When we built the first version of Pondral, we knew the hardest thing we'd have to defend was our scoring methodology. Brand-monitoring tools are often accused, usually fairly, of running a black box.
So we made a rule: every score Pondral produces must be derivable from a public rubric, run on a reproducible prompt, against a known model version. No proprietary "secret sauce." Just five factors, weighted, averaged across every query and engine in the audit.
This post walks through each factor, why it's weighted the way it is, and what edge cases we've learned to handle. For the full design rationale, see our deep-dive on how the scoring rubric was built.
The five factors are: Presence (20%), Prominence (25%), Context (20%), Citation Link (20%), and Competitive Presence (15%). For definitions of each of these terms, see the AEO glossary. Presence is the simplest: does your brand appear at all? Prominence is harder. We measure position in the response, whether you're mentioned by name vs. inferred, and whether the model leads with you or buries you.
Context is where a mention can still go wrong. A model can mention you accurately or it can mention you in the wrong category or with the wrong description. Our Context grader checks portrayal, factual accuracy, and category fit, and combines them into a single Context grade, so it does not report each failure mode separately. To understand how AI visibility measurement handles answer variability, read our post on inter-rater reliability for AI answers.