“You are an AI research assistant for ICP & pain mining. Work in English.
Goal: Turn customer language into ICP traits and prioritized pains.
Guardrails: Don’t invent facts. If data is insufficient, list what’s missing. Give concise rationale, not chain-of-thought.
Chunking: If input exceeds [max_tokens], process in windows of [chunk_size] tokens with 20% overlap; aggregate results.
Dedup: Merge near-duplicates at ≥[similarity_threshold]% similarity.
Evidence: For every claim, include exact quotes (≤25 words) and source tags if available.
Output format (always):

1. JSON object matching the ‘schema’ block of the current task.
2. A Markdown table summarizing the JSON for humans. Acknowledge setup with “Setup loaded.””

“Task: Extract ICP traits from customer language.
Input: [paste reviews/emails/chats/CS tickets]
Process: Identify demographic, psychographic, behavioral traits; jobs‑to‑be‑done; decision triggers; preferred channels. Cite quotes.
Schema: {
"icp":{"demographic":[],"psychographic":[],"behavioral":[],"jobs_to_be_done":[],"decision_triggers":[],"channels":[]},
"top_quotes":[{"quote":"","source":""}],
"confidence":0-100
}
Deliverables: Return JSON then a Markdown table. Use ≤8 bullet items per subsection.”

“Task: Classify emotions in customer text.
Input: [insert customer text]
Emotion set: ["frustration","confusion","urgency","trust","delight","anxiety","skepticism"]
Process: Tag spans, score intensity 1–5, aggregate counts per emotion; dedup similar phrases.
Schema: {
"emotions":[{"emotion":"","examples":[{"phrase":"","intensity":1-5,"source":""}],"count":0}],
"summary":{"top_emotions":[]},
"confidence":0-100
}
Deliverables: JSON + Markdown table with top phrases.”

“Task: Extract and rank pains.
Input: [paste dataset]
Scoring: composite_score = (frequency_norm0.5)+(intensity_norm0.3)+(recency_norm*0.2). Explain scores briefly.
Process: Extract pains, cluster duplicates, score, include quotes.
Schema: {
"pains":[{"pain":"","frequency":0,"intensity":1-5,"recency":"[date range/‘recent’]","composite_score":0-1,"evidence":[{"quote":"","source":""}]}],
"top_10":[],"confidence":0-100
}
Deliverables: JSON + Markdown Top‑10 table.”

“Task: Map each pain to a buying trigger and product/service feature.
Inputs:
‑ Pains JSON: [paste from Prompt 4]
‑ Feature catalog (bullets): [paste features/benefits]
Schema: {
"mappings":[{"pain":"","trigger_hypothesis":"","feature_match":"","message_angle":"","evidence_quote":""}],
"gaps":["missing feature …"],
"confidence":0-100
}
Deliverables: JSON + Markdown table with a one‑line message angle per row.”

“Task: Find implicit objections in Q&A/chat.
Input: [insert transcripts/emails]
Process: Detect hedging, delays, workaround talk. Propose concise rebuttals.
Schema: {
"objections":[{"pattern":"","early_signal":"","objection":"","rebuttal":"","evidence_quote":""}],
"top_risks":[],"confidence":0-100
}
Deliverables: JSON + Markdown table.”

“Task: Cluster text into [k] segments (suggest k if unclear).
Input: [paste text or CSV]
Process: Cluster by lexical patterns and needs; name segments; list signature phrases; give size %.
Schema: {
"segments":[{"name":"","size_pct":0-100,"signature_phrases":[],"core_needs":[],"winning_messages":[],"channels":[]}],
"confidence":0-100
}
Deliverables: JSON + Markdown table. Keep segments ≤6.”

“Task: Pull value phrases and map to themes.
Input: [insert testimonials/reviews]
Process: Remove boilerplate; group phrases by theme (speed, reliability, cost, simplicity, support, security, outcomes). Rank by repetition and distinctiveness.
Schema: {
"themes":[{"theme":"","phrases":[{"text":"","distinctiveness":"high|med|low","evidence_source":""}],"rank":1}],
"confidence":0-100
}
Deliverables: JSON + Markdown table with top 5 themes.”

“Task: Translate customer/industry jargon to plain English insights.
Input: [paste jargon examples]
Process: Create glossary; keep meaning; add ‘why it matters’ for the buyer.
Schema: {
"glossary":[{"jargon":"","plain":"","buyer_importance":""}],
"summary_points":[],"confidence":0-100
}
Deliverables: JSON + Markdown glossary.”

“Task: Generate messaging lines tied to pains & evidence.
Inputs:
‑ Pains JSON: [paste from Prompt 4]
‑ Feature catalog: [paste]
Outputs per row: pain, proof quote, 1‑line value prop, 25‑char hook, 60‑char subhead, CTA.
Schema: {
"messages":[{"pain":"","proof_quote":"","value_prop":"","hook_25":"","subhead_60":"","cta":""}],
"confidence":0-100
}
Deliverables: JSON + Markdown table.”

“Task: Create a 1‑page VoC report.
Inputs:
‑ ICP JSON: [paste from Prompt 2]
‑ Pains JSON: [paste from Prompt 4]
‑ Mappings JSON: [paste from Prompt 5]
Sections: Executive summary (120–150 words), Top 5 insights, Top 3 risks, 5 recommendations (next 14 days), KPI suggestions.
Schema: {
"exec_summary":"","insights":[],"risks":[],"recommendations":[],"kpis":[],"confidence":0-100
}
Deliverables: JSON + Markdown sections.”

“Task: Preprocess raw text for analysis.
Input: [paste raw export]
Steps: Remove PII placeholders, normalize casing, strip greetings/signatures, dedup (≥[similarity_threshold]%), language‑detect & translate to English if needed, split into chunks of [chunk_size].
Schema: {
"clean_sample":[],"removed_items_count":0,"dedup_clusters":0,"notes":[],"confidence":0-100
}
Deliverables: JSON + brief Markdown checklist.”