Build a system prompt for an AI cold email reply agent
Interviews you about your offer, booking flow, and voice, then generates a complete production system prompt for an AI reply agent — classification taxonomy, response rules, escalation conditions, and a strict JSON output contract an n8n or Zapier workflow can parse. You get the artifact teams pay thousands to have written, tuned to how you actually talk.
You are a conversation-design specialist who has written system prompts for AI reply agents handling thousands of cold email replies a month without one embarrassing send. You know an unattended prompt lives or dies on its output contract: rigid format, explicit escalation rules, zero room for improvisation. Your job: interview me about my business, then write a complete, production-ready system prompt for an AI agent that drafts replies to prospects responding to my cold emails. The system prompt you produce must contain: 1. ROLE AND CONTEXT — who the agent writes as, the company, the offer, and the tone, in 3-4 sentences. 2. REPLY CLASSIFICATION — categories (interested, question, objection, not now, wrong person, unsubscribe, out of office) with a one-line handling rule each. 3. RESPONSE RULES — length caps, banned phrases, when to include the booking link, how pricing questions are handled. 4. ESCALATION RULES — exact conditions where the agent must NOT reply and must flag a human: legal threats, angry replies, existing customers, multi-question emails, anything ambiguous. 5. OUTPUT CONTRACT — the agent returns only a JSON object with fields for classification, draft_reply, confidence, and escalate, so n8n or Zapier can parse it. No prose outside the JSON, no placeholder text ever inside drafts, and unsubscribe requests always classified for suppression, never argued with. Interview me ONE AT A TIME, waiting for my answer each time: 1. What do you sell, to whom, and what is the one outcome you lead with? 2. What does booking look like — a calendar link, a router, or human scheduling? 3. How should pricing questions be handled: deflect to a call, give a range, or share exact numbers? 4. Which replies must always reach a human untouched? 5. Paste one reply you personally sent recently that sounds exactly like you. Then output the full system prompt in a single code block, ready to paste into an n8n AI node, followed by three test replies I should run through it before going live.
How to use it
- 1
Copy the prompt into Claude, ChatGPT, or any LLM.
- 2
Answer the five questions; question 5 matters most — a real reply you wrote teaches the agent your voice better than any adjective list.
- 3
Paste the generated system prompt into your n8n AI node (or wherever your reply automation calls a model).
- 4
Run the three included test replies plus 10 real historical replies through it, and tighten escalation rules for anything it got wrong.
- 5
Rerun this builder whenever your offer or pricing posture changes — stale reply agents quote old positioning.
Best practices
Keep the agent drafting, not sending, for the first two weeks — review drafts in a Slack channel until confidence is earned.
Be ruthless on question 4; the cost of a human reviewing five extra replies a day is nothing next to one bad reply to an angry prospect.
Test the JSON contract deliberately: feed it a weird reply and confirm the output still parses before wiring it to a send step.
Log every escalate=true reply and review weekly — the pattern tells you what rule to add next.
Example: what this looks like in practice
A founder running 8,000 cold emails a month for her compliance-software startup answers the interview: they sell SOC 2 automation to seed-stage CTOs, booking runs through a Cal.com link, pricing questions get a range plus a call nudge, and anything mentioning an existing customer or legal language goes to her. She pastes a real reply she sent. The builder outputs a 700-word system prompt with seven classifications and a JSON contract. Wired into an n8n workflow with Smartlead's reply webhook, the agent drafts replies into a Slack approval channel. In week one it handles 61 of 74 replies cleanly, escalates 13, and her average reply time drops from 9 hours to 40 minutes — which converts two extra meetings from the same reply volume.
Best fit
This prompt is one gear in a bigger machine. We orchestrate 20+ tools into outbound systems our clients own — and guarantee the results.
Apply for a Pilot Spot → →Frequently asked questions
Yes, and at meaningful volume it is becoming standard practice — but the safe pattern is draft-first, not autonomous sending. A well-prompted agent classifies each reply, drafts a response in your voice, and escalates anything sensitive. Teams typically review drafts for a couple of weeks, then let clean categories like scheduling logistics send automatically while objections stay human-approved.
More automation & workflow design prompts
Turn a manual sales process into an automation blueprint
Converts a fuzzy 'we do this by hand every week' description into a structured blueprint with a detectable trigger, labeled steps, edge cases, human checkpoints, and failure alerts. It forces the design decisions people usually discover mid-build, so what you hand to n8n, Zapier, or an ops hire is assembly work, not archaeology.
Architect a Clay table before you burn the credits
Designs the Clay table before you build it: sources, column order, run conditions, waterfalls, outputs, and a credit estimate. Column ordering and gating are where most teams silently overspend, so the prompt enforces filter-first architecture and makes every paid column defend its run condition — typically cutting credit burn 40-70% versus enrich-everything builds.
Audit your full sales tool stack for overlap, gaps, and waste
Turns your tool list into a capability matrix that exposes overlaps, gaps, and zombie subscriptions, then delivers cut/keep/consolidate verdicts with savings math and safe migration notes. Teams typically find 20-35% of GTM tool spend is redundant — this makes the case with a matrix, not a hunch.