---
title: Collect Feedback on AI Smart Replies
slug: ai-assist/collect-feedback-on-ai-smart-replies
createdAt: 2026-05-04T08:52:13.280Z
---

Smart Reply Feedback lets agents rate AI-generated reply suggestions during live interactions. When enabled, agents can mark each suggestion as Good or Bad — and when marking Bad, select a specific reason. This gives administrators a clear signal on suggestion quality and helps improve the AI assistant over time.

## For Admins: How to Configure Smart Reply Feedback

![](https://api.archbee.com/api/optimize/SK2urIgCC2ml99YyqWOSl/1F2vRqINSvcZ64o34SlgC_image.png)

Smart Reply Feedback is configured in the assistant's setup flow, under the **Configuration** step.

1. Open the assistant you want to configure and go to **Step 2: Configuration**.
2. Scroll to the **General Configuration** section.
3. Locate **Smart Reply Feedback** and toggle it to **Enabled**.
4. Once enabled, a **Customise Bad Feedback Reasons** section appears. You can define up to three labels that agents see when they mark a suggestion as Bad. Each label can be up to 15 characters. The defaults are **Irrelevant**, **Wrong**, and **Outdated** — update these to suit your team's needs.
5. Proceed through the remaining steps and click **Publish & Go Live** to apply the changes.

:::BlockQuote
**Note:** Smart Reply Feedback requires the **Smart Reply** toggle to also be enabled. If Smart Replies are turned off, feedback cannot be collected.
:::

## For Agents: How to Give Feedback on Smart Replies

![](https://api.archbee.com/api/optimize/SK2urIgCC2ml99YyqWOSl/UIjr6RtmQ7ysJR-a2VYu7_image.png)

When Smart Reply Feedback is enabled, feedback controls appear next to every suggestion in the AI Assist panel during a live interaction.

1. During an active call or chat, open the **AI Assist** tab in the right-hand panel.
2. The **Suggestions** tab will show AI-generated reply cards for the current conversation.
3. To rate a suggestion as **Good**, click the **thumbs up (👍)** icon next to it. The icon turns green to confirm.
4. To rate a suggestion as **Bad**, click the **thumbs down (👎)** icon. Reason buttons will immediately appear below the card.
5. Select the reason that best describes the problem — for example, **Irrelevant**, **Wrong**, or **Outdated**. Only one reason can be selected per suggestion.

Feedback is optional and does not affect how the suggestion is used. You can still copy and use a suggestion even if you mark it as Bad.

## For Admins: How to View Recorded Feedback

![](https://api.archbee.com/api/optimize/SK2urIgCC2ml99YyqWOSl/7e0TNt3qPsElAUsxHxTZg_image.png)

All agent feedback is captured and available in **Interaction History**.

1. Navigate to **Interaction History** from the main menu.
2. Click the **Feedback** tab at the top of the page.
3. Use the filters to narrow the data by **Date**, **Assistant Name**, or **Feedback** type (Good / Bad).
4. Each row represents a smart reply suggestion that received feedback. The table shows: 
   - **Interaction ID** — the unique ID for the customer interaction
   - **Date and Time** — when the suggestion was shown to the agent
   - **Assistant Name** — the AI assistant that generated the suggestion
   - **Smart Reply** — a preview of the suggestion text
   - **Transcript** — the customer message that triggered the suggestion
   - **Feedback** — Good or Bad rating given by the agent
   - **Bad Feedback Reason** — the reason selected when feedback was Bad (e.g. Outdated, Wrong); a dash means no reason was provided or feedback was Good

**Using feedback to improve the assistant:**

- A pattern of **Outdated** feedback indicates the knowledge base has stale content that needs refreshing.
- Frequent **Wrong** feedback suggests the assistant is misinterpreting certain query types — review and refine the source content for those topics.
- Frequent **Irrelevant** feedback may mean certain customer intents lack dedicated coverage in the knowledge base.

Reviewing this data weekly and comparing trends over time is the most effective way to drive measurable improvement in suggestion quality.