Measuring honestly

Measurement only makes sense if the questions are the real ones: “which is the best”, “who do you recommend”, “who should I go to for”. These are choice questions, the ones where the answer has an immediate effect. Measuring on informational questions — “what is a window frame” — gives higher numbers and less use.

The brand appears or it does not. It is binary and looks poor, but it is the threshold: before it, no other metric exists. Presence must be measured on a stable set of questions, always the same, or comparison over time means nothing.

Being cited last of three is not the same as being cited first. In answers, order weighs more than in a list, because the names are few and the reading is short. It is the metric that moves first when an intervention works.

The words the system uses to describe the brand when it cites it. It is the most overlooked metric and often the most useful: it reveals which source the system is using. If the description is old, the source is old, and you know where to act.

Some numbers look rigorous and are not.

• A “visibility score” without the questions it is computed on: without the denominator, the number compares to nothing.
• Counting citations in a single session: answers vary between runs, and a sample of one is not data.
• AI traffic measured as organic traffic: it intercepts a different moment, and against volume it always looks like a failure.