Most AI cold email comes out sounding like everyone else’s AI cold email. The fix is to feed it your own sales calls instead of a template. I run this loop every Monday with last week’s discovery transcripts before touching the outbound queue.
Three prompts, each one handing back a finished asset you can drop into the pipeline that day.
[Claude]
[Claude] Here are 3 discovery call transcripts from last week. Extract the exact phrases prospects used to describe their pain, the objections that killed momentum, and the words they used for success. Output a table: Pain Phrase | Objection | Success Language | Which Deal. Then write 5 subject lines built only from that vocabulary. No invented metaphors.
[ChatGPT]
[ChatGPT] Using the pain phrases and objections below, rewrite my 4-email cold sequence for manufacturing ops managers. Each email under 90 words, one CTA, opens with the prospect’s own language not mine. Output as a plain text block I can paste into HubSpot, with a one-line note after each email explaining what changed and why.
[Cursor]
[Cursor] Build me a script that reads call transcripts from ./transcripts (txt files), extracts lines containing buying signals using these regex patterns, and writes a CSV sorted by deal size with columns: File, Signal Type, Quote, Timestamp. Include a –dry-run flag and print a summary count per signal type when done.
Full archive at aitoolalliance.com/tag/prompt-pack.