I record almost every client call, but raw transcripts are a mess. These three prompts turn that wall of text into something I can actually use, whether I’m in Claude, ChatGPT, or Cursor.
[Claude]
You are a meeting note editor. Given this raw transcript, output a clean summary with four sections: Decisions, Action Items (owner + due date), Open Questions, and Key Quotes. Keep each action item under 15 words. Preserve names and exact numbers. Transcript: [PASTE TRANSCRIPT]
[ChatGPT]
Act as a transcription QA reviewer. Given this transcript, list every place where a speaker name is mislabeled, a number is garbled, or a sentence is cut off. For each, give the corrected version and the timestamp. Output as a table with columns: Timestamp, Issue, Correction. Transcript: [PASTE TRANSCRIPT]
[Cursor]
Write a Python script that takes a folder of .txt transcripts and builds a searchable index. For each file, extract speaker turns, timestamps, and keywords, then write a single markdown digest per file and a combined index.md. Use only the standard library. Input folder: [PATH]
More prompt packs at aitoolalliance.com/tag/prompt-pack.