# Inbox

> Manage conversations in the Inbox — filter by status and assignee, read live AI Insights, see why the AI said what it said, take over from the AI, edit visitor details, and export conversations.

The **Inbox** is where you read, filter, and act on every conversation your AI Agent handles. You can take over from the AI when a conversation needs a human touch, reassign it back when the customer is unblocked, see how the AI scored sentiment / resolution / topic for each chat, and export everything for analysis.

## Optional example checks

- [Preview metadata keys][public-lab-metadata-keys-post]. Check how supplied field names become variable keys, including duplicate names and reserved-key errors. This does not inspect values or account-specific permissions.

## Inbox views

The Inbox sidebar groups conversations by status, with live counts next to each entry:

- **All** — every conversation the AI Agent has ever handled.
- **Open** — active conversations still in progress. Open splits into two sub-views:
  - **Assigned** — Open conversations currently assigned to a teammate (i.e., a human has taken over).
  - **Unassigned** — Open conversations sitting in the shared queue, waiting for a human (typically because a Human Handoff just fired). Treat this like an alert queue — if your Agent is configured to hand off, conversations land here when they need attention.
- **Closed** — resolved or auto-closed conversations. A conversation moves to Closed when the AI marks it resolved (e.g., after a long period of inactivity) or when a teammate explicitly closes it.

## Conversation list

The conversation list (middle pane) is sorted newest first. Each row shows:

- The visitor's name (or **Unknown Visitor** if no contact details have been captured).
- The channel icon (Widget, WhatsApp, Slack, Telegram, etc.).
- A preview of the latest message.
- A relative timestamp (e.g., `18 m`, `4 h`, `3 d`).
- The current **Assignee** (AI Agent or a teammate).
- An optional **[Conversation badge](#conversation-badge)** — a small chip summarising the conversation (its Topic, Sentiment, Language, a metadata value, etc.). Hidden by default; turn it on from the Inbox settings menu.

### Conversation badge

Every row in both the **Inbox conversation list** and the per-conversation **AI Insights** list can show a small **badge** that summarises the conversation at a glance. Open the Inbox settings menu and pick **Conversation badge** to choose what it shows (the **Badge shows** dropdown):

- **Topic** — the auto-detected or user-defined [Topic](/docs/conversations/insights/manage-topics). This is what AI Insights shows by default.
- **Custom topic** — only the Topic a teammate explicitly set on the conversation.
- **Sentiment** — Positive / Neutral / Negative.
- **Intent** — the detected user intent (e.g., purchase, support, complaint).
- **Resolution** — the Resolved / Unresolved outcome.
- **Language** — the conversation language.
- **Channel** — the source channel. Shown in AI Insights only, since the Inbox already shows a channel icon on every row.
- **Conversation data** — any value your integration injects into the conversation's [metadata](/docs/ai-agent/variables), surfaced as the badge so you can spot and visually group conversations by what matters to your business at a glance. Useful for things like:
  - the **product, model, or SKU** the visitor is asking about (e.g., a camera model or a specific plan),
  - the customer's **account tier or plan** (e.g., Free / Pro / Enterprise) so high-value conversations stand out,
  - the **store, brand, region, or storefront** the chat came from, when one Agent serves several,
  - an **order, ticket, or subscription status** (e.g., Shipped, Past due, Trial),
  - a **marketing campaign or UTM source**, to see which traffic a conversation came from,
  - an **A/B test or experiment variant**, to compare how each cohort chats.

  Choose the **Data field** from the list of keys seen in recent conversations.

By default the Inbox conversation list shows no badge and AI Insights shows the **Topic**, so existing setups are unchanged until you pick a different source. The badge reflects whatever the conversation currently holds: if the chosen metadata field is missing on a given conversation, that row simply shows no badge.

The only control at the top of the list is a free-text **search** ("Find words or phrases") that matches against message content. To filter by status or assignee, use the [Inbox views](#inbox-views) in the sidebar (All / Open [Assigned · Unassigned] / Closed); within **Open › Assigned** a teammate filter lets you narrow to a specific assignee.

## Conversation window

Clicking a conversation opens it in the centre pane. The header above the messages shows several chips that help you orient quickly:

- **Title** — the first chip shows the conversation title, prefixed with the channel source icon (Widget, WhatsApp, Slack, Telegram, etc.). It's a label, not an interactive control. (To copy the conversation id, use the per-field **Copy** icon in the right-hand details pane.)
- **Translate** — the chip reads **Original** by default; open it to translate the whole conversation into another language for reading. Useful when the visitor wrote in a language you don't read. While translating it shows _"Translating…"_, and once applied it reads **Translated**.
- **Rating** — hover the star to see the visitor's conversation rating (thumbs / CSAT and any written feedback). The star is filled when a rating exists.

