llms.txt

This skill implements /llms.txt — a proposed plain-Markdown file that points LLMs to a site's most important content — and, just as importantly, sets honest expectations about it. As of now, adoption is early and major answer engines largely do not parse it, so it is a low-effort, low-impact best practice rather than a citation lever. The work that actually moves AI visibility is clean rendering, structure, schema, and authority — do those first. It works for any business, B2B or B2C.

When to use this skill

• A user asks to add llms.txt / llms-full.txt.
• Deciding whether llms.txt is worth the effort (it is cheap, so usually a mild yes).
• Setting realistic expectations about what it does today.

The honest status

• Engines largely ignore it today. Independent testing shows the major crawlers usually fetch HTML directly and do not rely on llms.txt; sites adding it generally see no measurable citation change.
• Adoption is early. A small minority of sites have it; there is no formal standard that engines have committed to honoring.
• Why do it anyway: it is a ~30-minute task, it is a clear opt-in signal, and it is useful for documentation-heavy sites and for stating brand/entity facts plainly. Treat it as cheap insurance, not a growth tactic.
• Do not let it crowd out the real work — rendering, structure, schema, freshness, and authority are where the gains are.

What it is

• A Markdown file at the site root: /llms.txt.
• A short brand/site description, then curated links to the most important pages (docs, key products/services, about), each with a one-line note.
• Optionally /llms-full.txt with fuller content for sites that want it.

Format

# Brand Name

> One to three sentence description of what the site/company is and does.

## Key pages
- [Product](https://example.com/product) - what it is
- [Docs](https://example.com/docs) - how to use it
- [About](https://example.com/about) - company facts and contact

## Contact
- Email: hello@example.com

Serve it as UTF-8 plain text/Markdown at the root.

Workflow

1. Set expectations. Tell the user up front this is low-impact today and prioritize the higher-leverage skills.
2. Draft the file. Brand description plus a curated set of the most important links with one-line notes; keep it accurate and consistent with the entity facts.
3. Publish at the root as /llms.txt (and /llms-full.txt if useful).
4. Keep it current alongside the entity/foundation facts; a stale file is worse than none.
5. Re-evaluate periodically. If engines begin honoring it more, revisit its priority.

Output format

• The llms.txt content, ready to publish.
• A one-line expectation note for the user (low impact today).
• A pointer to the higher-leverage skills to do first.

Common pitfalls

• Expecting a citation boost — there is little evidence of one today.
• Prioritizing it over rendering, schema, and authority work.
• Letting it drift out of sync with the brand's real facts and key pages.

Tool

Generate a valid llms.txt from a small config:

python scripts/generate_llms_txt.py --sample            # example config
python scripts/generate_llms_txt.py --input site.json   # write the file

Optional. The script only automates this step; it needs python3 (standard library only, no install). No Python? Just follow the steps above manually.

Related Skills

• ai-retrievability — server-rendered, readable content (matters far more).
• structured-data — machine-readable signals engines actually use.
• entity-optimization — keep the brand facts in llms.txt consistent with the entity.