Short answer

An observation is one specific question, asked on one specific platform, at a specific date and time, with its full answer, the brands it mentions in order, the sources it cites and a status (valid, error or unavailable). It is the unit everything else is built from: rates, comparisons, variability. Without observations there is no measurement; there are opinions with decimals.

Anatomy of an observation

  • Question: the literal text and its version. “Which online accountant would you recommend if I am a freelancer in Madrid?”, version 1.
  • Platform: ChatGPT with search turned on, in a clean session.
  • Date and time: 11 September 2026, 10:40.
  • Language and country: English, Spain.
  • Full answer: the whole text, not a summary.
  • Brands mentioned, in order: A, B, C.
  • Explicit recommendation: yes or no, for each brand.
  • Sources: the domains and URLs cited.
  • Verifiable claims about the brand: and their status (correct, partial, incorrect, not verifiable).
  • Status: valid.
  • Reviewer: who transcribed and classified it.

Why the unit is not the position

Because many answers have no list, and without a list there is no position. And because a position without its question, its platform and its date cannot be repeated. An observation can.

Why the unit is not the score

Because a score is a summary of observations, and without them it can be neither audited nor compared. If someone gives you a score, ask for the observations behind it.

How many you need

Our AI diagnosis uses 24 (12 questions on 2 platforms). The ongoing partnership uses 40 a month (20 questions on 2 platforms, twice). It is not a magic number: it is the one that makes the work, the cost and the gaps visible without pretending to represent the whole market.