| llms.txt is a context map, not a ranking switch |
No. Treat llms.txt as a public, LLM-friendly context map that can help agents find important resources when they choose to read it. It is not a ranking, citation, or traffic guarantee. |
Validate structure and publish useful public links, then measure real clicks, referrals, or tool activations separately. |
A valid file is not proof of ranking, AI citations, or human traffic. |
| Root /llms.txt should be public and fetchable |
The practical first check is whether the public root /llms.txt file can be fetched, returns a useful text response, and is not hidden behind auth, redirects, previews, or staging URLs. |
Run the live validator for the domain after upload and keep the report with the launch record. |
A local draft is not proof that the live root file is reachable. |
| H1, summary, and interpretive notes |
The llms.txt proposal defines a Markdown file with an H1 for the project or site and encourages concise background information such as a short blockquote summary and notes that help interpret the listed files. |
Check for one clear H1, concise summary, and short notes before file-list sections. |
A heading alone does not make the file useful if links and notes are noisy. |
| H2 file lists with Markdown links |
Useful sections should use H2 headings and Markdown list items with required links, optionally followed by short notes. This makes the file compact for both humans and agents. |
Validate that important resources are listed as Markdown links with descriptive labels and optional short notes. |
Bare URLs or ambiguous labels are weaker than curated links with context. |
| Optional section controls context budget |
The Optional section has special meaning in the proposal: those URLs can be skipped when a shorter context is needed. Validators should not treat every link as equally important. |
Put secondary, deep, or nice-to-have resources under Optional rather than bloating Core sections. |
A long list is not better if it hides the most important links. |
| Curated context, not a sitemap dump |
No. The proposal distinguishes llms.txt from sitemap.xml because a sitemap can be too large and not focused on LLM-readable context. A validator should reward concise curation over complete URL dumps. |
Keep core public URLs, docs, guides, policies, tools, and answer pages that explain the site; move lower-priority links to Optional or omit them. |
More URLs do not prove better AI usefulness. |
| Public URLs only, no private paths |
Because llms.txt is public by design, it should not list admin, cart, checkout, account, customer, token, preview, staging, localhost, or filtered private URLs. |
Remove private/noisy URLs and run a vertical checker for Shopify or ecommerce drafts. |
A link that opens for the owner may still be unsafe or useless for public agents. |
| Validate both draft and live file |
Do both. Paste the draft before publishing, then fetch the live /llms.txt after upload. Keep a copyable report so changes can be reviewed in launch notes or pull requests. |
Run paste validation, copy the report, upload the file, run live validation, and keep both proof links. |
A pasted draft is not proof of the live file, and a live fetch is not proof of rankings. |
| Pre-AI answer pack and real measurement |
Use the answer pack or proof lookup endpoint to fetch one citation-ready answer with proof links before scanning broad guides. Then measure Search Console clicks, qualified referrals, and validator activations only. |
Cite the evidence matrix, publish pack, live validator, or answer pack; do not count fake searches, self-clicks, or crawler hits. |
Crawler hits, impressions without clicks, or self-tests are not human traffic. |