{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/88f85986e0014426be0acc47c5aef248\" frameborder=\"0\" width=\"1280\" height=\"960\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":960,"width":1280,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":960,"thumbnail_width":1280,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/88f85986e0014426be0acc47c5aef248-c70a600313111897.gif","duration":186.633333,"title":"When your voice prompt becomes the thing your tests wrote","description":"This Loom explains why voice prompts need a cleanup pass after iterative testing and rule additions. Himanshu describes building a Florida voice air receptionist called Olivia that answers calls, collects address and property type, identifies the call purpose, and routes it to the right team, starting with a well-structured system prompt. After a failed test and subsequent call issues, he added many rules over the day, resulting in prompt “scar tissue” where rules existed in multiple places and were hard to trace. He recommends consolidating duplicate instructions, using bullets for pattern lists, separating format from flow, and deleting or rewriting outdated rules, typically running cleanup every 8 to 10 edits and avoiding overly frequent cleanups that can cause testing loops."}