The framework
The AI Visibility Score.
A single number, 0 to 100, for how likely an AI assistant is to recommend a company. We built it because “are we visible in AI” was a feeling, and a feeling cannot be improved.
Definition
The AI Visibility Score is a Rothenhall Partners framework: a 0 to 100 measure of how likely an AI assistant is to recommend a company, rolled up from five weighted dimensions and measured across ChatGPT, Claude, Perplexity, and Google’s AI surfaces.
Ranking in Google no longer decides whether you make a buyer’s shortlist, because that shortlist is increasingly built inside an AI assistant first. The AI Visibility Score measures that new surface directly: not where you rank, but whether the models can reach you, know who you are, and name you when it counts.
What it measures
Five weighted dimensions.
Each dimension answers one question a model implicitly asks before it recommends you. The weights reflect what actually decides an AI answer, access and identity first, proof last.
Machine access
Can the AI crawlers actually fetch your site?
Whether GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot and the rest are served or refused, at both the robots.txt and the CDN layer. A silent block here zeroes everything downstream.
Entity clarity
Does a model know which company you are?
Whether the assistants resolve your name to you, and not a competitor or a same-named product, backed by Organization and Person schema and one consistent descriptor across the web.
Shortlist presence
Are you named when buyers ask?
Your mention and citation rate on the questions buyers actually type, and your share of voice against named competitors on the third-party lists assistants quote.
On-page extractability
Can a model read and quote your pages?
Whether your content is server-rendered, chunk-structured, answer-first, and full of extractable claims, or hidden behind JavaScript a crawler never runs.
Authority signal
Is there real-world proof to reward?
The founders, press, partners, reviews and original data that give a model a reason to trust you and repeat you.
How to read a score
Four bands.
The number is a communication device. The band is what a founder remembers.
0 – 40
Invisible
The assistants cannot find you, cannot read you, or do not know who you are. New or blocked companies live here.
41 – 60
Faint
You exist to the models but rarely surface. Strong companies with weak machine-visibility sit here.
61 – 80
Present
You appear for some of the queries that matter, and you are read and understood correctly.
81 – 100
Recommended
You are named, cited, and characterised well across surfaces for the queries your buyers ask.
How it is measured
Distributions, not a single screenshot.
A defined query set.
We never score against a vibe. The score is computed against a named, versioned set of the prompts your buyers actually type, which you own.
At least five runs, at least two geographies.
AI answers are non-deterministic. A single run is noise. Every prompt runs five or more times, across geographies, so the number is a rate, not a lucky screenshot.
Rates, never rankings.
There is no rank in an AI answer. We report how often you appear, are cited, and are described accurately, as distributions across ChatGPT, Claude, Perplexity, and Google.
Every number sourced and caveated.
Referral data is an undercount and we say so. We do not promise placement in any individual answer, because nobody controls a probabilistic synthesis step.
Claims discipline
What the score is not.
- Not a ranking. There is no position to hold in a synthesised answer.
- Not a guarantee. We move rates, never the outcome of any single response.
- Not a one-off. A single measurement decays within days, so the score is a tracked line, not a certificate.
- Not a black box. Every sub-score traces back to the runs and checks that produced it.
Scores in the wild
The same measure, three companies.
The score is only useful if it is comparable. Here it is, applied to three real diagnostics.
Invisible
BetterWaves
A new app with almost no web footprint to be found by.
Faint
Napkin
A strong studio the AI engines were being blocked from reading.
Present
DayOne Technologies
Open and credible, but its pages were hidden behind JavaScript.
What is your AI Visibility Score?
We run the diagnostic on your public footprint and return the score, the five sub-scores, and the specific, reproducible reasons behind each one. The findings are yours to keep.