Why Bluemoon never invents a number
AI visibility tools are easy to fake: a confident percentage with nothing behind it. Here is the rule we build every screen around, and what it costs us.
Ask three AI visibility tools for your brand's share of voice and you will get three numbers. They will disagree, sometimes by a factor of two, and none of them will say how they were made. That is the problem we set out to avoid, and it shaped the product more than any feature request.
A dash is not a zero
Zero means the engines were asked and never named you. A dash means we did not measure it: the plan did not include that engine, the market was added yesterday, or the scan failed and we said so. The product never confuses the two, and neither should your report. The difference sounds small until a board asks why visibility in Germany fell to zero when the truth is that nobody was measuring Germany.
Every figure carries its denominator
Recommendation rate reads “31% of 547 answers naming you”, never just 31%. A lift reads “over 12 questions, against the control group”. If a rate cannot carry its count, it is not shown. This is why some screens in Bluemoon look emptier than a competitor's demo: there was nothing measured to put there yet.
Illustrative. Bluemoon shows the first two; it refuses to show the third.
Claims come with evidence
Every mention links to the stored answer it came from: the question, the engine, the market, the date, the full text and the sources it cited. If we say a competitor is recommended, you can read the sentence. If we say a source is cited by 3 of 6 engines, you can open the three answers. Evidence is not a premium feature; it is what makes the number a number.
Observed and inferred are different words
Some things cannot be observed: an engine's exact retrieval index, the prompt volume behind a question, the share of purchases influenced by an answer. When we show anything derived, it is labelled as an estimate, its inputs are listed, and it never sits next to an observed rate as if it were one. Our methodology page keeps the current list of what is observed and what is inferred.
- 1Ask
A buyer question, in a market, sent to an engine at a recorded time.
- 2Store
The full answer and the sources it cites are kept, unchanged.
- 3Read
Brand mentions and roles (recommended, listed, mentioned, caveated) are detected and can be reviewed.
- 4Count
Rates are computed over the stored answers, with the count shown next to the rate.
Every step keeps the link to the previous one, so any rate can be traced back to answers.
If you have seen a number in Bluemoon that broke one of these rules, tell us. We treat it as a bug.
Related
What “observed” means: how we count a mention
Every rate in Bluemoon is built from stored answers. Here are the definitions behind the count, the limits of a daily sample, and the short list of things we infer instead of observe.
Playbooks6 min read
How to read an AI answer like an analyst
A stored answer holds more than your brand name. Five steps that take you from the question to the roles, the sources and the gap, and out the other side with one brief.