The llms.txt Experiment: What It Can and Cannot Do

What llms.txt is for

The proposed format gives site owners a simple way to point language-model systems toward important content. Its practical value depends on whether individual systems choose to use it.

What a sensible test looks like

Publish a concise file, include only canonical resources, record the deployment date and monitor crawling, citations and answer changes over a meaningful period. Avoid changing several variables at once.

What it does not solve

It does not create authority, fix weak content, clarify a confused business entity or guarantee inclusion in generated answers.

The low-risk implementation

Keep the file factual, current and aligned with the site. Do not expose confidential processes or create an enormous duplicate sitemap.

Cited’s position

We treat llms.txt as a supporting signal worth testing, not a core visibility strategy. The broader audit determines whether more material barriers are limiting discovery.

Define success before publishing

Possible outcomes include crawler requests to the file, discovery of linked resources, more consistent citation of canonical pages or no observable change. Defining these outcomes in advance prevents the test from being rewritten around whatever happens.

Separate discovery from influence

A crawler fetching llms.txt shows access, not that the file changed an answer. Evidence of influence would require a sustained difference in retrieval or citation while other material variables remain stable.

Maintain the file like an editorial index

Include a concise description of the organisation and a curated list of canonical, high-value resources. Remove outdated links and avoid claims that are not supported on the linked pages.

What would count as evidence

Useful evidence would include dated crawler requests, discovery of resources listed in the file and repeatable changes in which canonical pages are retrieved or cited. Those observations are stronger when page content, internal links and other technical signals remain stable during the test.

A fetch of the file alone does not show that an AI system used it to form an answer. Cited therefore treats llms.txt as an experimental discovery aid rather than a ranking factor or guaranteed visibility mechanism.