If llms.txt is installed but not referenced by AI, is the direction wrong?
llms.txt was installed but not referenced by AI, is the direction wrong? In most cases, it is a mistake in the object. This file addresses whether the model side should read a site description, not whether some part of your content can be excerpted into the answer. Both public measurements give the same direction: installed does not equal referenced, and referenced does not equal because of this file.
What did the two public data sets say separately
SE Ranking has a large sample size, covering 300,000 domain names, of which only 10.13% contain llms.txt, the study also concludes that there is no correlation between this file and AI citations.Another group is a small sample experiment, ezy.ai deployed the file to 83 sites and counted the crawling records, OpenAI has read it 7 times.A large sample indicates that it has not yet become the industry default configuration, while a small sample indicates that the number of times it is captured is very low. The combined data can refute the hypothesis that 'putting a file is the same as turning on the reference switch', but cannot refute that 'the model will read the text on your page'.
Where are the boundaries of what llms.txt can do
It can do three things: provide site orientation, guide to key pages, and indicate how it should be used. What it cannot do is supplement the text with extractable answers, nor can it control whether the engine wants to summarize this section. It is more practical to treat it as a 'site description page for machines', similar in role to robots and site maps in the crawling chain - it is an entry agreement, not the content itself.
Installed but not referenced, common three misjudgments
The first misjudgment is treating the file as content: The file contains the site introduction, but the main text does not have a paragraph that can stand alone as an answer. The second misjudgment is not checking the crawling entry: The robots rules block the user agent on the model side from the outside, and even if the file is well written, it cannot be read. The third misjudgment is not checking the stability of the address: After the column is revised, the old address does not do a 301 redirect, the historical pages have already changed to new content in the index, and the anchor text is still pointing to the old address. These three situations do not require a new plan, just complete the existing configuration.
Three more effective things than placing files
| Direction | Specific actions on the site side | Position in the background | How often to review |
|---|---|---|---|
| Paragraph can be answered independently | Each section gives the conclusion first and then the conditions, and the title is written as a question the user might ask | Article editor and content model | Each version update |
| Entry alignment | Robots allow retrieval class crawling, only block meaningless parameter paths; site map is split by category and kept up to date | Overall site settings and site map generation | Monthly |
| Address stability | After the static rules are determined, they will not be changed frequently. Add 301 redirect where changed, and anchor text points to the new address | Static URL and link push management | Each version update |
How do these three things land in AnQiCMS?
AnQiCMS supports automatic generation of LLMs.txt, and the GEO and SEO related backend modules are separated: automatic generation of site map, active push of new content to Baidu and Bing, static URL rules and 301 redirect management address form, robots configuration determines which crawling is allowed. That is to say, this file is just one of the switches, and the real quality of the cited content is determined by whether the text structure and entry configuration are the same logic.
Common questions
Should we keep llms.txt or not?The cost of keeping it is very low, it is a descriptive entry. The judgment standard is not to place the expectation of effects on it, and also not to let it conflict with robots rules.
Should we optimize the paragraph first or do the promotion first?The paragraph structure comes first. The promotion solves the problem of 'new content being discovered as soon as possible'. If there is no extractable answer paragraph on the page itself, it will not be cited even if it is caught.
How do I know if the model has read my page?View the access records of user agents retrieved from the server log, compare them with the crawling and collection status of the background, which is more verifiable than looking at the third-party 'reference rate' panel, as the criteria vary from one company to another.
Will changing the CMS be automatically cited by AI?No. The system changes are the configuration capabilities and structural constraints, and the premise for being cited is still that the content can be an independent answer and the address can be crawled.