Do you need to be the smartest prompt engineer in the room to build an AI role-play that feels real? No. You need a clear structure and the right context, and the AI does the rest.
This guide is about the writing, not the buttons. If you want the click-by-click of the editor, start with how to create a role-play. Everything below lives on the Persona and Scenario tabs.
Start with the Persona
Name
Give the AI customer a name. Any style works: "Jack", "Jordan Lee", "Dr Lisa Smith".
Job title
For B2B, the title carries a lot. A VP of Sales doesn't talk like a CEO or a RevOps lead. Titles change vocabulary, expectations, and how the AI pushes back.
For B2C, "Customer" is usually fine, unless a specific role would change the conversation: "Teacher", "Accountant", "Parent of two teenagers on a budget".
Customer company and link
B2B: add the customer's company, real or fictional, and a link. If it's a real account and you plan to share the scenario, use their actual site, ideally the page they'd have open during the call. For sensitive or generic training, a fictional company works and you can skip the link.
B2C: skip both.
Difficulty
Think of difficulty as seasoning.
Medium is where to start. Well balanced, challenging, conversational, and it won't collapse into a loop of objections.
Hard adds teeth. More pushback, and a higher bar on how reps handle objections and hold the structure.
Insane is a sparring partner who never tires of "Yeah, but...". Expect sweat.
Conversation Setup and Context
First, pick the category:
- Choose one: General, Cold Call, or Discovery Call.
- If you’re unsure, choose General Conversation; you’ll still shape everything with Conversation Steps.
Role-play context
This block is the AI's primary source of truth. Write it well and the AI thinks, speaks and decides like your ICP. Write it vaguely and the AI falls back on general knowledge of your industry, which is where the generic-sounding buyer comes from.
For B2B, work through:
- What the prospect is looking for, or what triggered this meeting
- Current state: tools, processes, who's involved
- Top challenges and pains
- Motivations and goals: targets, deadlines, personal stakes
- What's unusual about the organisation: constraints, compliance, market dynamics
- What's specific about this meeting: a referral, competitive pressure, timing
For B2C:
- What triggered this conversation
- Current setup, and who else is involved if it matters
- Top frustrations
- Goals and preferences: time, budget, health, style
Use light numbers wherever you can. Specifics sharpen the AI more than adjectives do.
Conversation steps
Write steps as the shape of the dialogue, not a script. Brief, and pointed at an outcome.
Discovery call, for example:
- Open briefly, set an agenda and success criteria.
- Map the current state, widen, then narrow into pains.
- Quantify the impact.
- Map the buying process.
- Book the next step with a specific date and attendees.
Advanced mode lets you add conditions and build custom flows. 
A few that earn their keep:
- If the rep skips the agenda, the buyer interjects after the first discovery question: "Before we dive in, when do you plan to show me the product?"
- If the rep doesn't quantify impact, the buyer leans status quo: "We're fine as we are."
- If the rep proposes next steps with no date or attendees, the buyer defers: "Just email me."
Steps vs Scorecard—how they work together
Conversation steps drive what the AI decides during the role-play, including whether it accepts a next step. The scorecard judges the recording afterwards.
So if you want the AI to react in real time to a buying signal or agree to a follow-up, that logic belongs in conversation steps and their conditions, not in the scorecard.
Objections
Don't load 15 objections into one scenario. Real buyers don't work that way. A price-sensitive buyer keeps returning to price. They don't pivot from budget to integration risk to procurement in the space of ten minutes.
Pick one storyline per scenario - budget, status quo, internal politics, a competitor feature - and build three or four objections around it. Reps get to actually master a theme.
If you want 10 to 20 mixed objections, that's a separate rapid fire exercise. Keep your main role-plays focused.
Advanced mode lets you add three elements per objection, and it's worth the extra minutes:
Trigger: bring this up after value is discussed, before pricing is shared.
Handling guidance: the rep should acknowledge the constraint, diagnose current spend across tools, hours and discount leakage, build a simple ROI from the buyer's own numbers, offer phased options, then confirm fit.
Consequence: if the rep mishandles it, the buyer asks for details by email instead of agreeing to a meeting.
At minimum, write a trigger and handling guidance.
Conversation starter
The AI buyer speaks first. The AI now writes this opening line for you from the rest of your setup, so your job is to check it sounds like what your reps actually hear when they pick up.
Outbound: "Jordan here, who's this?"
Inbound or a known contact: "Thanks for making time, what's the plan for today?"
If your reps normally open the call themselves, give the buyer a neutral line and let the rep take it from there.
One of our partners put it well: your first scenario feels like a C-. Tweak it once and it's a solid B. A few iterations in, you're building A-class role-plays without breaking a sweat.
Draft, test, iterate. Ask a couple of teammates to run it before you launch it to the team.
Short on time? Build from a link, a file, or a real recorded call and PitchMonster drafts the whole thing in a minute. Then edit the parts that don't sound like your buyer.
Questions? Reach out to your Customer Success Manager. We're always there for you.