How it works
Four steps, and the ability to walk back from any conclusion to the original answer that produced it.
-
You say what matters
Register the brand, the products, the categories and the competitors. Then write the questions your customers actually ask, or start from the suggestions generated out of your catalog.
A badly framed question measures the wrong thing precisely. That is why a question is a versioned object: when you change it, a new series begins instead of contaminating the previous one.
-
GeoTopX asks and preserves
Each question runs repeatedly across several generative engines over the period. Every answer is stored in full, with engine, observed version and timestamp.
Preserving is the part almost nobody does. Models change without notice and the same question answers differently tomorrow: what was not stored today cannot be checked later.
-
Evidence becomes measurement
Answers are normalized and turned into metrics with the sample size, the run-to-run variation and the confidence level stated next to the result.
A number without a stated sample is not a measurement, it is an opinion wearing the appearance of precision. Every GeoTopX metric carries the interval it varies within.
-
Measurement becomes decision
You receive recommendations that point at the evidence that prompted them, and you can walk the path backwards: from the suggested action to the observed answer that produced it.
That path back is what separates a verifiable recommendation from a well-presented guess.
The evidence chain
Every link stays reviewable once created. No step discards what the previous one produced.
- Observed answer
- Preserved artifact
- Normalized observation
- Metric with uncertainty
- Recommendation
- Report