How the score is built

Every brand gets one score out of 100. It combines what actually happens in AI answers (Outcome, weighted 60%) with how well the brand's site is built to be found (Readiness, weighted 40%). Outcome is the result that matters; Readiness is the fixable groundwork that moves it.

We put a standardised set of buyer questions to the major AI providers and measure, across many samples: how often your brand is mentioned, how often it's actively recommended, and how often your own site is cited. We report each as a rate with a 95% Wilson confidence interval, because AI answers are probabilistic — a single screenshot proves nothing.

We also decompose how you surface: whether the model knew you from training (parametric) or found you by browsing (retrieved). That distinction determines whether a problem is fixable in weeks or months.

Each category has a fixed, versioned set of questions — UK cigar retail is uk-cigar-retail.v2, spanning five intents: discovery, comparison, reputation, product and service. The version is recorded with every reading, so any result can be reproduced and audited. The set is deliberately identical for every brand in a vertical: a leaderboard only means something if everyone sat the same exam.

How the set is built, honestly. The questions are authored from category expertise. They are notsampled from real AI query logs — because no provider publishes them. There is no Search Console for AI: nobody, ourselves included, has ground truth on what people actually type into an assistant. Anyone claiming otherwise is estimating too. We would rather show our hand than imply a precision we don't have.

Known limits of v1.It covers online, national intent (“the best online cigar shop in the UK”). It does not yet measure local intent (“near me”, “open now”), so a brand that is strong in local discovery is under-represented in this reading. The next version will derive questions from measured search demand and from brands' own customer questions, and publish the source of each one.

Share of voice is a fraction: your mentions over the whole field's mentions. So the figure only means something if the field is defined — otherwise “share” is a share of an arbitrary list. We define it, and we publish the rule.

The field is measured, not guessed.We read what the assistants actually name in a category's answers and extract every business mentioned — hundreds of them — then include the ones that meet a stated criterion. For UK cigar retail that criterion is: specialist premium-cigar merchants a buyer of premium cigars could choose. Not budget or general tobacco and vape shops, not distributors or importers, not supermarkets or department stores, and not cigar marques — a manufacturer is a product, not a shop, and a distributor sells to no consumer. Including them would dilute every share figure toward meaning nothing.

The honest limit. A field is never finished — assistants name new retailers as the web changes, and we add the ones that qualify. So a share of voice is a share of the defined field as it stands, not a claim about every shop that exists. We publish the field and its rule so a disagreement is about a stated criterion, not a hidden list.

We crawl your site for the signals AI systems rely on across seven pillars — crawlability, machine readability, structured data, entity clarity, content coverage, freshness, and agent readiness. Each contributes points toward the Readiness score, and each failing check maps to a concrete fix.

The AVI is a repeatable audit, not a guarantee of AI recommendations. We measure via provider APIs, which can differ from the consumer apps; our prompt set is a benchmark, not the full universe of real queries. We state these limits rather than hide them. Weights are provisional and will be validated against observed outcomes over time.

The index is published under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). Quote a score, reproduce a category table, chart the numbers, use them in commercial work — you need no permission from us. The one condition is attribution: credit The AI Visibility Index and link to the page you took the figures from.

That is deliberate. A standard nobody may quote is not a standard, and a ranking is worth more to the brands in it when the press can reproduce it. If you are writing something up and want the underlying figures or a comment on method, ask — theaivisibilityindex@outlook.com.