AI sales roleplay is a way to rehearse a conversation with a simulated buyer before speaking to a real person. Give the model an approved offer description, a fictional buyer situation and a specific practice goal. Then review the transcript for listening, truthful claims and a clear next decision, not whether the simulated buyer agreed to buy.
The exercises below are designed for a founder practicing discovery, an unsupported-requirement question and a commercial objection. They include an original prompt, three fictional branches, a corrected replay and a copyable review sheet. A simulation is not customer research, proof of demand or a prediction of sales performance.
Choose the practice job before the tool
Practice one difficult moment at a time. You might want to ask a better follow-up question, acknowledge a product limitation without bluffing, or end a conversation with a mutually useful next step. “Make me better at sales” gives the exercise no clear finish.
Specialist coaching products can offer structured scenarios and feedback. For example, Salesforce's Sales Coach teaching separates roleplay from feedback and discusses scenario context and testing. The plain-text exercise here does not claim the same voice, scoring, administration or coaching features.
You can adapt the prompts for Aurora AI in Finta, Claude and Codex where the relevant text interaction is available. Their tools, connected context, models and controls differ. Use a fictional or appropriately redacted brief rather than assume any tool may access private customer material.
Build an honest practice brief
| Input | What to provide | What to leave out |
|---|---|---|
| Offer facts | Current approved capabilities, limitations and any approved commercial facts | Roadmap items described as live or invented customer outcomes |
| Buyer situation | A clearly fictional role, current workflow, constraints and alternatives | A real person's sensitive traits, hidden motives or private data without permission |
| Practice objective | One behavior you can observe in the transcript | A target win rate or automatic deal qualification |
| Scenario boundary | Which objections the simulator may introduce and which facts it must not change | Unlimited invented product claims or false credentials |
| Review standard | Evidence from the transcript and approved fact sheet | A confident score without supporting examples |
Prepare the fact sheet yourself. If a capability or price is unknown, put “unknown, must verify” in the input. The exercise should reward admitting that gap, not a more persuasive invention.
Copy this roleplay prompt
Task
Run a fictional sales practice exercise. Do not contact a customer.
Tool: [Aurora AI in Finta, Claude or Codex]
Inputs
Approved offer facts and limits: [authorized fact sheet]
Fictional buyer role and situation: [scenario]
My practice objective: [one observable behavior]
Allowed objection branch: [choose one branch from this guide]
Roleplay
Play the buyer one turn at a time. Ask one question, wait for my answer and respond naturally without becoming automatically agreeable. Stay within the fictional scenario.
Do not invent facts about the product, customers, certifications, pricing or real market. If a question goes beyond the brief, say the answer is unknown in this exercise.
Review
When I type REVIEW, stop roleplaying. Review only the transcript and approved fact sheet. Identify specific evidence of listening, useful follow-up questions, truthful claims, handling of unknowns and next-step clarity. Separate observation from interpretation.
Flag unsupported statements and propose one corrected response for me to practice again. A person must check the facts before using that response outside the exercise.
Constraints
Do not treat the fictional buyer's reaction as demand evidence or predict a win rate. Do not send messages, access new private records or change any system.Keep the actual fact sheet visible while you practice. If the simulator invents a supporting fact, stop and correct the brief. Do not absorb that invention into your real pitch.
Try three distinct objection branches
All people, products and dialogue in these branches are fictional. Use your own approved offer facts in a real practice session. The example offer is a simple reporting service with file-based input; no live integration or security certification is claimed.
Branch one: the buyer asks for a demo before the problem is clear
Fictional buyer: “Can you just show me the dashboard?” The practice goal is to learn enough context to make the demonstration relevant, without turning the exchange into an interrogation.
A useful response: “Happy to show the relevant part. Before I choose an example, what do you currently do when a report is missing information?” If the buyer answers, follow the actual workflow and ask about a recent case. If they prefer to see the product first, respect that and avoid inventing a discovery requirement.
The exercise ends when you can explain the current process and agree which part, if any, deserves a closer look. A simulated request for a demo is not a qualified opportunity.
Branch two: the buyer asks for a capability you cannot verify
Fictional buyer: “We require a live database integration and a particular security certification. You have those, right?” The goal is to stay accurate under pressure.
