---
name: visibility-audit
description: >
  Audit a brand's AI visibility, sentiment, response volume, mentions, and competitor
  trend in Promptwatch. Use when someone asks for an AI visibility audit, GEO health
  check, brand mention report, sentiment trend, or model-by-model visibility summary.
---

# Visibility Audit

## Prerequisites

Requires the Promptwatch MCP server. Add to your agent config:

```json
{
  "mcpServers": {
    "promptwatch": {
      "url": "https://server.promptwatch.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}
```

Get your API key at https://app.promptwatch.com -> Settings -> API Keys. Run `setup-promptwatch-mcp` if connection is not verified.

## Tools

- `list_projects`: resolve project.
- `get_visibility_time_series`: pull visibility scores over time.
- `get_sentiment_time_series`: pull sentiment trend.
- `get_brand_visibility_over_time`: compare brand visibility against competitors.
- `get_responses_summary`: get response and mention totals.
- `get_mentions_time_series`: inspect mention frequency.
- `list_responses`: sample actual AI responses by sentiment, model, or date range.

## Workflow

1. Resolve the project. Ask only if `list_projects` returns multiple plausible projects.
2. Use the last 30 days by default with daily granularity.
3. Pull the current period and the previous equivalent period for visibility, sentiment, mentions, and responses.
4. Pull brand visibility over time with competitors included when configured.
5. Sample responses from low-visibility models and negative-sentiment slices.
6. Compare current vs previous period before drawing conclusions.

## Analysis Rules

- Label visibility as `up`, `down`, or `flat` using current vs previous period.
- Flag any platform/model with 0 visibility as invisible.
- Flag sentiment shifts over 10 percentage points.
- Treat low sample size as a caveat, not a trend.
- When visibility is low, suggest `content-gap-audit` or `competitor-benchmark`.

## Output

Produce:

1. Executive summary with overall visibility score, period-over-period change, and confidence caveats.
2. Platform breakdown showing which LLMs mention the brand and which do not.
3. Sentiment summary with positive, neutral, and negative movement.
4. Brand vs competitor comparison, focused on the top 3 competitors.
5. Evidence from sampled responses.
6. Top 3 opportunities with concrete next steps.

If tools return no data, say which dataset is empty and recommend checking project setup, tracked prompts, model coverage, or date range.