Each prompt below is designed to feed into Claude or ChatGPT. Run them in order — outputs of earlier prompts feed into later ones. Don't skip the synthesis prompts — that's where the work actually happens.

1. Customise the script. "My M1 positioning is: [paste]. My ICP guess is: [paste]. Customise the 12 problem-discovery questions below for this exact buyer. For each question, add a 'watch for' note — what answer would be a real pull signal vs a polite signal: [paste 12-question script]."
2. Build the target list. "I need 15 problem-discovery interviews with [persona]. Generate three sourcing strategies that don't rely on my warm network: cold LinkedIn outreach, niche communities, and proxy events. For each, give me the exact search terms or platforms, and a 400-character outreach DM I can adapt."
3. Write the outreach DM. "Here's my positioning [paste] and the target persona [paste]. Write three LinkedIn DM variants, each under 400 characters, asking for 20 minutes of their time. No selling. No 'pick your brain'. Frame me as someone studying the problem, not pitching the solution. Three angles: peer curiosity, research framing, problem-shared framing."
4. Build the objection bank. "Given my positioning [paste], list the 10 most likely pushbacks I'll hear in discovery calls. For each, give my honest response — not the rehearsed one. If the honest response is weak, flag that as a real problem with my offer, not an interview problem."
5. Live follow-up coach. "I'm 20 minutes into an interview. Here's the transcript so far [paste chunk]. Give me three follow-up questions that dig into past behaviour, not future intention. Specifically chase any moment where they mentioned a workaround they've already paid for."
6. Bias-check this interview. "Here's a full interview transcript [paste]. Flag every question where I led the witness. Flag every moment I sold instead of listened. Flag every time I let a polite signal pass without checking whether it was real pull. Be specific — quote the line."
7. Extract pull signals. "Below are 3+ interview transcripts [paste]. Extract: (a) verbatim pain phrases — the exact words they used, in quotes; (b) workarounds they've already paid for, with the amount if mentioned; (c) willingness-to-pay signals — anything that hints at price, frequency, or budget shape. Rank each list by how often it recurred across interviews."
8. Cluster the personas. "Here are 15 per-interview synthesis sheets [paste]. Find 3 emergent ICP segments. Score each segment across 5 dimensions: pain intensity (1-5), pain frequency (1-5), current spend on workarounds ($), willingness to pay ($), accessibility for me to reach (1-5). Be brutal — if one of the three segments is weak, say so, don't pad."
9. Find the disconfirming evidence. "Here's my synthesis output [paste]. I'm about to write a go verdict. Show me the 5 strongest pieces of evidence in this data that I'm ignoring because they hurt the thesis. Quote verbatim. If the disconfirming evidence is stronger than the confirming evidence, tell me to pivot, not go."
10. Stress-test the verdict. "Here's my draft go/pivot/kill verdict [paste]. Steel-man each of the other two options. For 'go': what's the most expensive way this is wrong? For 'pivot': what's the version of my offer the data actually supports? For 'kill': what would I rather hear instead, and is that wishful thinking? Three paragraphs. Don't soften."