# Writing assistant

> Draft and rephrase your own replies in your agent's tone, without handing the conversation back to Autopilot.

- Source: https://support.unifunl.io/en/help/inbox/writing-assistant
- Category: Inbox (`inbox`)
- Language: en
- Updated: 2026-08-20

---

Sometimes you want to answer a customer personally but you do not want to spend five minutes finding the right wording. The writing assistant sits in the message composer and drafts or rephrases for you, in the tone your agent already uses.

![The buttons under the message box: 1 = writing assistant, 2 = Send.](https://support.unifunl.io/help/screenshots/25-composer.png)

## Write with the assistant

1. Open the conversation and click into the message box.
2. Either write a rough version of your reply, or leave it empty.
3. Use the **writing assistant** button under the box.
4. Read what it proposes, edit it if you want, then **Send**.

Nothing is sent until you press Send. The assistant only ever fills the box.

> **Good to know** — the assistant follows the personality and the rules set in [Agent behaviour](https://support.unifunl.io/en/help/ai-center/agent-behavior), so its drafts sound like the rest of your conversations rather than like a generic chatbot.

## The assistant is not Autopilot

These are separate features and it is worth keeping them apart:

- **AI Autopilot** replies to the customer without you.
- **The writing assistant** proposes text to *you*, and you decide whether to send it.

You can use the assistant on a conversation where Autopilot is off. That is the normal way to take over a customer personally while still writing quickly.

## FAQ and troubleshooting

### The assistant's draft is in the wrong language

The agent answers in the customer's language, including when several are mixed in one message. If a draft comes back in a language you did not expect, check which languages the agent is allowed to use in the AI Center.

### Can I attach an image or an emoji?

Yes. The composer carries emoji and attachment buttons next to the assistant.

## What's next?

When the agent's own reply is wrong, flag it so the team can fix the cause.

[Flagging a bad reply](https://support.unifunl.io/en/help/inbox/flagging-a-reply)

## Related articles

- [AI Autopilot](https://support.unifunl.io/en/help/inbox/ai-autopilot.md): Let the agent answer on its own, and take back a single conversation without stopping it everywhere else.
- [Using the inbox](https://support.unifunl.io/en/help/inbox/using-the-inbox.md): Work every conversation from one screen — the list on the left, the exchange in the middle, and the contact record on the right.
- [Agent behaviour](https://support.unifunl.io/en/help/ai-center/agent-behavior.md): Pick a personality, set the tone, choose the languages and script, and set the guardrails with Always do, Never do and escalation rules.