The sentiment attached to a brand wherever it is discussed, and what an AI model says about a brand once it has recalled it.
Brand reputation is the sentiment attached to a brand wherever it is discussed, and in an AI context it is what a model says about a brand once it has recalled it. It is the valence layer sitting on top of brand authority: authority decides whether a model brings you up at all, reputation decides whether that mention helps or hurts you.
The distinction matters because the two move independently. A brand can be highly embedded in a model’s parametric memory and still be recalled alongside a recall, a lawsuit or a scandal. Counting a mention tells you the brand surfaced. It does not tell you what was said, which is why number of mentions and share of voice are volume measures rather than reputation measures.
Reputation is formed from what the training corpus and retrieved pages say: reviews, news coverage, forum threads, complaints and rebuttals. Because a model compresses all of it into a single characterisation, one well-covered incident can dominate a brand’s description for as long as that coverage stays in the corpus. This is distinct from source authority, which judges whether a source is worth believing, and from E-E-A-T, which is a rater framework for assessing content quality rather than a measure of standing.
Managing it is a matter of what the corpus contains, not what the brand asserts about itself. Correcting the record where the coverage lives changes what a model can say; publishing a rebuttal that nothing cites does not. The effect on AI visibility is asymmetric, since a model that has learnt to hedge about a brand can decline to recommend it while still recalling it perfectly.
