June 25, 2026 · Abel Ko
Before you can improve how AI models talk about your brand, you have to see what they already say. Share of voice is our name for that measurement: across the questions a buyer might ask, how often does a model mention you, and how favorably, compared with the alternatives. Here is the method, described plainly and without any numbers, because the method is the part worth sharing.
Start from real questions, not keywords
Traditional search measurement starts from keywords. Answer-engine measurement starts from questions, because that is how people actually talk to a model. We build a set of prompts that mirror the decisions in a category: broad questions that ask a model to survey the field, comparison questions that pit options against each other, and specific questions about a use case, a price point, or a constraint. The prompt set is the instrument. If it does not reflect how buyers really ask, nothing measured on top of it will mean much.
Query each engine the way a person would
We run the same prompts across the major answer engines, including ChatGPT, Perplexity, Gemini, and Claude, because each has its own training data, retrieval behavior, and style. A brand can be prominent in one and absent in another, and that gap is itself a finding. We repeat prompts rather than asking once, since these systems are probabilistic and a single answer is a sample, not a verdict. Running a prompt several times tells you whether a mention is reliable or occasional.
Score presence, then framing
For each answer we ask two questions. First, presence: is the brand mentioned at all, and how prominently, whether it leads the answer or trails at the end of a long list. Second, framing: when the brand is named, how is it described, and is the description accurate, flattering, neutral, or wrong. A mention buried in a footnote is not the same as a recommendation, and a confident but incorrect description is a problem to fix, not a win to celebrate. Scoring both presence and framing keeps the measure honest.
- Presence: was the brand named, and where in the answer did it appear.
- Prominence: did it lead, sit in the middle, or trail the list.
- Framing: was it recommended, described neutrally, or criticized.
- Accuracy: did the model state true things about it.
Compare against the real field
A mention count in isolation means little. Share of voice is a share, so it only makes sense against the set of competitors a buyer would actually consider. We define that comparison set explicitly and score every brand in it with the same prompts and the same rules. That turns a vague sense of visibility into a relative position you can act on: where you lead, where a competitor owns the answer, and where the whole category is up for grabs.
Track over time
A single reading is a snapshot, and these systems change. Models get updated, retrieval indexes refresh, and the public record about a brand shifts as new pages are published. So we measure on a schedule and watch the trend. The trend is where the value is: it tells you whether a change you made to your content, your positioning, or your public presence actually moved how the models describe you, or whether the movement was just noise.
What the number is, and is not
Share of voice is a directional instrument, not a guarantee. It cannot promise that a model will name you tomorrow, and it should never be dressed up as precision it does not have. What it does is replace guessing with observation. Instead of wondering how AI engines see your brand, you can look, compare, and track, and then spend your effort where the measurement says it will matter.