A useful response: “The scope I can confirm uses an approved file, not a continuous database connection. I cannot confirm that certification from the information here. Are those mandatory requirements for your evaluation? I can have the responsible person verify the exact position before we continue.”
Do not substitute an unrelated security feature, claim a roadmap date or let the roleplay write an authoritative security answer. A clear not-fit outcome is a successful practice result when it reflects the facts.
Branch three: the buyer challenges value and price
Fictional buyer: “We already do this in a spreadsheet. Why would we pay for another service?” The goal is to investigate the status quo without making up savings or pushing an unauthorized discount.
A useful response: “If the current process works well, changing it may not be worthwhile. Where, if anywhere, does it fail today? We can compare the work and a small test against your existing approach before deciding whether an approved quote makes sense.”
If the buyer says there is no material problem, practice closing respectfully. If they raise a real constraint within the fictional brief, ask one specific follow-up. Do not invent the cost of their time or promise a return on investment.
Worked replay: correct one unsupported claim
This transcript is an original illustration, not a recorded customer call or output from a product test.
| Moment | Fictional transcript | Evidence-based review |
|---|---|---|
| Buyer question | “Can you connect to our production database continuously?” | The approved offer sheet supports file-based input only. |
| First seller attempt | “Yes, our team can automate that for you.” | Unsupported capability and implied delivery promise. Stop the exercise and correct it. |
| Corrected replay | “The scope I can confirm uses an authorized file. Continuous connection is not part of that scope. Is a live connection mandatory, or would a limited file-based evaluation answer a useful question?” | States the verified limit, asks a relevant decision question and does not pretend the gap is solved. |
| Possible next step | “If it is mandatory, let's pause until the responsible technical person can confirm whether there is a suitable option.” | Preserves the buyer's choice and keeps an unknown out of the proposal. |
The correction improves the practice response. It does not demonstrate that a real buyer would accept it. Validate the language and product facts before taking them into an actual conversation.
Review behavior, not simulated persuasion
| Criterion | Observable evidence | Reason to replay |
|---|---|---|
| Listening | The next question follows what the buyer actually said. | The seller returns to a prepared pitch regardless of the answer. |
| Truthful claims | Material statements match the approved fact sheet. | A capability, credential, price or outcome is invented. |
| Useful curiosity | A question resolves a practical uncertainty. | The seller asks leading questions to force agreement. |
| Unknown handling | The gap and responsible verification step are explicit. | Confidence replaces evidence. |
| Next decision | The buyer can accept, decline or clarify one useful step. | The seller treats politeness as consent or commitment. |
Mark each item observed, missing or unsafe and quote the relevant line. This is an editorial practice rubric, not a validated assessment instrument. Avoid turning ten practice conversations into a percentage claim about likely sales performance.
Copy the practice and review record
Practice setup
Fictional scenario: [scenario]
Objection branch: [branch]
Approved fact-sheet version: [version]
Redaction and permission check: [evidence]
One behavior to practice: [behavior]
Transcript review
Repeat for each criterion.
Criterion: [behavior being reviewed]
Transcript evidence: [exact passage]
State: [observed, missing or unsafe]
Correction: [specific change]
Unsupported or changed facts introduced by either participant: [facts]
One response to replay: [response]
Corrected replay: [revised response]
Human fact check: [review and source references]
Before a real conversation
Unknowns outside the simulation: [unknowns]
Facts to verify before use: [verification needed]
Useful question for a real buyer: [question without assuming the answer]When the simulation is too easy or too convincing
If the buyer agrees with everything, strengthen the fictional constraint and ask it to wait for evidence. If the feedback is vague, request the exact transcript line and a single revision. If the model keeps adding offer facts, reduce the brief and restate that missing facts must remain unknown.
For real discovery, use customer interview questions when you are learning about a problem, not selling covertly. For an active account, the relationship-selling playbook connects preparation with respectful follow-through. Keep actual customer statements and fictional practice in separate records.
Sources and disclosure
Prepared by Finta Editorial Team. Primary sources checked October 20, 2026, including GitLab's discovery guidance and Finta's current Aurora description. The prompts, branches, transcript and rubric are original educational materials. No customer interview, specialist coaching benchmark, model comparison or sales outcome is claimed.