Understanding metrics

Onsomble turns the AI answers collected during a scan into a set of comparable metrics. This page explains how each metric is calculated and how to interpret it. You do not need to calculate these values yourself. The formulas are included so you can see exactly what sits behind each result.

​

Start with the measured responses

Each active prompt runs against the enabled models and regions. Each resulting AI answer is one measured response. The filters you apply determine which responses are in scope. Onsomble then calculates the metrics again using that selection. A response contributes only when it contains the data needed for that metric, so Position, Sentiment, and source metrics can have less data than Visibility.

​

The five headline metrics

​

Visibility

Visibility is the percentage of eligible responses that mention the selected brand. responses mentioning the brand / eligible responses × 100 If your brand appears in 12 of 20 responses, its Visibility is 60%.

• Direction: Higher is better.
• What it tells you: How often the brand appeared in the answers you measured.
• What it does not tell you: Whether the mention was prominent, favourable, accurate, or persuasive.

​

Share of Voice

Share of Voice is the selected brand’s share of all measured mentions of tracked brands. mentions of the selected brand / mentions of all tracked brands × 100 If your brand receives 30 of the 100 tracked-brand mentions, its Share of Voice is 30%.

• Direction: Higher is better.
• What it tells you: How much of the tracked-brand discussion belonged to the selected brand.
• What it does not tell you: Whether those mentions were positive or whether the brand appeared first.

​

Sentiment

Sentiment is the average tone of what the AI models actually said about the brand. Onsomble finds the individual statements a response makes about a brand, classifies the tone of each one, and averages them. It displays the result on a scale from 0 to 100.

• 0 is strongly negative.
• 50 is neutral.
• 100 is strongly positive.

If the statements about a brand average out slightly favourable, the Sentiment score lands a little above 50 — for example 62.

• Direction: Higher is better.
• What it tells you: Whether the statements about the brand were favourable, neutral, or unfavourable overall. Every score is backed by the quoted statements that produced it.
• What it does not tell you: The percentage of responses that were positive. A score of 62 does not mean that 62% of responses were positive.

A dash means the brand has no Sentiment score for the selection — it was not mentioned, or it was mentioned without any statements that carry a tone. That is missing evidence, not a bad score. Sentiment is an automated interpretation of language. Read the quoted statements behind a surprising score, especially when only a small number of them contributed to it.

​

Gap Score

Gap Score is the percentage of eligible responses that did not mention the selected brand. It is the inverse of Visibility. 100 - Visibility If Visibility is 60%, Gap Score is 40%.

• Direction: Lower is better.
• What it tells you: How much of the measured response set did not include the brand.
• What it does not tell you: The distance to the strongest competitor. It describes missed brand presence, not a competitor-performance gap.

​

Overall Health

Overall Health is a rounded average of Visibility, Share of Voice, and Sentiment. (Visibility + Share of Voice + Sentiment) / 3 If Visibility is 60, Share of Voice is 30, and Sentiment is 45, Overall Health is 45 after rounding. When a brand has no Sentiment score, Health averages the remaining components instead of treating the missing score as zero.

• Direction: Higher is better.
• What it tells you: A broad summary of presence, relative attention, and tone.
• What it does not tell you: Which part of performance is strong or weak. Always inspect the three component metrics before acting on a Health score.

​

Prompt-level measurements

​

Position

Position is the average order in which the selected brand appeared when a response provided a measurable position.

• 1 means the brand appeared first.
• 2 means it appeared second, and so on.
• Lower is better.

Responses where no position can be measured do not contribute to the average. A dash means there is not enough position data for the current selection.

​

Volume bars

The Volume column in Prompt Results is a visual summary, not a separate score.

• Each bar represents one selected model.
• A green bar means the response mentioned the brand.
• A taller green bar means the brand appeared earlier in that response.
• A muted or red bar means the response did not mention the brand.

Use Position for the numeric average and open the response when you need the full context.

​

Competitors

The Competitors value shows how many tracked competitors appeared in responses to the prompt. It shows how crowded the measured answer was, but it does not indicate whether those competitors were described positively or recommended.

​

Reference measurements

These are the numbers you’ll see on the References page. Here, a “source” is any website AI pointed to in its answers.

​

Reach

Reach shows how often a website turns up across the answers Onsomble collected, as both a percentage and a count. answers that used the site / all answers in view × 100 If a site showed up in 8 of 20 answers, its Reach is 40%. The count underneath can run higher than the number of answers, because a single answer can point to the same site more than once.

​

Cited

A Citation is a reference AI turned into a clickable link, rather than leaning on quietly in the background. Cited is a subset of Reach, because AI doesn’t link to every site it uses. If a site showed up ten times and six of those included a link, it has 10 in Reach and 6 Cited.

​

Pages

Pages counts how many separate pages on a site AI actually used. Reach and Cited count how many times pages showed up; Pages tells you whether AI keeps returning to one strong page or draws from many.

​

Influence

Influence measures how much a site shapes what AI says about the brands in your market, on a scale of 0 to 100. It’s based on the positive or negative things AI said about a brand, yours or a competitor’s, while pointing to that site in the same breath. Plain, factual mentions don’t count towards it. A higher score means the site is doing more to steer AI’s view of your market, whether what it’s saying is flattering or not. The score reflects the prompts you’re currently looking at, and each site is scored against the highest-scoring one on screen. Influence appears once a scan has fully analysed what each answer said. You won’t see it on older scans, or on a scan that’s still processing.

​

Claims backed

Claims backed counts the specific statements in AI’s answers that a site stood behind, where AI’s sentence drew on a passage from that page. A high number means AI is treating the site as real evidence, rather than just dropping its name in.

​

Brand coverage

Brand coverage shows how often AI actually named your brand when it cited a site about your market, rather than just linking to a related page. references to a site that also name your brand / all references to that site × 100 If a site showed up 20 times and your brand was named in 15 of those answers, its Brand coverage is 75%. This is a link, not proof that the site caused the brand mention. On the References page, sites that get cited often and never mention your brand are called silent sources. They’re the clearest gaps to close, because they’re already covering your market without you.

​

Narrative measurements

Polarity describes whether an attribute was expressed positively, negatively, neutrally, or in a mixture of ways. Relevance is Onsomble’s estimate of how important an attribute is within the measured category narrative. It appears as High, Medium, or Low. Relevance is not a count of exact phrase matches, so inspect the supporting responses before deciding how much weight to give it.

​

Regional metrics

Region Exploration uses the same Health, Share of Voice, Visibility, and Sentiment definitions, calculated for the measured responses in each region. Compare regions with care when they use different prompts or contain very different numbers of responses. A regional result describes the questions measured for that location, not the size of your business or customer base there.

​

Changes between scans

A change beside a metric compares the current result with the previous completed scan in scope. It is shown in points, not as a percentage increase. For example, a move from 40 to 46 is an increase of 6 points. It is not a 6% increase. Prompt, model, region, competitor, and filter changes can all affect the comparison. A movement tells you what changed in the measurement. It does not prove why it changed.

​

Missing and limited data

• A dash means the current selection does not contain enough data for that measurement.
• An empty filtered view means no measured results match the filter combination. It does not mean the metric is zero.
• A zero is a calculated result. For example, 0% Visibility means none of the eligible responses in scope mentioned the brand.
• A result based on only a few responses can move sharply between scans. Read the underlying prompts and responses before treating that movement as a trend.

​

What’s next