Strategy
How to measure: repeated questions, not rankings
There is no ranking to check. Visibility inside generated answers is measured by asking systematically and recording what comes back.

Why rank trackers do not help
A rank is a stable number: today you are third and tomorrow fourth. A generated answer has no ranks. It has presence or absence, and it varies between one query and the next even with identical wording, because generation carries randomness and because the retrieved set of sources changes.
The consequence is that a single query says nothing. What says something is frequency: out of twenty times the same question is asked, in how many do you appear.
The question panel
Start by writing fifteen to twenty five questions a real customer would ask in their own words. Not keywords, whole questions. Split them into three groups.
- Discovery: the customer does not know the brand and is looking for a category. "Where can I eat good Mallorcan food in Palma?"
- Comparison: the customer is weighing options. "What is the difference between structural carpentry and a prefab installer?"
- Brand: the customer already knows you and wants confirmation. "What is known about Talleres Sorell?"
The third group exposes problems fastest. If asking your own name returns wrong facts or mixes your brand with another, the entity layer is broken and there is no point investing in the other two until it is fixed.
The three figures
Appearance rate
Out of every hundred panel queries, in how many the brand appears. This is the headline figure and the one to track month by month.
Accuracy of the descriptor
When you do appear, what the system says about you. Record the literal sentence. A wrong descriptor repeated is worse than not appearing, because it propagates.
Sources cited
Which pages the system names while answering. That list tells you where to work, because those are the sources it treats as authorities in your field.
The cadence
Once a month, same day, same panel, across at least three different systems, in a clean session with no history. Record it in a simple sheet with one column per month. The three month curve says more than any single measurement, and it is the only way to know whether the work is landing.
Asking from a session with your own history contaminates the result: the system may remember earlier conversations about your brand. Measurement is always done in a fresh session, signed out where possible.
Keep reading
More in Strategy

Backlink and frontlink: what actually changes
Both are mentions on somebody else’s site. The difference lies in who processes them, what they take from them, and how long they last.

Entity, source and consensus: the three layers
A mention only works if the model knows who it is talking about, can read where it reads it, and finds it repeated somewhere else. All three, not two.

How to write a quotable paragraph
Writing to be quoted is different from writing to persuade. A quotable paragraph stands on its own, carries facts, and does not ask the reader to take anything on trust.