Standards

AI Visibility

The AI Discovery Files specification, a free checker, a free WordPress plugin, a verified directory and quarterly adoption research.

Visit AI Visibility All five brands

The brand

What AI Visibility is

AI Visibility publishes the AI Discovery Files specification: a set of ten files a website can serve at its own root to state, in a form a machine can read, who runs it, what it publishes and what it permits. The specification is the brand's main output, and it is the only thing in the group whose main output is a document rather than a job.

Around the specification it runs four other things, and between them they are most of what the site actually does. A public checker that grades any domain against all ten files. A free WordPress plugin that writes them for you. A verified directory of sites that implement them. And quarterly research that crawls around two thousand of the largest global and UK domains to measure how many have bothered.

It also sells one thing, which is covered further down this page rather than left for you to find on its pricing page.

The specification

The ten AI Discovery Files

Ten files, and each does one job. A site can implement one, some or all of them.

  • llms.txt v1.7.0 A plain-text index of what a site publishes and who publishes it.
  • llm.txt redirect A compatibility alias for the singular spelling. It 301s to llms.txt rather than duplicating it, so a crawler learns the canonical name.
  • llms.html v1.7.0 The same content in a form a person can read as well.
  • ai.txt v1.7.0 What an AI system may do with the content: permissions, restrictions and how attribution is expected.
  • ai.json v1.8.0 Structured site facts, including featured articles, validated against a JSON Schema.
  • identity.json v1.8.0 Who the organisation is, in machine-readable form, aligned with Schema.org Organization.
  • brand.txt v1.7.0 How the brand should be described and named.
  • faq-ai.txt v1.8.0 Question and answer pairs for assistants to quote.
  • developer-ai.txt v1.8.0 Technical facts about how the site is built, for an AI assisting a developer.
  • robots-ai.txt v1.7.1 Stated permissions for AI crawlers, separate from robots.txt.

This site implements all ten, which is the least we can do given who publishes the specification. You can read ours at /llms.txt and /ai.json. Versions above were read off the specification index on , and a check in this repository re-reads that index and fails the build if this list drifts from it.

Free to use

The checker, the plugin and the directory

Three things, none of which costs anything or asks for an email address. They are the reason the specification can be adopted by people who will never be customers, which is the only test of an open standard that means much.

The plugin is on WordPress.org at version 2.1.2, published under the bsolveit account, which is a public link between the brand and this company that neither of us controls. WordPress.org reports 50+ active installations, read from its API on . That is a small number, and it is small because the plugin was only added on . It is quoted here rather than left out because a figure held by somebody else is worth more than a flattering one held by us, and because the directory count above depends on people finding this brand at all.

The measurement

What a crawl of 1,744 domains found

Every quarter the brand crawls the top 1,000 global and top 1,000 UK domains, deduplicated to about 1,995, and counts which of the ten files each one serves. It is the only published series we know of that measures adoption of this specification rather than asserting it.

Quarter Domains crawled With at least one file Adoption Avg readiness
Q1 2026 1,460 95 6.5% 2.2
Q2 2026 1,905 137 7.2% 2.2
Q3 2026 1,744 164 9.4% 2.2

Read off the three reports on . Adoption has gone from 6.5% to 9.4% across the three quarters, driven almost entirely by llms.txt, which grew from 93 domains to 122. The last column is scored out of 5.0, and it has not moved once. So more sites are publishing something, and not one of them is publishing more of it than a year ago.

The figure that says most about the state of things is none of those. It is that 84.2% of the domains crawled have no AI-specific crawler policy of any kind in their robots.txt. Not a restrictive one, not a permissive one. The overwhelming majority of the largest sites on the web have made no decision about AI crawlers whatsoever.

The honest caveat belongs with the numbers, and the Q3 report states it before anyone else can: only 1,744 of the 1,995 domains were crawled successfully this quarter against 1,905 the quarter before, because 251 errored. The denominator moves, so quarter-on-quarter comparisons are softer than a clean percentage makes them look. A research series that flags its own weakest quarter is worth more than one that does not, which is the whole reason it is quoted on this page.

The awkward part

AI Visibility sells the files it specifies

A specification maintained by the company that sells compliance with it is not really a specification, it is a product with a rulebook attached. Anyone evaluating one has to ask whether a rule exists because it is useful or because it is billable. That question is fair, and this brand is exposed to it, because alongside the free specification it sells the files as a job: £295 plus VAT, paid once, for the ten files researched and written for your business, usually inside two working days. That price was read off the page it links to on .

An earlier version of this page dealt with that by saying the brand sold nothing at all, which was simply untrue and is the reason this section now exists. The defensible answer is narrower. Every rule in the specification can be satisfied for nothing, by hand from the published documents, or in a few clicks by a plugin the same brand gives away under a licence it cannot revoke. A rule cannot be billable when the free tool we publish satisfies it in one click. What the fee buys is somebody else doing the work, which is the same thing every other brand in this group sells.

What does still hold from the old argument is the domain. The specification lives on its own site rather than inside 365i Web Design, because a standard published on a page that also sells implementation of it would not be credible whatever the licensing said. The practical division: this brand publishes the specification and the conformance rules, and the web brands implement it on client sites.

The disclosure that matters most is the one nobody asks for. This company owns both the brand that writes the rules and the brands that get paid to follow them. That is a real conflict, it cannot be designed away with a separate domain, and the only honest thing to do with it is print it.

The limits

What AI Discovery Files do not prove

Serving these files does not make an AI assistant cite you, and anyone claiming otherwise is selling something. What the files do is remove ambiguity about who you are and what you publish. Whether a given model reads a given file, and what weight it gives it, is not something a publisher can observe from outside.

What can be observed is narrower and more useful: Google's Lighthouse now includes an agentic browsing category which checks for a valid llms.txt at the domain root, among other things. That is a real, checkable signal that machine-readable inputs are being taken seriously by mainstream tooling. It is not proof that any assistant reads the file, and the standard's own documentation says so rather than implying more.

The adoption research above should be read the same way. It measures who is publishing these files, which is a fact. It does not measure what any AI system does with them, which is not a fact anyone outside those companies currently has.

Where to go next

Check a domain against the specification, free

Run the checker Visit AI Visibility

The checker asks for a domain and nothing else. If you run WordPress, the plugin writes all ten files for nothing, and this company would rather you used it than paid anyone to do it by hand.

Related on this site: the build standard, which applies the same distinction between what can be measured and what can only be asserted.

By Mark McNeece, Director. Published . Last reviewed .

Editorially reviewed by Mark McNeece on . Author and reviewer are the same person on this site, and we say so rather than inventing a second name.

Editorial integrity: read our editorial standards.