How to Measure Your Brand’s AI Visibility

Most teams "measure" AI visibility by typing their name into ChatGPT once and reacting to whatever comes back. That is not measurement — model answers are probabilistic, vary by phrasing, and differ across engines. The same question asked five times can produce five different competitor lists. Building Sourceable forced us to treat this rigorously; here is the framework, whether or not you use our tooling.

The goal is a number you can track monthly and defend to a board — not a screenshot.

The four metrics that matter

Mention rate: of the prompts a real buyer would ask in your category, in what share do you appear at all? Accuracy: when mentioned, are the facts right — services, location, pricing — or is the model improvising? Sentiment and framing: are you "a leading option" or "a smaller alternative to X"? Citation share: when engines cite sources, are any of them yours? Track all four; they fail independently and have different fixes.

“A screenshot of one good answer is anecdote. Fifty frozen prompts, four engines, sampled monthly — that is data.”

Build a prompt panel, not a vanity check

Write 30–50 prompts stratified across intent: category discovery ("best X for Y"), comparisons ("X vs alternatives"), problem-first questions your buyers actually ask, and named-brand checks. Freeze the panel. Run every prompt against each engine you care about — sampled multiple times, because variance is the point — on a monthly cadence. Now movement means something: it is the same questions, asked the same way, drifting over time.

Reading the numbers without fooling yourself

Expect brutal baselines: most mid-size B2B brands start near zero on category prompts — that is the market reality, not a broken methodology. Watch for the traps: named-brand prompts inflate scores (of course it knows you when you say your name); a single engine is not the market (we routinely see a brand strong in Perplexity and absent from Gemini, because they weight corroboration differently); and month-one deltas are noise — trends need three points.

Manual panels are workable at small scale — a spreadsheet and discipline. Continuous coverage at scale is what we built Sourceable for. Either way: measure first. Every AEO tactic in this series is a hypothesis until a panel says it moved something.