How do I control exactly what my AI agent says, word for word?

Build the agent with the visual drag and drop workflow builder instead of only a system prompt. With a system prompt, the AI decides what to say and varies from one conversation to the next. The visual builder gives you control and predictability, for example for a business that needs a specific script or compliance wording.

1. Start with the Conversation started trigger. It covers every conversation type: voice, chat and SMS, plus connected email and DM channels.
2. Add a Conversation node for what the AI says. This node is the core of the builder and is how the AI talks on the phone.
3. Choose how the node speaks. Static response: the AI says exactly the words you wrote and never deviates. AI generated: you give instructions, such as greet the user by first name and then tell them the reason we are calling, and the AI words it.
4. Add routes, for example a yes route and a no route, or sales and support. The AI decides which route the caller's answer matches, and it can wait or stay on the step if the answer is not a clear match, like a salesperson with a script.
5. Tick Auto continue when a step should go straight to the next one without waiting for the caller to speak.
6. Add an Ask and Save step to capture information, a bookable step to book an appointment, or a Decision node to branch on data.

You can edit the wording live while testing, save, and rewind the test conversation to hear the change. Because the AI always says what you wrote, you can optimize the conversation by changing one line and seeing how that path performs.

If the agent also has an opening greeting set and the first node already greets the caller, the AI says two things at once, so remove the opening greeting in that case.

If the client does not need this level of control, or the conversation is not linear, the system prompt way or AI generated nodes are more flexible.