I want the chatbot to only chat and not provide descriptions of the scenario
I am creating a chat simulation but the chatbot always gives me details of the environment surrounding the conversation but i want convesrsation only so when I phone the chatbot I can have a proper conversation - how do i disable the scenario parts of the conversation?
1 Reply
Here’s how you can stop your Voiceflow agent from adding the extra “scene-setting” lines and keep it strictly conversational:
1. Find the prompt that’s driving the bot’s replies
• Either open the Prompt Step inside the flow, or open Agent Settings → Behaviour → System Instruction if you placed the main prompt there. 2. Rewrite the instruction so it’s crystal-clear what the model can (and cannot) output. A short template that works well is: 3. Add a quick example so the model sees the format you expect: 4. Save and test. If any step still produces narrative text, locate the Prompt/Message step that sent it and apply the same instruction. Why this works
Voiceflow simply forwards your prompt to the LLM, so the model follows whatever rules you give it. Including an explicit “no environment descriptions” line plus a formatting rule (“only what you would say next”) removes the scenario text. The doc on prompt design (“Crafting great prompts”) recommends adding clear Instruction, Formatting and Examples exactly for cases like this. Need to tweak further?
• Reduce “Max tokens” on the Prompt step to keep answers short.
• Turn the guidance into a global style rule under Agent Settings if you want it to apply everywhere in the project. Docs: https://docs.voiceflow.com
• Either open the Prompt Step inside the flow, or open Agent Settings → Behaviour → System Instruction if you placed the main prompt there. 2. Rewrite the instruction so it’s crystal-clear what the model can (and cannot) output. A short template that works well is: 3. Add a quick example so the model sees the format you expect: 4. Save and test. If any step still produces narrative text, locate the Prompt/Message step that sent it and apply the same instruction. Why this works
Voiceflow simply forwards your prompt to the LLM, so the model follows whatever rules you give it. Including an explicit “no environment descriptions” line plus a formatting rule (“only what you would say next”) removes the scenario text. The doc on prompt design (“Crafting great prompts”) recommends adding clear Instruction, Formatting and Examples exactly for cases like this. Need to tweak further?
• Reduce “Max tokens” on the Prompt step to keep answers short.
• Turn the guidance into a global style rule under Agent Settings if you want it to apply everywhere in the project. Docs: https://docs.voiceflow.com
Voiceflow
Welcome to Voiceflow's docs
Build, manage, and deliver chat and voice agents for customer support and beyond.