration — they are the structure. The single H1 (#) names your site and should appear once, on the first line. H2 headings (##) mark the sections that group your links — Services, Docs, About, and so on. This gives a model a skeleton to hang meaning on: it can tell where one topic ends and another begins without guessing. Keep the hierarchy flat and predictable; a wall of same-level text is far harder to interpret than a clearly sectioned outline.

The blockquote is your summary slot

The Markdown blockquote (>), placed right after the H1, is the conventional home for your one- or two-sentence description. Setting it off as a blockquote signals “this is the summary” rather than burying it in a paragraph. It is the highest-value real estate in the file, so make it a precise, plain-language statement of what you do and for whom.

# Riverside Veterinary

> Full-service animal hospital and 24/7 emergency
> care for pets in the Sacramento area.

Link lists do the heavy lifting

Under each H2, a simple bulleted list of Markdown links is the workhorse of the format. The pattern is always the same: a link, a colon, and a short description.

## Services
- [Wellness exams](https://example.com/wellness): Annual checkups and vaccines.
- [Surgery](https://example.com/surgery): Routine and advanced procedures.
- [Emergency care](https://example.com/emergency): 24/7 urgent treatment.

The description after the colon is not filler — it is the part a bare URL cannot express. The bullet structure keeps each entry discrete, which helps a model treat them as separate items rather than run-on text. Use absolute URLs so each link stands on its own.

Formatting to avoid

Heavy inline styling

Bold, italics, and emphasis scattered through the file add visual noise without adding structure a model can use. Reserve any emphasis for genuine need, which is rare here.

Tables and images

Tables complicate parsing and images carry nothing useful in a text curation file. Leave both out; they belong on your actual pages, not in llms.txt.

Deep nesting

Multiple levels of nested lists or H3/H4 sub-sections make the outline harder to follow. A flat set of H2 sections with simple bulleted links is the easiest shape to parse.

Structure helps, but promises nothing

Clean Markdown structure earns its keep by making your file unambiguous — a model that reads it can reconstruct your site's shape without guesswork. That is a real benefit. It is also the limit of what structure can do. The llms.txt proposal is young, support varies across platforms, and there is no guarantee any given AI system reads your file at all. Good structure maximizes clarity for whatever does look; it does not compel anything to look.

Frequently Asked Questions

Why does llms.txt use Markdown instead of JSON or XML?

Markdown is lightweight, human-readable, and interpreted natively by language models without a parsing library. Its headings and lists map cleanly onto the structure llms.txt needs — a title, a summary, and grouped links — while staying easy for a person to write and edit.

Do heading levels matter?

Yes, they carry hierarchy. A single H1 names the site, H2s define sections, and the link lists sit under them. Using headings consistently is what lets a model infer structure rather than reading an undifferentiated wall of text.

Can I use nested lists or sub-sections?

You can, but keep nesting shallow. A flat structure of H2 sections with simple bulleted links is easiest to parse. Deep nesting adds complexity without adding much clarity for this use case.

Should I include code blocks or images?

Generally no. llms.txt is a curation file, not a content page. Images add nothing a text model can use here, and code blocks are rarely relevant. Stick to headings, a blockquote, and link lists.

Does clean Markdown guarantee my file gets used?

No. Good structure makes your file easy to parse, which helps, but adoption is still emerging and no AI system is obligated to read it. Clean Markdown improves your odds; it does not guarantee an outcome.

Get well-structured Markdown automatically

Our generator outputs clean headings, a blockquote summary, and described link lists — the structure done for you.

https://