Claim
The standard evidence cited to validate llms.txt as a GEO tactic fails the most basic empirical test: a joke file containing fictional information about office cats passes all four proofs at the same level as a genuine llms.txt, proving the proofs measure AI system defaults rather than any meaningful signal influence.
Mechanism
Each of the four standard proofs confuses a neutral fact about how AI systems work with evidence that llms.txt changes AI behavior:
- Crawler activity: AI crawlers fetch many files, including irrelevant ones. Fetching does not indicate reading, weighting, or trusting the content.
- Google indexing: Google indexes anything accessible. Indexing does not indicate use by LLMs.
- LLM content retrieval: when a retrieval-augmented system is given the contents of a file and asked about it, it will return that content. This is RAG behavior, not evidence of organic citation influence.
- LLM endorsement: when asked "does llms.txt help?" a model returns the consensus answer in its training data, not the result of reasoning about the file.
As Williams-Cook describes the fourth failure:
"When you ask a model whether llms.txt helps, it is not reasoning...It is returning the most common thing it has seen written on the subject."
The structural issue is that none of these proofs test the counterfactual: would AI systems cite a site more or less without the llms.txt file?
Conditions
Holds when: practitioners are evaluating GEO tactics on the basis of the four standard proofs. The critique applies regardless of niche or content type.
Fails when: a practitioner runs an actual A/B study with and without llms.txt controlling for content quality and authority signals. That study has not been published. Until it is, the effectiveness claim remains undemonstrated.
Evidence
Williams-Cook created cats.txt, a file declaring fictional office cats as official company members, and submitted it to the same four proofs:
- Crawlers fetched it.
- Google indexed it.
- LLMs returned its content when asked about it.
- ChatGPT confidently recommended cats.txt for ranking before the community revealed it was satire.
Supporting data from Ahrefs: of roughly 38,000 valid llms.txt files, 97% received zero crawler requests in May 2026. John Mueller (Google) stated that "no AI system currently uses llms.txt."
"A crawler fetching a file tells you nothing about whether the contents are read, weighted, trusted, or acted upon."
Signals
- If your evidence for a GEO tactic would also apply to a file about fictional office cats, the evidence is not measuring what you think it is.
- Ahrefs crawler-request data per llms.txt file is a better signal than anecdotes about model mentions.
- Model endorsement of a tactic that appears frequently in its training data is circular.
Counter-evidence
Some implementations pair llms.txt with higher-quality structured content, making it hard to attribute any observed citation lift to the file versus the content improvement. Llms.txt may eventually matter if AI crawlers adopt it as a standard; the current evidence says they have not.
Cross-references
- AI search visibility is an off-site corroboration problem with an on-site quality floor, not the inverse: Aleyda Solis on third-party offsite corroboration as the actual floor for AI citation.
- UGC platforms hold 4x more AI-cited domains than publishers; community signals are AI search's largest unoptimized variable: Kevin Indig's data on what sources AI actually cites at scale.