Roleplay is the one sales use of AI that beats the human alternative outright. Your manager is busy, your peers go easy on you, and nobody is available at seven in the morning before the call that matters. A model will play a difficult buyer for forty minutes at any hour and never once let you off the hook because it likes you.
What follows is a working set: how to build a buyer who behaves like a real one, prompts for each call type, and, the part most roleplay articles skip, a scoring prompt that grades the transcript afterward, so the practice produces a note you can act on instead of a vague feeling that it went fine.
Setting up the buyer: the six details that make a persona behave
Most roleplay fails at setup. "Pretend to be a CTO" produces a helpful assistant wearing a CTO name tag. Six details fix it.
Give the model the role and company shape, what the buyer is personally measured on, their current mood about vendors in your category, the internal politics around the purchase, one thing they will not volunteer unless asked directly, and their rule for ending the call. That last one matters most: without it, the model stays on the phone forever, which no real buyer does.
Persona setup prompt: "You are [role] at a [size] [industry] company. You are measured on [metric]. You have been pitched by four vendors in this category this quarter and you are tired of it. [Internal politics]. You will not mention [hidden constraint] unless I ask a direct question that surfaces it. If I talk for more than ninety seconds without asking you something, you will say you have a hard stop and end the call. Stay in character. Do not coach me, do not break character to give feedback, and do not be more helpful than the real person would be. I will open."
That final instruction block is the difference between practice and a conversation with an assistant.
Cold call roleplay prompts
The standard cold call: Use the persona setup above, then add: "This is a cold call. You did not expect it. Your first reaction is mild irritation. You will give me fifteen seconds before deciding whether to hang up. If my opener is generic, hang up."
The gatekeeper: "You are an executive assistant screening calls for [role]. You are polite and completely immovable. You will let me through only if I give a specific reason that names something real about the company."
Discovery call roleplay prompts
The vague buyer: "You have a real problem but you describe it in generalities. If I ask an open question, answer it generally. Only give me a specific number or example if I ask a question that requires one. Do not help me get there."
The over-sharer: "You will talk at length about a problem that is not the one your budget is attached to. The real priority is [X], and you will only bring it up if I ask what is on your roadmap this year."
Pair this with the question sets in meeting prep and discovery.
Objection and pushback roleplay prompts
The objection drill: "Play a buyer who raises this objection: [objection, verbatim]. When I respond, do not accept the first answer. Push back twice more, escalating, the way someone would who has heard this response from a vendor before."
The polite no: "You have decided not to buy but you do not want the conversation to be awkward, so you will say things like 'let me discuss internally' and 'send me something'. Never say no directly. Make me work out that the deal is dead."
The polite-no drill is the most useful twenty minutes on this page, because it rehearses the failure that costs reps the most pipeline hygiene. More responses in deal strategy and stakeholder management.
Pricing and negotiation roleplay prompts
The procurement call: "You are in procurement, not the business. You do not care about the product. You are measured on savings and you have a target of [percent]. You will open by asking for a discount before discussing anything else. You have one piece of bargaining power: [what it is]."
The anchor: "You will open with a number forty percent below our list price and treat it as reasonable. If I concede anything without asking for something in return, you will ask for more."
See negotiation, procurement and closing for the preparation work that should happen before this drill.
Multi-stakeholder roleplay prompts
The committee call: "Play three people on one call: [role A] who is enthusiastic, [role B] who is skeptical and technical, and [role C] who is silent for most of the call and then asks the one question that decides it. Label each speaker. Let them interrupt each other."
This one is harder on the model and worth running anyway. Committee calls are where deals are actually lost, and reps almost never rehearse them.
Losing on purpose
Run a drill where the buyer is difficult in one specific way and see whether you notice. Ask for a buyer who agrees with everything and has no budget. Ask for one who is using the call to get a quote to pressure their incumbent. Ask for one who has already decided and wants confirmation. The skill being practiced is diagnosis, not persuasion.
Scoring the rep: the rubric prompt
This is the prompt that turns a roleplay into a coaching artifact.
Scoring prompt: "Break character. Here is the transcript of our roleplay: [paste]. Score me from one to five on each of: opening relevance, ratio of questions to statements, depth of the best question I asked, whether I surfaced the hidden constraint, how I handled the first pushback, and whether I secured a specific next step. For each score, quote the exact line that determined it. Then give me the single change that would have made the largest difference, and the question I should have asked and did not."
Two notes on limits. A text model scores what you said, not how you said it: pace, filler words, tone and confidence are outside what it can see, and the vendor pages in this space are right to say so. And the score is only as good as your criteria, so replace the six dimensions above with your own team's scorecard.
Running this as a team drill
Give every rep the same persona setup and the same scoring prompt, have them run it alone, then compare the transcripts in your next team meeting. The variation between reps on the same buyer is more instructive than any recorded call, because the buyer is held constant. Managers will find the surrounding coaching structure in leadership, coaching and people management and in AI prompts for sales managers.
Where to keep the personas you build
A good persona takes fifteen minutes to write and is worth reusing for a year. Keep them somewhere searchable rather than buried in a chat history you cannot search by situation. Promptifi organizes sales prompts by role, stage and task, so the objection drill you want on a Tuesday afternoon is findable in under a minute. The same personas run on ChatGPT, Claude or Gemini without rewriting.
Related: ChatGPT prompts for objection handling.
Take one for a test run, no account needed: Browse the library →
Frequently asked questions
Can ChatGPT roleplay a sales call?
Yes, and so can Claude and Gemini. The quality depends almost entirely on the setup prompt. Without an explicit instruction to stay in character and not coach you, the model reverts to being helpful within a few turns, which defeats the exercise.
How do I make an AI buyer persona realistic?
Give it six things: the role and company shape, what the buyer is measured on, their mood about your category, the internal politics, one fact they will not volunteer unless asked, and a rule for ending the call. The last one is what stops the model behaving like an infinitely patient assistant.
What is the best prompt for cold call practice?
A persona setup that includes a hang-up rule: fifteen seconds to decide, and a generic opener ends the call. Practising against a buyer who cannot leave teaches you nothing about openings.
Can AI score my sales roleplay?
It can score the transcript against criteria you define, such as question ratio, whether you surfaced the constraint, and whether you secured a next step, and quote the line behind each score. It cannot judge tone, pace or filler words, which need a voice tool or a human listener.
Is Claude or ChatGPT better for sales roleplay?
Claude tends to hold a difficult character longer without softening. ChatGPT is quicker to produce a sharp scoring rubric. Both work; the setup prompt matters more than the choice.
Pro is $12 per month, $89 per year, or $149 for lifetime access. See pricing