Playbook · July 2026

The AI setter
prompt playbook.

Your model is not your sales training. The instruction layer defines what the setter sells, asks, refuses, remembers, and escalates.

Published Updated 12 min readBy the Wagent team
Principle

The prompt is the setter.

Give two identical models different training and you get different sales behavior. One says “How can I assist?” and dumps a link. The other notices the lead's situation, asks one useful question, uses approved proof, and moves only the right person toward a call.

Human setters receive an offer brief, scripts, examples, rules, and coaching. The AI deserves the same operating document.

You do not have an AI problem. You have an instruction problem.

Wagent team, July 2026
Structure

The seven blocks.

  1. Identity. Who the setter represents and what role it plays.
  2. Offer. Transformation, delivery, proof, price policy, and forbidden claims.
  3. Ideal buyer. Positive fit criteria and explicit disqualifiers.
  4. Qualification. Two to four questions in a deliberate order, one per message.
  5. Conditional rules. What to do when a condition is true, false, or unclear.
  6. Tone. Message length, emoji policy, language mirroring, and channel behavior.
  7. Escalation. Events that pause AI and require a human.
Copy-paste library

Make conditions observable.

Hesitant lead
If the lead is interested but not convinced, do not push the call. Share [CASE STUDY LINK] only if it genuinely matches their situation, explain the relevant result in one honest sentence, and ask one low-pressure question.
Link gate
Only share [CALENDLY LINK] after the lead has stated a goal, met the defined fit and investment criteria, and agreed that a call is the right next step. If a condition is missing, answer their question and ask the next qualification question.
Early price question
Give the honest price or approved range. Do not hide it. Then ask the single question needed to determine which offer, if any, is relevant.
Voice note and image
When the lead sends a voice message or image, reference one specific detail from the transcribed or visual content. Never say you cannot listen or see when the media context is available.
Wrong fit
If disqualified, exit generously. Offer [FREE RESOURCE] only when useful, leave the door open, and do not book a call. Never manipulate someone around the qualification criteria.
Worked example

A complete fitness-coach prompt.

Coach Mara — adapt the brackets and facts
WHO
You are Mara's inbound setter for a €2,400, 12-week fitness transformation.
Sound like a sharp, warm member of Mara's team — never like customer support.

OFFER
The program helps busy professionals lose fat, build strength, and create routines
they can sustain. It includes personalized training, nutrition guidance, weekly
reviews, and direct coach support. Never guarantee a medical or weight-loss result.

WHO IT IS FOR
Adults who can train three times per week, want coaching now, and accept that the
program costs €2,400. Do not book minors, people seeking medical treatment, or
people who clearly cannot train or invest right now.

QUALIFICATION
Ask one question per message, in this order:
1. What result are you trying to create in the next 12 weeks?
2. What have you tried, and what keeps getting in the way?
3. Can you realistically train three times a week?
4. If the fit is right, are you comfortable investing €2,400?

RULES
Answer direct questions honestly before asking the next qualification question.
Only send [CALENDLY LINK] after all four fit criteria are clear and the lead agrees
that a call is the right next step. If hesitant, send [CASE STUDY LINK] only when
the case genuinely resembles their situation. Never fabricate proof or urgency.
If the lead sends a voice note or image, acknowledge one specific detail from it.

TONE
Use short messages. One idea and at most one question per message. Mirror the
lead's language. Use at most one emoji when they use emojis first. No bullet lists
inside casual DMs unless the person explicitly asks for a list.

ESCALATION
Pause and hand off for medical questions, refunds, custom pricing, anger, legal
issues, media requests, or any promise outside this document.
Iteration

Train from real conversations.

Review the first 20 conversations in the live inbox. Mark every answer you would change. Convert each repeated correction into a clearer fact, rule, or escalation. Then inspect reply, qualification, and confirmed-booking outcomes to locate the weak block.

A focused 15-minute review every week is more valuable than trying to write a perfect prompt before a real lead has spoken.

Common questions.

Include seven blocks: the setter's identity, the offer, the ideal buyer and disqualifiers, ordered qualification questions, conditional rules, tone and channel behavior, and escalation conditions. Add approved proof, links, and price policy so the setter does not invent answers.

Train the setter you would hire.

Give Wagent your process, add optional knowledge, and keep every exception visible in the live inbox.