AI Visibility Report Metrics Explained

An AI visibility report is useful only when its metrics can be traced back to real prompts and answers. Start with the evidence, not the headline number.

Mention rate

Mention rate is the share of valid monitored answers that contain the brand. It answers “Did the model include us?” but not “Did the model prefer us?”

A mention can be neutral, negative, or incidental. Open the response to understand the context.

Recommendation rate

A recommendation requires stronger language: the assistant presents the brand as a suitable option for the user's need. A brand may have a high mention rate but a low recommendation rate when its category is understood but its advantages are not persuasive.

Position and list rank

Earlier placement can matter in list-style answers, but prose answers do not always have a meaningful rank. Use position as supporting context rather than an absolute search ranking.

Competitor presence

Competitor counts reveal which alternatives occupy the same answers. Inspect the reasons they are recommended: stronger proof, clearer use cases, broader recognition, or citations from trusted sources. A discovered brand is not automatically a direct competitor, so confirm it before adding it to the project.

Citations

When a model returns sources, Dottly AI records the available URLs and domains. Citations show which pages help shape the answer. No recorded citation can also mean that the provider or model did not expose source data; it should not always be read as “the model used no sources.”

Trends

Compare equivalent runs over time. A small sample can move sharply after one answer changes, so consider the denominator and read the changed responses. Prompt edits, model changes, and temporary provider behavior can all affect a trend.

The strongest workflow is simple: find a repeated gap, inspect the supporting answers and sources, make a verifiable improvement, then observe future runs under the same conditions.

Investigate the next layer

The Dottly AI report documentation explains how to inspect prompt-level evidence. Learn how to separate signal from AI visibility fluctuations, turn repeated brand displacement into a GEO competitor analysis, and investigate AI search citation gaps.

Teams comparing monitoring platforms can use the AI search visibility tools for SaaS evaluation guide to test whether aggregate metrics remain traceable to prompt-level evidence.

For a wider channel mix, use the brand tracking software selection framework to assign each metric to the system that actually observes it.

When these measurements need to reach clients or executives, the SEO report PDF framework shows how to turn denominators, evidence, uncertainty, and next actions into an auditable document.

Check the denominator before the percentage

Every rate should show its numerator, denominator, and eligibility rule. A compact metric ledger can look like this:

Do not compare two percentages when their eligibility rules differ. For example, a route that does not expose citation data should not be treated as a route with zero citations. Keep unavailable, not found, and not evaluated as different states in exports and dashboards.

Use a metric review workflow

Start at the response level, then move upward:

1. Confirm the prompt version, model, market, language, and completion state.
2. Read the answer and verify the brand and competitor classifications.
3. Check citation URLs and whether they were available on that route.
4. Recalculate the aggregate from the visible eligible rows.
5. Compare only equivalent runs and annotate any configuration change.

If the aggregate and the rows disagree, preserve the raw response and fix the calculation before publishing a trend. The report documentation shows where to inspect prompt-level evidence and failure states.

Frequently asked questions

Is mention rate the same as recommendation rate?

No. Mention rate records whether the brand appears. Recommendation rate asks whether the answer presents it as a suitable choice. A neutral or critical mention can increase the first metric without increasing the second.

Does position represent a conventional search ranking?

No. It describes where the brand appears in a list or answer when that position is meaningful. Prose answers may not have a stable list rank.

How should failed model calls affect a trend?

Keep failures visible as operational outcomes, but exclude them from rates that describe valid answers. A higher failure count is a reliability problem, not evidence that the brand was absent.