How We Measure AI Visibility for a Local Business

The question a local business is really asking

When an owner in Boca Raton asks “are we showing up in AI?”, they usually mean one thing: when a customer asks ChatGPT, Gemini, Perplexity or Google’s AI features who to call, does our name come up? That is a fair question and a hard one to answer honestly, because AI-generated answers vary by platform, model, query wording, time, location and personalization. Two people asking the same question on the same afternoon can get different lists.

So we do not try to answer it with a single number. We answer it with two separate measurements that are kept deliberately apart, plus a plain statement of what neither one can claim.

Measurement one: the AI Visibility Readiness Score

The readiness score looks at the things you control: your website, your local and entity signals, your service and location clarity, the evidence and authority you make available, and whether your content actually answers buyer questions. It is scored out of 100 across five categories:

1. Technical Discoverability (20 points), 9 checks, for example “https available” and “homepage is publicly fetchable”.
2. Entity + Local Clarity (20 points), 8 checks, for example “business name is clear” and “primary city/service area is visible”.
3. Service + Location Relevance (20 points), 7 checks, for example “primary service is clearly explained” and “dedicated service content exists”.
4. Evidence + Authority Readiness (20 points), 8 checks, for example “real proof elements are present” and “case studies/examples are available when appropriate”.
5. Answer-Ready Content (20 points), 8 checks, for example “key services/concepts are defined directly” and “process/how-it-works is clear”.

Every check is explained in the report, and every check lands in one of three states. Pass means we verified the signal is there. Gap means we verified it is missing. Not verified means we could not read enough to judge it, because a page timed out, a profile could not be associated, or bot protection turned us away.

A check we could not verify earns no credit toward the score. That is a deliberate choice, and it is the part of this methodology we would defend hardest. The tempting alternative is to quietly give unverified signals the benefit of the doubt, which produces a friendlier number and a worse assessment: it flatters exactly the sites we understood least. Instead we report a second figure, Assessment Coverage, which says how much of the intended assessment we could actually verify. When coverage drops below 80 percent we do not publish an overall score at all, we publish a partial assessment and say what we could not see.

The score is our internal diagnostic methodology. It is not a ranking-factor score from Google, OpenAI, Gemini, Perplexity or any other platform, and we say so on every result.

Why start here? Because the official guidance from the platforms themselves points at fundamentals. Google’s documentation on AI features says the same SEO practices that help in traditional Search remain relevant to AI Overviews and AI Mode, that there are no special requirements, that important content should be crawlable and textual, and that structured data should match visible content. OpenAI’s publisher guidance says public websites can appear in ChatGPT search and that blocking its search crawler removes that possibility. None of that is a secret; all of it is measurable on your own site today.

Measurement two: the Observed AI Recommendation Sample

The second measurement is what AI-assisted platforms actually said. When a sampling provider is enabled, we run a fixed set of commercial queries, the questions a real buyer in your category and area would ask, and record the exact query text, the platform and model where known, the timestamp, whether your brand was mentioned, which competitors were mentioned, and which sources or citations supported the answer where they are available.

We call it a sample on purpose. An API query does not exactly reproduce what every consumer sees, and a result on Tuesday is not a promise about Thursday. What a sample does give you is evidence: which businesses recur, which websites, reviews, directories and publications the answers lean on, and whether the facts used to describe your company are accurate. That is the raw material for deciding what to fix.

Why we keep the two apart

Because they answer different questions and fail in different ways. Readiness can be high while observed visibility is low, usually a sign that third-party evidence, not your own site, is the missing piece. Observed visibility can be decent while readiness is weak, usually a sign of reputation carrying a site that is easy to overtake. Blending them into one score would hide exactly the distinction that tells you what to do next.

Tracking change over time

For clients we maintain a fixed benchmark prompt set, repeat it on a schedule, record the platform and model each time, compare trends rather than single snapshots, and keep exploratory prompts separate from the benchmark. Alongside that we watch traditional search and local metrics, and Google Search Console’s generative-AI performance reporting where it is available, so AI visibility sits next to the numbers the business already trusts.

What none of this claims

• that any score or sample proves your brand is, or is not, recommended in every AI engine;
• that a higher readiness score guarantees inclusion, ranking or recommendation;
• that we have access to any platform’s ranking system;
• that a tactic is officially endorsed unless the platform’s own documentation says so.

What we can say is narrower and more useful: here is what your site and profiles make available to be retrieved and verified, here is what a fixed set of questions returned on a given date, here is what changed after the work, and here are the sources we relied on. The full method, including the official references, is on our methodology page.