Field notesFundamentalsMethodology
What an AI readiness score cannot tell you, including ours
Every proprietary score in this category has the same four limitations. Knowing them is the difference between using a score well and being misled by one.
AI Detective6 min read
We publish a score out of 100. It is useful, and it is worth being precise about what it is not, because the failure modes of proprietary scores are well established and we are not exempt from them.
It is not used by any AI platform
Our score is calculated by us, using our published methodology. OpenAI, Google, Anthropic, Perplexity, and Microsoft do not use it, have not endorsed it, and are not aware of it. It is not an official ranking factor anywhere.
This applies to every third-party score in this category and in SEO generally. Domain Authority is Moz's metric; Domain Rating is Ahrefs'. Both are useful comparative diagnostics and neither is used by Google. Any vendor implying otherwise about their own number is misrepresenting it.
It measures readiness, not observed visibility
This distinction matters more than any other, and it is where category marketing is loosest.
| What it means | What we do | |
|---|---|---|
| Readiness | Whether your site can be accessed, parsed, identified, and trusted — assessed from your site itself | This is what we measure |
| Observed visibility | Whether specific AI systems actually mention or cite you, for specific prompts, in specific configurations | We do not currently measure this, and we do not claim to |
Observed visibility is genuinely harder than it sounds. Answers vary by prompt wording, model version, location, language, personalisation, and time, and querying an API does not reproduce what a person sees in a consumer product. Tools that track it are doing real work, but a number derived from a fixed prompt sample is a measurement of that sample, not of your visibility in general.
A high score guarantees nothing
Improving readiness removes reasons for exclusion: pages that cannot be fetched, identity that cannot be resolved, claims nobody stands behind, facts that cannot be dated. Removing barriers is worth doing and it is not the same as being chosen.
Some checks are inference, and can be wrong
Several of our checks infer a signal by pattern matching — looking for an About link, a byline, a visible date. If your site expresses one of those unusually, we may report it missing when it is present.
We label these as heuristic in every report and state how each can be wrong, rather than presenting an inference as an observation. If one is wrong about your site we would like to know; those corrections are how the patterns get better.
Where our own accuracy stands
We have measured that audits complete reliably and cover the pages they intend to. We have not yet finished establishing how often individual findings are correct against a human-labelled benchmark, and until that is done we are not making an accuracy claim.
Saying so costs us something. It seems the only defensible position for a product whose proposition is trustworthy evidence, and we would rather be publicly incomplete than quietly overstated.
How to use a score like this well
- Treat it as a diagnostic that points at specific fixable things, not as a target to maximise.
- Read the evidence behind each finding rather than the number. The evidence is the product; the number is a summary.
- Compare a site against itself over time, not against a competitor's score from a different tool or methodology version.
- Ignore any movement you cannot trace to a specific change you made.