# Reading the Score



## Visibility Score [#visibility-score]

**How often you appear in the answer across your tracked prompts.** Compare it
to your own previous number, and to your competitors' on the same prompt set.
Nothing else.

The score only means something next to the prompt set behind it. The same number
on a tight set of buying prompts and on a set padded with generic category
questions mean two different things.

## Content Coverage [#content-coverage]

**How much of the tracked prompt set you have material addressing.** Coverage is
the ceiling on visibility: a model cannot cite a page you never wrote.

Reading them together is a way of using the two numbers, not a view in the
product:

| Coverage | Visibility | Reading                                                                                                                    |
| -------- | ---------- | -------------------------------------------------------------------------------------------------------------------------- |
| High     | High       | Working. Keep the prompt set fresh.                                                                                        |
| High     | Low        | You wrote it and it is not being cited. A [sources](/platform/ai-visibility/cited-sources) problem, not a writing problem. |
| Low      | Low        | Straightforward. Write the missing pages.                                                                                  |
| Low      | High       | Rare. Someone else's page is carrying you: fragile, and worth fixing.                                                      |

## Per-model Breakdown [#per-model-breakdown]

ChatGPT, Gemini, Perplexity and Claude are scored separately, and in practice
they disagree on the same prompt. The split tells you where to look:

* **Strong on one model, weak on the rest.** The material exists somewhere the
  other models are not reading. Compare the cited sources for those prompts.
* **Weak on one model everywhere.** Look at what that model cites instead, in
  [cited sources](/platform/ai-visibility/cited-sources).
* **Even and low.** Nothing you have published is addressing the question. A
  content problem, not a distribution one.

`get_tracked_prompt_history` filters by model, so you can pull the per-model
evidence behind any of these over time.

## Competitor Visibility [#competitor-visibility]

A visibility percentage for each competitor you track, on the same prompts. Read
where the gap concentrates: a competitor who owns comparison prompts is worth a
campaign this quarter.

## What Movement Means [#what-movement-means]

<Steps>
  <Step title="Ignore single scans">
    Models do not give the same answer every time. One scan is one sample.
  </Step>

  <Step title="Read the trend">
    Look at several scans in a row, not at one point.
  </Step>

  <Step title="Attribute against publishes">
    A rise after publishing pages that targeted those prompts is a story you can
    defend. Scan *before* you publish. See
    [tracked prompts](/platform/ai-visibility/tracked-prompts#scans).
  </Step>
</Steps>

<Note>
  Reports for a period download as PDF or DOCX via `list_ai_visibility_reports`
  and `get_ai_visibility_report` over
  [MCP](/mcp-server/tools#ai-visibility). For the underlying prompt, model and
  competitor evidence, use `get_tracked_prompt_history`.
</Note>