Skip to content
GeoAI Optimization

Methodology

Measurement principlesNumbers that explain themselves

AI-search measurement is young and easy to oversell. These six principles apply to every figure in our research, client reports and the future GEO Measurement Platform.

The principles

Principle

No Black-Box GEO Scores

We will not publish a number simply because a number looks impressive.

Every measurement must identify its methodology, sampling conditions, date, system context, and limitations. Until a scoring method is published with its dimensions, weighting, sampling and uncertainty, there is no GEO score.

  1. 01

    Every metric reports its method

    Query set, systems, sampling and definitions travel with the number.

  2. 02

    Every result has a date and sample

    When it was observed, and how many runs it is based on.

  3. 03

    AI-system conditions are documented

    System, surface, locale, language and session state.

  4. 04

    Mention, citation and recommendation are distinct

    They move independently, so they are never blended.

  5. 05

    Measurements are observations, not guarantees

    A sample describes what was seen, not what will happen.

  6. 06

    Results state their limitations

    What was not covered, and how much results varied.

Before any GEO score is published

A composite score will only be published once its methodology is public, including:

  • dimensions and their definitions
  • weighting and the reasoning for it
  • sampling, AI systems and query methodology
  • locale and dates
  • limitations, confidence and uncertainty
  • a version number, so scores from different versions are never compared

The full method

Query selection, sampling, systems, conditions and metric definitions are documented on the methodology page.