Visibility Score

The Visibility Score is a 0–100 composite metric that quantifies how prominently your brand appears across AI engines for a defined set of queries. It’s the primary KPI for AI visibility: the AI equivalent of a search ranking position or organic traffic number.

What goes into the score

A visibility score typically combines three dimensions:

1. Mention rate: the percentage of tracked queries where your brand appears in the AI’s response. A brand mentioned in 60 out of 100 tracked queries has a 60% mention rate.

2. Mention positioning: where in the response your brand appears. First mentions score higher than prominent mentions, which score higher than listed mentions.

3. Sentiment: the tone of the brand mention. Positive mentions score higher than neutral; negative mentions may be weighted against the score.

In LLM Metrix the weights are published rather than proprietary:

Engine score = 100 × (
  (Mention rate      × 0.50) +
  (Prominence factor × 0.35) +
  (Sentiment score   × 0.15)
)

Three inputs and nothing else. Citations, competitor mentions and the claims an answer makes are recorded on the same answer and drive their own pages, but none of them carries any weight here. Nor does your position relative to a competitor: a headline number that fell because someone else rose would be unreadable as a measure of your own work, which is why share of voice is reported beside the score rather than inside it.

The denominators differ, and that is the part worth internalising. Mention rate is taken over every answer the engine returned; prominence and sentiment are averaged only over the answers that mentioned you, because an answer that never names you has no position and no tone to measure. So a brand named once in ten answers (first and positively) can score the same as a brand named in five of ten, around third place and neutrally. Same number, opposite problems, opposite fixes. Read the mention rate next to the score, never the score alone.

Two guards fall out of the same reasoning. An answer whose engine call failed is excluded from all of it, because a timed-out request is missing data rather than an absence of your brand, and counting it as “not mentioned” would let a flaky provider manufacture a drop you would then go and try to fix. And when an engine mentioned you in nothing at all, the sentiment term is forced to zero instead of defaulting to neutral. Otherwise a brand no engine has ever heard of collects points for the tone of mentions that do not exist.

That formula produces one score per engine. The headline figure is their mean, weighted by how many answers each engine actually delivered: an engine that limped in with one answer out of twenty counts a twentieth as much as one that answered every prompt, rather than carrying equal authority and dragging the headline number by tens of points during a partial outage. An engine whose every call failed is dropped from the average entirely rather than averaged in as a zero, since a row of zeroes is indistinguishable from a brand nobody mentions.

How to read the score

Tracking score over time

The visibility score is most useful as a trend metric. A single score tells you your current standing; the score charted across scans reveals whether your AEO/GEO efforts are working and in which direction your visibility is moving. Automated refreshes run daily on every paid plan (weekly on free), so the series is daily on paid plans unless you add on-demand scans of your own.

Each score is frozen onto its scan at write time and never recomputed, which is what makes the trend line comparable at all: a figure that recalculated under a changed weighting would silently rewrite your own history.

Where the aggregate hides variation

A brand-level score averages away differences that are worth looking at directly. Two cuts are available and one is not, and the difference matters when you plan a review:

• By engine: available, and worth opening first. You may hold a strong position on Perplexity and be near-absent on ChatGPT. The per-engine breakdown is a point-in-time read of your latest scan: the frozen figures behind the trend chart are brand-level, so comparing one engine across two dates means opening both scans.
• By competitor: available, but it is a scan rather than a slice. Scoring a competitor means asking the engines about them over your prompt set, which is a separate scan that bills like one, and the benchmark table holds each subject’s latest result rather than a series.
• By query cluster: not a product view. Clustering exists in prompt discovery, where suggested prompts are grouped by topic, but no metric is joined back to a cluster. Cluster-level reading is something you do by eye over the per-prompt rows for a scan, or by pulling GET /api/v1/scans and grouping the answers yourself.

So an overall score of 55 says “you’re competitive” and the per-engine row underneath it is what says where. Reaching a cluster-level answer is real work, not a filter, which is a reason to keep the tracked prompt set tight and thematically deliberate rather than exhaustive.