> ## Documentation Index
> Fetch the complete documentation index at: https://docs.recepai.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Chat

> How your receptionist behaves when a guest types — her manner, how she hands over to your team, and what you count as a good conversation.

**Chat** sets how your receptionist behaves in written conversations, wherever a guest types to her.

She already knows your hotel from [Training](/ai-manager/training/overview) and she already has an identity from [Profile](/ai-manager/receptionist/profile). This screen is about *manner*, not knowledge.

***

## What is on the screen

### Core chat personality

Her instructions for written conversation. Once you have published her identity, this is written from it and the box is locked, with her identity shown above it — so what you read here is what she is actually running.

If you need to change how she comes across, the right move is almost always to edit her identity in [Profile](/ai-manager/receptionist/profile) and publish again. The screen does offer a way to unlock the text and edit it directly, for the rare case where you need a wording change that the identity does not cover.

<Warning>
  **Do not put hotel facts in here.** Prices, hours, room types and policies belong in [Training](/ai-manager/training/overview). A fact typed into this box is invisible to your other channels and easy to forget when it changes. This box is for behaviour.
</Warning>

### Connect to Staff

Where a guest goes when your receptionist cannot help — a phone number, an email address, a messaging channel, or all three. She offers these when a guest asks for a person, or when the request is beyond what she can do.

Fill in at least one. Without it, a guest who needs a human is left without a next step.

### Thinking engine

Which engine she thinks with. This is set for your hotel; on most accounts it is shown here for information rather than as something to change.

### Quality criteria

What you count as a good written conversation. Each criterion has a **name** and a **success description** — plain language, in your words.

> **Clear and complete answer** — *the guest's question was answered fully, without them having to ask again.*

She grades herself against these after conversations, and what she flags is what you read later in [Analytics](/ai-manager/analytics/chat-history).

### Advanced settings

Behind a separate control, there are finer settings that change how she composes an answer. They are tucked away because a wrong value is hard to notice and easy to blame on something else. Leave them alone unless someone has asked you to change one.

***

## What you do here

<Steps>
  <Step title="Add your handover contacts">
    Fill in Connect to Staff so a guest who needs a person always has one.
  </Step>

  <Step title="Write two or three quality criteria">
    Start small and specific. Two criteria you actually care about beat eight generic ones — she reports against every one of them, and vague criteria produce vague reports.
  </Step>

  <Step title="Read her personality once">
    Check that it matches how your hotel talks. If it does not, fix it in Profile rather than here.
  </Step>

  <Step title="Save">
    Changes take effect on the next conversation.
  </Step>
</Steps>

***

## What happens next

From the next guest message, she behaves the way this screen says. After conversations she grades herself against your criteria and reports what fell short, along with the topic that was missing.

That report is the loop that makes her better every week — it lands in [Analytics](/ai-manager/analytics/chat-history).