Customize the assistant's behavior with its own instructions

For the assistant to work, you don’t need to add any instructions of your own. In most cases, we recommend that you start without instructions and only add them if you see a need to influence the assistant’s behavior.

Start simple

Feel free to test the assistant without your own instructions first. If you notice behaviors you’d like to adjust, you can add instructions gradually as needed.

Keep in mind that AI models often interpret instructions more literally than humans. Instructions that might be perceived as nuanced by a human therefore risk having a stricter effect than intended.

The assistant already handles this

You do not need to write instructions to:

• Describe what information is available to the assistant.
• Specify that the assistant should only respond based on available information.
• Prevent the assistant from making up information (hallucinating).
• Instruct the assistant to ask for clarification when a question may have multiple possible answers.

These behaviors are already handled by the assistant’s default settings.

Examples of things you can influence

If you want to customize the experience, you can, for example, provide instructions regarding:

• Role and context—briefly describe the role the assistant should play and the context in which it is used.
• Tone—for example, formal, friendly, or professional.
• Response structure—for example, that responses should preferably be presented in bullet points.
• Level of detail—for example, whether answers should be concise or more detailed.
• Justification and transparency—whether you want the assistant to clearly show how a response is based.
• Salutations—whether you want to use a specific opening or closing.