Understanding your scores

Sentiment score

Every mention Reputably finds is scored for sentiment — broadly, whether the person is speaking positively, negatively or neutrally about the subject. Scoring is done by a language model reading the actual text, not by keyword matching, so sarcasm and context are handled far better than a simple word list would manage. Sentiment lets you triage: an angry post about your category is worth reading before a neutral passing mention.

Because scoring uses a model, it is a judgment, not a measurement — occasionally a mention will be read more positively or negatively than you'd score it yourself. Treat sentiment as a fast sort, then read the ones that matter. If a model call ever fails, Reputably falls back to a neutral default rather than dropping the mention, so a bad response never blocks a sync or hides a conversation from you.

Buying-intent score

Sentiment tells you the mood; buying intent tells you the opportunity. Reputably scores how strongly a conversation looks like someone actively looking to buy — a recommendation request ("can anyone suggest a good physio in Brunswick?") scores high, while general chatter about the category scores low. High-intent conversations surface first, because those are the ones where a timely, helpful reply can win a customer. This is the difference between monitoring for mentions and monitoring for leads.

AI-visibility results

AI visibility is not one score. The dashboard combines three answer-level measures: visibility (the share of checked answers that name the brand), sentiment (how positively the answer describes it, from 0–100) and position (the average place the brand appears when the answer orders alternatives). Read them by prompt and model before relying on the aggregate.

The evidence views measure different stages and should not be blended into the same denominator. A brand can be named from the model's learned memory even when its website was not shown in a disclosed retrieval trail. A site can be surfaced by a search and then passed over. A source can be cited even when ChatGPT's partial trail did not list the search that found it. Those are different facts, not contradictory scores.

Because AI answers vary by phrasing, model, location, session and time, the honest use is direction of travel across repeated checks — not a permanent grade. See Set up AI visibility tracking for the full view-by-view explanation and what is AI visibility tracking? for the background.

Reading scores honestly

All of these scores exist to save you time, not to make decisions for you. They sort a large, messy stream of public conversation into "look at this first" order. The judgment — whether to reply, how to reply, whether a review needs a human touch — stays with you. Used that way, the scores are a genuinely powerful filter; used as gospel, any AI score will occasionally mislead.

Where to go next

• Set up lead monitoring — control what gets scored.
• Safe reply guidelines — act on high-intent conversations without spamming.
• Data sources & limitations — what's covered and what isn't.