You can do this by structuring the script as modular intents and guardrails instead of a rigid word-for-word script, then letting the agent paraphrase within those boundaries. With Vapi, this usually means using a strong system prompt, dynamic variables, and fallback responses to keep it on-script while still sounding natural. I’ve implemented scripted yet conversational voice agents before, so I can help you set up the prompt design and flow logic to balance consistency with authenticity. @Harley