Does llms.txt Help AI Visibility? Four Reports Separate the Questions

The lesson

Test whether the file is requested, whether it helps discovery and whether visibility changes as separate outcomes.

What people reported

Lee Davies: Explicit instructions worked; ordinary discovery did not

Lee Davies describes building a path from llms.txt through a manifest to structured answers. When explicitly instructed to follow that path, the assistant could use it. Asked an ordinary customer question without those instructions, it fetched the homepage and stopped. A successful guided demonstration was therefore different from spontaneous discovery. Read the original report: Lee Davies — LinkedIn

Zoë Elizabeth Blogg / Reboot: Three months without discovery through the file

Zoë Elizabeth Blogg reports Reboot’s two-domain test: new pages were referenced only in llms.txt, with no internal or external links. Logs showed bots visiting the sites but not the file or its listed pages over three months. That finding concerns this discovery setup; it does not test every linked-file or interactive-agent scenario. Read the original report: Zoë Elizabeth Blogg / Reboot — LinkedIn

GO MO Group: A longer client visibility test found no measurable lift

GO MO Group reports a six-month experiment across clients in different industries and no measurable AI-visibility benefit from implementation. The public post gives the conclusion and a link to a longer write-up. It does not provide enough detail in the post to estimate a detectable effect or conclude that every AI platform ignores the file. Read the original report: GO MO Group — LinkedIn

Hristo Stanchev: A positive result with a different likely explanation

Hristo Stanchev reports roughly three times as many AI citations for a dynamic setup as for a static one across two sites. He explicitly says the file is not proven to be the cause: the same pipeline improved sitemaps, stale-URL handling and crawlable product content. This is a bundled freshness intervention, not a clean llms.txt experiment. Read the original report: Hristo Stanchev — LinkedIn

What the experiences have in common

The shared lesson is that existence, access, discovery and selection are different stages. A successful fetch establishes access. It does not establish that the assistant would have found the file on its own, used its links or recommended the business. Equally, no observed request during one test cannot prove that no agent will ever use the format.

The positive report sharpens the question rather than resolving it. If several useful publishing changes happen together, the team needs a second test to identify which change made a difference. Otherwise a convenient filename can receive credit for ordinary improvements elsewhere in the site.

For Google AI Overviews and AI Mode, Google says ordinary SEO requirements apply and no special additional optimization is necessary. That guidance is specific to Google Search, not every AI agent. Read the official guidance: Google Search Central documentation

What these reports cannot establish

These are small or incompletely specified observational tests, with different bots, prompts and endpoints. Server user-agent strings alone do not conclusively authenticate a crawler. The reports cannot support either a guaranteed visibility boost or a universal claim that the file is useless.

A test you can run: proposed protocol

1. State one hypothesis: discovery of a controlled page set, agent task completion, or a change in citations. Do not combine all three into one success score.
2. Record baseline logs and repeated answers for a fixed set of relevant prompts. Separate crawler requests, user-triggered fetches and monitoring tools.
3. Keep HTML, sitemaps, publishing frequency and other changes stable while adding the file. If testing freshness too, use a separately labeled intervention.
4. Track file requests and subsequent page access separately from citations and referrals. Keep explicitly guided prompts in their own test group.
5. Report a null finding with its site count, duration and observation limits. Treat positive changes as provisional until repeated with appropriate controls.

The practical takeaway

Implementing a file and proving a distribution benefit are different jobs. These reports make direct measurement more useful than either blanket enthusiasm or blanket dismissal.

Sources and research notes

Sources reviewed on September 15, 2026. Public social pages and search extracts sometimes expose inconsistent relative dates; unverified publication dates are omitted. Reported results are attributed claims, not independently audited facts. Reposts of the same underlying campaign are not counted as additional experiments.

• Lee Davies — LinkedIn — Public post text retrieved.
• Zoë Elizabeth Blogg / Reboot — LinkedIn — Public post text retrieved.
• GO MO Group — LinkedIn — Public post text retrieved.
• Hristo Stanchev — LinkedIn — Public post text retrieved.
• Google Search Central — Official documentation — Public post text retrieved.