Being Mentioned Is Not the Same as Being Chosen

Almost every organisation now measuring its presence in AI answers is measuring one thing: whether the brand came up. It is the easiest signal to collect, the easiest to chart, and the least connected to any decision a customer actually makes.

Mention counts were borrowed from an older discipline. In paid and earned media, an impression was the unit because exposure was genuinely what was being bought. An answer is not an impression. An answer is an act of selection carried out on the reader’s behalf, and the interesting question is not whether a brand survived into the text. It is which of four quite different states it reached.

Four states, not one

Presence. The brand appears somewhere in the answer. This is the state most tools report, and on its own it establishes only that the model holds the name.

Consideration. The brand appears inside the set the answer treats as relevant to the question asked. A brand named in a list of things that exist has not been considered. A brand named among the options for this patient, this indication, this market, has.

Recommendation. The answer moves from describing to directing. Language shifts — commonly used, first line, the preferred option where cost is a factor. A brand can be present and considered in every answer and never once reach this state, and a competitor with lower presence can occupy it consistently.

Decision survival. The recommendation holds when the question gets harder. Add the payer, the contraindication, the price, the prior therapy, the local guideline, and watch what happens. Many recommendations do not survive the second question, and the second question is the one a real buyer asks.

These are not degrees of the same measurement. They are different measurements. A single number that averages them produces a chart that goes up while the commercial position goes down.

Why the distinction is commercial, not semantic

Consider two brands in the same category. The first is named more often. The second is named less often but is consistently the one the answer reaches for when the question narrows to who should actually receive this. On a share-of-voice view the first brand is winning. In the only place the answer touches revenue, it is not.

The failure is not that the first brand is invisible. It is that the metric it is being managed by cannot see the difference. And unlike a media plan, there is no second impression to buy: a buyer who asks a question receives one answer, and the brands that did not reach recommendation are not underweighted in it. They are absent from the part that counts.

This is also why sentiment, taken alone, misleads. A brand can be described warmly throughout an answer and still be the one the reader is steered away from at the decision point. Tone is not direction.

What this changes for the people who own the number

For Commercial and Brand, the practical consequence is that share of voice should stop being the headline. The number worth putting in front of leadership is the rate at which the brand reaches recommendation on the questions that precede a real decision, and how that rate moves.

For Medical and Regulatory, the four states matter for a different reason. A claim that appears at the presence layer is a description. The same claim appearing at the recommendation layer is functioning as guidance to a reader who may act on it, and it should be held to the standard that implies — checked against the approved record rather than counted.

For Market Access, decision survival is the whole question. An answer that recommends a therapy and then collapses the moment coverage, prior authorisation or price enters the conversation is not a win that needs defending. It is a gap that needs closing.

What to watch next

Two things are worth establishing before the next set of evidence enters your category. First, a baseline: where the brand sits across all four states today, on the questions your buyers actually ask rather than the ones that flatter the result. Second, the competitor that occupies recommendation language most often — not the one mentioned most, which is usually a different company.

Both are measurable now. Neither is visible in a mention count.

How each of these is measured, and where the measurement stops, is set out on our methodology page. To see the four states on your own portfolio, request a briefing.