Below the header you'll see the full message history. Every AI message has a **Why AI said that** link directly under it (see next section), next to the thumbs you use to [rate the reply](#rate-ai-replies).

### Why AI said that

Click **Why AI said that** under any AI message to open a traceability panel that explains how the AI produced that specific response. It surfaces:

- Which **Knowledge Base articles** the AI retrieved and used for the answer.
- Which **AI Actions** (if any) the AI decided to call, with the request payload and the response it received back.
- The reasoning the AI used when assembling the final answer.

**Score = Influence** is an AI-generated estimate, from 1 to 10, of how much a Knowledge Base source contributed to that particular reply. It is produced by a separate analysis after the reply is generated.

A higher score means greater estimated influence. For example, an article providing the main answer may score **10/10**, while another supplying supporting detail may score **7/10**. These are illustrative examples, not fixed scoring bands. A short passage can have a high score if it provides a key fact.

The score does not grade the article's quality or accuracy, measure the percentage of content used, or express confidence that the answer is correct. Each source is scored separately; scores do not have to add up to 10. The same article can receive different scores for different replies. Use the quoted snippets and explanation alongside the score.

Use this when:

- A response looks wrong or hallucinated — Why AI said that tells you whether the AI used the wrong KB article, called the wrong action, or invented something on its own.
- You want to verify that a new KB article or AI Action is actually being picked up.
- A user disputes an answer and you need an evidence trail for compliance or QA.

<div id="copyable-drafts"></div>

### Copyable text boxes

In the **Inbox**, **Chat Page**, and website widget, an AI reply can show text in a box with its own **Copy** button. The button copies only the contents, preserving paragraphs, spacing, and URLs. Markdown and HTML inside remain literal text. For recognized blocks, it replaces whole-message copying. Each alternative has its own box.

Use this for email drafts, social posts, reusable replies, or code snippets you want to copy as plain text. These boxes do not add syntax highlighting. Regular `python` or `js` fences keep their existing rendering; use the `draft` tag for this copy target.

![An email reply in a text box with a Copy button in the top-right corner.](https://quickchat.ai/docs/_astro/copyable-text-box.EByo0hfB.png)

This format is available to all users. Quickchat does not add these instructions to your prompt automatically. To opt in, add an instruction like this to your Agent's **Main Prompt** or **Guidelines** in [Identity](/docs/ai-agent/identity):

````text
When providing text for the user to copy:
- Put each copyable item in its own fenced Markdown block.
- Open with exactly three backticks followed by draft.
  Close with three backticks on their own line.
- Put only the text to copy inside. Keep explanations and notes outside.
- Use plain URLs. Close every block, including revisions and translations.

Example:
```draft
Hi Alex,

Thank you for your message. The next steps are here:
https://example.com/support

Kind regards,
Customer Support
```
````

Copying is unavailable during generation or when a recognized block has incomplete or mismatched boundaries. There is no whole-message fallback in these cases. Inbox translations copy the displayed language only when complete blocks match the original; return to the original message if the boundaries are lost.

Review the box before pasting or sending. Everything inside is copied, and notes outside remain visible in the conversation. Prompt instructions cannot guarantee correct formatting. If the AI omits the `draft` marker entirely, the reply keeps ordinary whole-message copying.

<div id="rate-ai-replies"></div>

### Rate AI replies

Under every AI reply except the welcome message, click the thumbs up (**Good response**) or the thumbs down (**Bad response**). Your rating is saved immediately, even if you close the details without submitting. In the details, you can pick reasons such as **Accurate answer** or **Incorrect information**, add an optional comment, and click **Submit** to save them. Click the selected thumb again to edit or remove the rating, or the other thumb to change it.

Ratings stay within your team: the visitor never sees them, and they are separate from the visitor's own **Rating** in the conversation header. Both **Export Conversations** and **Export Gathered Data** include them, so you can review every rating and comment in a spreadsheet instead of opening each conversation.

## Right-hand details pane

The right rail shows everything Quickchat knows about the conversation, broken into three sections.

### Insights

Live AI analysis of the conversation, refreshed as new messages arrive:

- **Sentiment** — Positive / Neutral / Negative, based on the user's tone and the resolution they got.
- **Resolution** — whether the conversation ended Resolved or Unresolved. Used in the [Insights dashboard](/docs/conversations/insights/analytics) to track resolution rate over time.
- **Topic** — the auto-detected (or user-defined) [Topic](/docs/conversations/insights/manage-topics) that best matches what the visitor was asking about.

While analysis is running, each Insight shows an "analysis in progress" placeholder until it finalises. Insights typically finalise within a few seconds of the conversation closing.

