SEO myths and dead features guide
What llms.txt Does, and When It Is Worth Maintaining
llms.txt can point an agent to useful documentation. Learn what the proposal covers, what Google says about it, and how to test a concrete use.
On this page
An llms.txt file is worth considering when you can name the system that will use it and the task it should help with. “Improve AI visibility” is too broad to be a useful test.
A documentation assistant looking for setup instructions is one possible consumer. A search ranking system is a different one. Evidence about the first does not establish an effect on the second.
What the proposal describes
The llms.txt proposal describes a Markdown file that provides background, guidance, and links to useful material for agents. Its current version is v2, dated August 10, 2026. It includes guidance for documentation and general websites, with a site heading and optional summaries and grouped links. Checked September 13, 2026.
Think of the file as an additional reading guide. You can point to installation instructions, an API reference, support boundaries, or another useful starting point.
It does not make inaccurate source material accurate. A file that links to outdated instructions gives a consumer a convenient path to outdated instructions.
The proposal also discusses Markdown versions of pages. Publishing an index file and providing those page versions are separate implementation tasks; do not assume that enabling one creates the other.
What it means for Google Search
Google explicitly says Search ignores llms.txt and that creating it neither helps nor harms Search visibility or rankings. It also says maintaining the file for other systems is fine. Google’s AI optimization guide, checked September 13, 2026.
That is a specific statement about Google Search. It does not establish that every agent or documentation tool ignores the file.
Nor does a request in your access log establish a ranking benefit. A fetch shows that something requested a resource. You still need to determine what made the request and what happened afterward.
Choose a test you can observe
A useful experiment begins with a narrow question. For example: “Can the assistant our support team uses find the current authentication instructions from this index?”
Run the test before and after adding the file:
- Choose a real task and record the expected source page.
- Record the assistant, its available tools, and the prompt.
- Confirm whether the workflow can request your file.
- Check which source it used and whether the resulting instructions are correct.
- Repeat the task, recording failures as well as successes.
This can help evaluate that workflow. It cannot show what all assistants do, or establish that the change increased demand for the business.
When reviewing logs, distinguish your own tests and monitoring from other traffic. Treat a claimed crawler identity as something to verify. Count requests separately from successful tasks, and keep the testing period comparable.
What WP Visibility provides
WP Visibility has an optional llms.txt feature, off by default. When enabled, it serves a virtual text file with site information, owner supplied text, and links to recently updated published posts and pages. The generated list excludes content marked noindex by the plugin and password protected posts.
That output is a starting point to inspect. It is not a complete export of the site, and it does not create a Markdown version of every linked page.
Follow the llms.txt setup instructions, open the resulting URL, and check it as a visitor would. Confirm that the most useful material is easy to find and that no link implies a capability or policy the source page does not support.
Give the file an owner
If you keep it, include it in documentation maintenance. Update references when setup steps change, remove links to retired material, and check that custom summaries still agree with the linked pages.
Publish only information intended to be public. An index file is not an access control mechanism, and an instruction in it is not a substitute for authentication.
For a small service site with no identified consumer, improving the actual service pages may be the more useful next task. For a documentation site with an agent workflow you can test, a maintained index may earn its place. Make that decision from the use case and observed results.
If you are evaluating other machine readable additions, the structured data guide explains a similar distinction between valid markup and a supported search feature.