# Set AI instructions (/core/govern/ai-control-and-optimization/how-to/set-ai-instructions)



<CalloutContainer type="info">
  <CalloutDescription>
    **Target Audience**: Organization admins and owners configuring AI for their team.
  </CalloutDescription>
</CalloutContainer>

### TL;DR [#tldr]

AI instructions let you define how the assistant behaves across your whole organization — what format to use, what to avoid, and any guardrails. These instructions apply to every user by default, but can be overridden per user when generating a token.

***

### When to use this [#when-to-use-this]

Use this page when you want the assistant to follow consistent rules: always respond in a specific language, avoid mentioning certain topics, structure answers in a particular format, or stay within a defined scope.

***

### Who can change this [#who-can-change-this]

Changing AI instructions requires the **organization update** permission, which **member**, **admin**, and **owner** roles have. Users with the **explorer** role cannot change it.

This setting is available on plans that include AI customization. If the section is locked, contact your account manager.

***

### How to set AI instructions [#how-to-set-ai-instructions]

1. Open **Settings** and select **AI behavior** in the left sidebar.
2. In the **AI instructions** card, type your instructions in the text field.
3. Click **Save**. A confirmation ("Context updated") appears once the change is applied.

<CalloutContainer type="info">
  <CalloutDescription>
    **Examples of useful instructions**

    * `Always respond in bullet points.`
    * `Never mention competitor products.`
    * `Include the data source in every answer.`
    * `Always respond in French, regardless of the language of the question.`
  </CalloutDescription>
</CalloutContainer>

***

### Good to know [#good-to-know]

* **User-level override**: instructions set here apply to all users by default. You can override them for a specific user by passing a custom system prompt when generating that user's token. See [Generate a token](/core/embed/authentication/how-to/authentication-and-tokens).
* **Character limit**: AI instructions are capped at 3,000 characters.
* **Scope**: instructions apply to the AI assistant across all contexts — the main chat, dashboard editor, and embedded experiences.

***

### Related pages [#related-pages]

* [Set company context](/core/govern/ai-control-and-optimization/how-to/set-company-context)
* [Set tone of voice](/core/govern/ai-control-and-optimization/how-to/set-tone-of-voice)
* [Define metric definitions](/core/govern/ai-control-and-optimization/how-to/define-metric-definitions)
* [Choose your AI provider and model](/core/govern/ai-control-and-optimization/select-your-ai-provider)