# User sentiment

Track the AI-estimated emotional tone of conversations — Great, Good, or Negative.

**Path:** Analytics → Dashboard → Conversations

User Sentiment scores the emotional tone of a conversation from the user's messages, so you can track how conversations feel, not just whether they resolved.

## Ratings

Every conversation gets exactly one rating (the ratings are mutually exclusive, so shares sum to 100%):

| Rating | Meaning |
| :--- | :--- |
| **Great** | Clearly positive tone. |
| **Good** | Neutral to positive tone. |
| **Negative** | Frustrated or negative tone. |

The widget headlines the **Good-or-better share** with its count (for example, "95.1% — Good or better · 620 of 652") and a trend chip in **percentage points** against the previous period (for example, "-0.4pp"). The legend shows the count per rating. From **⋯ → Chart Type**, display it as a **Stacked Bar** (ratings over time) or a **Summary** (a single horizontal bar of the current mix).

## Use it

- **A rising share of Negative** is an early warning that quality is slipping, often before Resolution Rate moves. Drill in with [Session replay](/user-guides/analytics/session-replay/).
- **Filter by Agent or time range** to find where the experience is worst.
- **Cross-check with retention**: if users with negative first conversations drop off, first-session quality is a retention lever.

## Related

- [Conversation metrics](/user-guides/analytics/conversation-metrics/): volume, resolution, and handoffs.
- [Signals frequency](/user-guides/analytics/signals-frequency/): what users asked for and where the agent fell short.