### Conversation

Metadata about the conversation itself:

- **Assignee** — the AI Agent or a specific teammate (see [Reassigning](#reassigning-a-conversation) below).
- **Channel** — Widget, Embed, Chat Page, WhatsApp, Slack, etc.
- **URL** — the page the visitor was on when they started chatting (where applicable — populated from `{{metadata_fullPathURL}}`).
- **Language** — ISO code of the conversation language, or `–` if not yet detected.
- **Started at**, **Last active**, **Duration** — timestamps and elapsed time.

Click the small **Copy** icon next to any field to copy its value to your clipboard.

### Visitor

Contact details for the visitor. Each field is editable inline:

- **Name**
- **Email**
- **Phone**

If Smart Data Gathering or an Action input mapping in [Save to memory](/docs/ai-agent/actions#save-to-memory) captured these during the chat, they appear automatically. Otherwise click **+ Add** next to each field to enter them manually. This is handy when you took a phone call after the chat and want to record the lead's contact info against the conversation.

Anything you add here is also available in [AI Action Variables](/docs/ai-agent/variables) as `{{metadata_visitor_name}}`, `{{metadata_visitor_email}}`, and `{{metadata_visitor_phone_number}}` for any future actions the AI runs in this conversation.

## Reassigning a conversation

At the bottom of every conversation you'll see two controls:

- **Assign to me** — instantly takes the conversation from whoever (or whatever) currently holds it and assigns it to you. Use this when you spot a conversation in the **Unassigned** queue and want to handle it yourself.
- **Assignee dropdown** — shows the current assignee (a teammate's name, **AI Agent**, or **Unassigned**); open it to pick any teammate, **Unassigned**, or **AI Agent**. Reassigning has these effects:
  - **AI Agent** → the AI takes over again and will reply to the next visitor message.
  - **A specific teammate** → that teammate becomes responsible; the AI stops replying.
  - **Unassigned** → the conversation returns to the shared Unassigned queue.

Once you've taken over, you can reply to the visitor with text directly from the conversation composer, on the **Website Widget**, **Embed**, and **Chat Page**, and on **WhatsApp**, **Telegram**, **Messenger**, **Instagram**, **Discord**, and **Email**. **Images and files** can be sent on the **Website Widget**, **Embed**, and **Chat Page**, and on **WhatsApp**, **Instagram**, and **Messenger**.

On Instagram and Messenger, use **Attach file or image** to send a **JPEG or PNG image** (up to 5 MB) or a **PDF** (up to 16 MB). Other file types and audio are not supported on these two channels. The file is sent as its own message without a caption. Any text you've typed stays in the composer; send it separately after the file finishes sending.

Conversations are **automatically closed** by the AI after a period of inactivity. To explicitly close or re-open a conversation, use the assignee + status controls on the conversation row or in its header menu.

## Export options

Click the export icon at the top of the conversation list to access the export dropdown:

- **Export Conversations**: full conversation transcripts (CSV or XLSX), one row per message, within a date range you specify. When a teammate has [rated an AI reply](#rate-ai-replies), that reply's row also shows the rating, its reasons and comment, who left it, and when. Requires the **Business** plan; see [Plans & limits](/docs/pricing-and-plans).
- **Export Gathered Data**: one row per conversation. It holds the visitor details captured by Smart Data Gathering (Name, Email, Phone, plus any custom fields you configured), the conversation's Insights (topic, sentiment, resolution status, flags) and conversation data fields, the visitor's conversation rating, and your team's ratings of AI replies (positive and negative counts, plus every comment). Use it to feed leads into your CRM, or to review answer quality across many conversations.
- **Export AI Insights**: every item from the AI Insights section (flagged conversations, market insights and matched custom alerts), one row per item, with its type, detail, topic and reviewed state.

These exports support a date range picker and a format toggle (CSV / XLSX).

Conversation and gathered-data exports are prepared in the background. Keep Quickchat open for the download to start automatically when the file is ready. If your browser blocks it, click **Download** in the completion notification. With **Email download link** enabled, you also receive a link at your account email when the file is ready, even if you leave the page. Emailed links are valid for 24 hours.

Only one gathered-data export can be prepared at a time for each AI Agent. Wait for it to finish before starting another with different settings. A conversation export can run alongside it.

Gathered-data XLSX exports support up to 100,000 conversations per file. Choose CSV for larger exports or when the file has more columns than Excel supports.

For conversation and gathered-data exports, use **Advanced filtering** to restrict the file to one observed **Conversation data field** and an **Exact value**. For example, select `modelName` and enter a model value. Matching is exact. Select **All conversations** to leave metadata filtering off.

[public-lab-metadata-keys-post]: https://app.quickchat.ai/v1/api_core/public-question/quickchat/tools/metadata-keys/lab/post/?placement=docs-inbox-metadata-keys-post-v1
