[ Analysis ]
llms.txt: what it does, what it cannot promise and why I publish one
[ In short ]
llms.txt is a proposal for giving AI systems a curated index of a website's content. It can help people consulting documentation, but does not guarantee indexing or citations. Google says Search ignores it. For other systems, a lack of documented support does not establish that the file is never read.
- Published
- Updated
- Reading
- 4 min
On this page
What is llms.txt?
It is a Markdown text file that introduces a site and links to selected useful pages. The proposal distinguishes the index from complete documents: one helps readers navigate, while the others contain the information. It does not replace robots.txt, a sitemap or public pages.
Its clearest value is editorial: selecting content and describing it without ambiguity. An index might distinguish a current guide from a historical note. That helps someone using the file; it does not establish that a search engine treats it as a ranking signal.
[ Sources ]
Which support is documented?
Separate three questions: can a system retrieve the file, does it consult it for a particular task, and does it use it to select sources? These are different behaviours. A request in the server logs answers the first question, not the other two.
| System | What is documented | What this does not establish |
|---|---|---|
| Google Search | Official guidance says Search ignores llms.txt | The behaviour of other providers |
| OpenAI | Crawler documentation distinguishes search, training and user requests | A commitment about the role of llms.txt |
| Other systems and integrations | Check documentation and the particular use case | Missing documentation does not prove that nobody reads the file |
Does publishing the file mean using it as a signal?
No. A provider can publish its own documentation in Markdown to make it easier to consult. That does not establish that its search engine uses other websites' files to decide which pages to cite.
A correlation between the file and citations also needs care: sites publishing it may have better content, more links or greater recognition. Without a design that separates those factors, the result cannot be attributed to the file. This is a rule for interpreting evidence, not a rejection of the format.
Which observations help you decide?
Before measuring, state which use you are investigating. For documentation, check whether the index leads to the right pages. For crawlers, observe requests and response codes. For citations, use a separate protocol that preserves questions, conditions and answers.
A request for the file is not a citation; a citation does not establish that the file caused it. Recording the events separately prevents a technical signal from becoming a commercial promise.
Why does Creaitivo publish one?
As a public index and a consistency check. It presents the four core services, distinguishes specialist engagements and links to supporting articles. It should say the same thing as the pages, structured data and public profiles.
I do not present it as a way to increase citations. It is a low-cost way to explain the site's structure to anyone consulting it. The real cost includes maintenance: a polished index with incorrect links does not help.
[ Sources ]
How do you write a useful index?
Start with canonical pages and the question each one answers. An accurate description is worth more than a long keyword list. Someone unfamiliar with the business should be able to navigate without interpreting internal names.
- Use the same descriptions of identity, services and boundaries as the website.
- Link to accessible pages and check redirects.
- Distinguish current content, examples and historical material.
- Exclude confidential data and instructions that contradict the pages.
- Name the person who updates the index when the offer changes.
When is it not a priority?
When commercial information is incomplete, important pages are inaccessible or nobody can maintain the index. Fix those foundations before adding another format.
For a brand seeking to be understood in AI answers, the first question remains practical: does the website explain the offer, intended customers and supporting evidence clearly? llms.txt can summarise that answer, not replace it.
[ What to take away ]
- Distinguish a documentation index, crawler access and source selection.
- For Google Search, llms.txt is not a visibility lever.
- For other systems, missing documentation does not establish non-use.
- Keep file requests separate from citation observations.
- Publish an index only if you can keep it consistent with the pages.
How to measure your brand's visibility in AI answers. Read the article
[ Related services ]
[ Apply this note ]
Help customers find your brand in AI answersCheck how your brand is described and cited in answers to your customers' questions.
[ Related notes ]
[ Sources ]
How is your brand being read?
Tell me about the market, the questions that matter, and what you cannot observe today. Within 48 hours, I will indicate whether you need a baseline, a technical intervention, or a comparison over time.