Every agent you set up in a Hub acts according to the instructions you give it: its role, its tone, what it can promise, and what it should avoid. The quality of these instructions directly determines the quality of its replies and actions. This article gathers best practices for writing them effectively, whichever Hub you use.
Where to enter your agent's instructions
The field's name and layout vary from one Hub to another, but the goal stays the same: tell the agent how to behave.
Phone Hub (Lily) — the System prompt field, in Customization > Communication, which you can write freely or build with the Guided form and Generate with AI buttons.
Acquisition Hub (Arthur) — several blocks in Customization, including Skills (agent goal, link to promote), Communication (tone, register, response delay), and Disqualification (behavior rules, automatic stop situations).
Mail & Chatbot Hub (Margot) — the Skills block in Customization, with What the agent can do and promise, Behavior rules, Response instructions by situation, and the Generate from AI Brain button. For a CHATBOT agent, Behavior rules are replaced by a free-form Instructions field, with the Improve with AI button.
The Creative Hub and the SEO Hub don't offer agent customization: Charlotte works from your campaign briefs, and Amandine relies solely on the URL being analyzed.
Be clear and specific about the agent's role
A vague instruction like “reply nicely to customers” leaves the agent guessing what you expect. Instead, specify the context, the goal, and the expected outcome: “You are the customer support assistant for an online clothing store. Answer questions about delivery times and returns based on our terms and conditions, and always include the order tracking link.” The more concrete the situation the instruction describes, the better suited the agent's reply will be.
Rely on the AI Brain instead of rewriting everything
You don't need to copy your positioning, products, or brand tone into every instructions field: that information already lives in the AI Brain and is available to all your agents. An agent's instructions should focus on what's specific to it (its role in this particular Hub, the actions it can take) rather than on company information already centralized elsewhere.
Set the agent's tone and boundaries
Beyond the role, specify the register (informal or formal address), the tone you want, and the topics to avoid. The Behavior rules block (Acquisition Hub and the MAIL agent in the Mail & Chatbot Hub) and the Automatic stop situations block (Acquisition Hub) let you set explicit rules: never criticize a competitor, don't make up features, stop the conversation if there's a refund request or aggressive language. These boundaries prevent off-topic or inappropriate replies without having to spell everything out in the prompt itself.
Use the writing assistance tools
You don't have to write complex instructions on your own. The Guided form breaks the writing down into simple questions, the Generate with AI button writes a first draft from a free-form description of what the agent should do, and Generate from AI Brain automatically composes instructions consistent with your brand. You can then review and adjust the result instead of starting from a blank page.
FAQ
Do I need to fill in every Customization field for the agent to work?
Do I need to fill in every Customization field for the agent to work?
No. An agent works as soon as a minimum of instructions is filled in. The additional fields (sales strategies, instructions by situation, etc.) fine-tune its behavior, but they aren't all mandatory.
Can I change the instructions after the agent goes live?
Can I change the instructions after the agent goes live?
Yes. You can change the instructions at any time from the Hub's Customization section, and the changes apply as soon as you save, without interrupting the agent.
Does the Generate with AI button replace my own instructions?
Does the Generate with AI button replace my own instructions?
No, it suggests a starting draft that you can freely edit before saving it. Nothing is applied without your approval.
Are the instructions the same for all assistants in a Hub?
Are the instructions the same for all assistants in a Hub?
No. In Hubs that allow multiple assistants (Acquisition, Mail & Chatbot, Phone), each assistant has its own instructions, tone, and context.