llms.txt for Web3 sites: put the file to a practical test

1 file, 4 checks before it earns a place

First 24 hours: decide whether one curated index fixes a real navigation problem. llms.txt is a proposed Markdown convention. It offers a short guide to selected pages, not a replacement for those pages.

The file works as a human-readable index when its links are useful. Its descriptions must match the pages. The file does not make a page authoritative by itself.

For the wider technical picture, see technical AEO: schema, llms.txt, and crawlers. The llms.txt and AI crawler setup guide covers related setup.

How do you move from a page list to a live llms.txt file?

5 checks take the file from an idea to a verifiable page. Keep the content owner and deployment owner clear from the start.

1. Name the job: Write down what the file should help a reader find. If current navigation already does that, improve the navigation instead.
2. Pick public sources: Gather current pages that explain the project, product, security, or support. Record one owner for each page.
3. Trim the list: Remove duplicates, temporary announcements, drafts, and pages that no longer match the product.
4. Write the descriptions: Group links by topic. Describe what each page contains, without claims the page cannot support.
5. Publish and verify: Put the approved Markdown file at the intended site root. Open its public URL and check every listed link.

Before deployment, ask the documentation owner to confirm the facts and names. The site administrator checks the hostname and public access. After deployment, compare the live text with the approved draft. This catches a wrong path or older version before the file enters routine documentation.

6 drafting rules for a useful llms.txt

6 drafting rules keep the file short enough to scan and specific enough to maintain. Each entry should point to a public page. That page must support its own description.

• Purpose first: Explain what the site or project does in plain language.
• Source pages: Favor durable documentation over short-lived announcements.
• Clear groups: Use topic headings that match how readers look for information.
• Accurate notes: State what a linked page covers. Do not add unsupported claims.
• Short list: Include selected pages, not every URL the site exposes.
• Named owner: Keep one person responsible for checking links and descriptions.

A compact outline can use a project heading and a one-sentence summary. Add a few topic groups. For example, a Web3 project might separate these sources:

• Protocol overview and security information.
• Governance and support resources.

These are examples, not required headings. If an important answer is scattered across posts, create or maintain a durable source page first. The citable content guide covers the related job of making individual pages easy to understand and quote.

LLMs.txt vs schema.org: which job belongs to each?

llms.txt points readers toward selected pages. schema.org vocabulary describes content or entities in a structured format. Use each for its own purpose. Check that published information matches the visible site.

Neither format repairs a weak source page. A curated file cannot substitute for accurate markup. Structured data cannot replace useful documentation. Treat the source pages as material that needs to stay current. Then assess each format against its intended task.

For implementation detail, compare the schema markup for AI search guide with our notes on schema markup. The distinction is practical. One format lists chosen resources. The other describes content using structured terms.

llms.txt evidence: inspect the file, not assumed effects

Review 3 things: the approved page list, the live file, and working links. Record why each page was included. Note who approved its description. Keep a change note when the list or wording changes.

For a site review, check the file at its public URL. Compare it with the approved draft, then open each link.

For AI answer reviews, record:

• Platform and prompt.
• Observed response and cited sources, where available.

Keep that record separate from the file change. A different answer does not show that llms.txt caused it. The AI search visibility hub covers broader visibility work.

A published llms.txt file cannot require Google or another platform to consult it. It cannot require a platform to retrieve or cite a listed page. Each platform controls those choices. There is no established universal evidence that publishing the file changes search results or AI citations.

Send these 5 inputs for an llms.txt review

5 inputs let DigitalMedia1 check whether the draft matches the live site. The review compares the proposed page list with public URLs. It flags descriptions that go beyond their sources. It also checks the published file against the approved version.

• Site URL: The exact hostname where the file should appear.
• Draft file: The current Markdown, if one exists.
• Priority pages: The sources you want readers to find first.
• Page owners: The people who can confirm product and security details.
• Deployment contact: The person who can publish and verify the file.

Send the list even if the draft is incomplete. We can mark missing source pages and unclear descriptions. We can also flag links that need a live check.

You get focused edits and a clear publishing handoff. We do not claim the file will change how a platform responds. Send: site URL, draft, priority pages, page owners, and deployment contact.

How it works

1. 1. Name the jobWrite down what the file should help readers find. Keep the task only if current navigation leaves a real gap.
2. 2. Select public sourcesCollect current pages and confirm an owner for each. Leave private, temporary, or outdated material out.
3. 3. Draft the descriptionsGroup the links and describe each page accurately. Remove duplicates and claims the source page does not support.
4. 4. Publish at the rootHave the site administrator deploy the approved Markdown file at the intended hostname and path.
5. 5. Verify the live versionOpen the public file, compare it with the approved draft, and check every listed URL.