Got it. And definitely these rule prompts are very useful for making sure the assistant (LLM) performs and behaves exactly as you want it to, following your protocols However the difference here is that this logic/rule only applies to the output of the LLM, and not the output of the TTS. We should remember that the whole Vapi voice orchestration layer looks somewhat like this: Caller says something -> Caller's speech is transcribed to text via STT -> This text is passed onto the LLM (such as GPT4) -> GPT4 generates a response to the caller's speech using the logic specified in the prompt given earlier by you, the developer -> This response is passed onto the TTS (such as 11Labs) which then converts it into speech and sends it back to the caller The rule prompt works great for anything within the LLM, however this logic will not pass onto the TTS, only the response "script" will be sent to it. So the TTS will generate the speech normally as it does, ignoring the logic or instruction to speak slowly. However, if you instruct the LLM to add a considerable amount of commas and pauses in the generated output which will be passed on to the TTS provider, possibly it will influence it to speak slower. For example if you ask it to always structure output something like "Hey this is Ryan from Ryan's Dental how can I help" to "Hey, this is Ryan, from Ryan's Dental. How can I help?"