You can build a sales roleplay partner in ChatGPT tonight, for free or for twenty dollars a month, with no code and no vendor call. This tutorial walks through exactly how.
TL;DR
- Yes, you can build it today: you can build a working AI buyer persona in ChatGPT this afternoon using the Custom GPT builder or Advanced Voice Mode.
- What you actually get: one persona, one conversation at a time, with no scorecard and no way for a manager to assign it or see who finished it.
- Fine for solo prep, not a program: that's enough for a rep prepping solo before a call, but it isn't a training program.
- Honest verdict: closing that gap, turning a solo trick into an assignable, trackable program, is the specific thing Outdoo AI is built to do.
At a glance:
| Outdoo AI | DIY build (ChatGPT) | |
|---|---|---|
| Persona creation | No-code, from a prompt, template, call transcript, or LinkedIn profile | Write the Instructions field yourself, by hand |
| Multi-stakeholder calls | Up to three AI stakeholders in a single call | One persona per conversation |
| Scoring | Scorecards graded against a set rubric, built from your own playbook | Ask for feedback, get an inconsistent answer |
| Manager visibility / assignment | Courses, batches, and call-blitz assignment with dashboards per rep | Workspace sharing (Team/Enterprise) makes a GPT visible, not assignable |
| CRM / LMS integration | Built-in CRM, LMS, and Slack integration, plus SCORM export | None out of the box; Actions require custom API work |
(The full breakdown of each gap is below, after the build.)
Build the Persona in the GPT Builder
Click your name in the top corner of ChatGPT, then My GPTs, then Create a GPT.
You land on two tabs: Create, a conversational builder that fills in fields for you as you describe what you want, and Configure, the plain form underneath it.
Use Configure. You want direct control over each field, not a summarized guess at what you meant.
Start with Name and Description, something like "Skeptical VP of Ops, mid-market SaaS deal." That's just identity.
The field that actually does the work is Instructions, which functions as a system prompt run ahead of every message the persona sends.
Write the character properly: title, company size, the specific objections they raise and the order they raise them in, how much patience they have before they check out.
Then write down what the persona should never do. Say it plainly: don't break character, don't coach the team member mid-conversation, don't offer feedback until they type "end roleplay."
Skip that last line and the model will cheerfully pause the scene to explain what you should have said instead, which defeats the point of a roleplay.
Below Instructions sit Conversation Starters, canned openers so a team member isn't staring at a blank box wondering how to begin, and Knowledge, where you can upload up to twenty files (PDF, Word, CSV, and a handful of other formats) so the persona argues from your actual pricing sheet and competitor battlecard instead of a generic one.
Then Capabilities: checkboxes for Web Browsing, Code Interpreter, Canvas, and Image Generation. Leave all of them off for a roleplay persona. You don't want it browsing the open web mid-scene instead of staying in character.
Test the persona live in the Preview pane on the right, and rewrite Instructions until the objections actually sound like your buyer, not a generic skeptic.
When it's ready, hit Publish. You get three visibility choices: Only me, Anyone with the link, or Everyone on the GPT Store.
On a Team or Enterprise plan, there's a fourth option to share it inside your company workspace, either to everyone or to specific people you pick.
Or Skip the GPT Entirely
You don't need to build anything to start practicing tonight. Open Advanced Voice Mode and say out loud, "play a skeptical VP of Operations pushing back on our pricing," and go.
It responds in real time, it interrupts naturally, and it's genuinely useful for rehearsing tone and pacing, not just word choice. For a rep with twenty minutes before a call, this is the fastest path there is.
Where This Breaks Down for a Team
None of the above is a knock on ChatGPT. It's what happens when you run a training program out of a tool built for one person having one conversation.
Start with the call itself. You can't put a skeptical CFO and an eager end-user in the same conversation arguing with each other and with the rep, which is closer to how real committee-based B2B deals actually unfold.
Ask the GPT for feedback afterward and it will give you something, but no rubric gets applied the same way call after call, rep after rep. The grade depends on how you happened to phrase the question that day.
Workspace sharing on Team and Enterprise plans is a genuine feature, worth using if you're building GPTs at all, and it does make a persona visible to everyone in your org.
What it can't do is let a manager say "practice this objection set by Friday" and then check who actually did it.
The CRM and LMS gap works the same way: Actions can call an external API if you write an OpenAPI schema for it, but wiring that into Salesforce or your LMS turns into custom engineering per company.
Even the analytics OpenAI does offer sit at the wrong altitude for this. The Global Admin Console gives Enterprise customers credit consumption and adoption by seat.
It has nothing to say about whether a specific rep is getting better at handling a pricing objection over the last quarter.
And there's no built-in way to drop a live practice agent into your own CRM or internal portal, no purpose-built layer for the compliance requirements around sensitive sales conversations and coaching data.
Where Outdoo AI Picks Up
Outdoo AI starts at the point this tutorial stops, and the difference comes down to what the scoring is actually grounded in.
A GPT's feedback is whatever the model improvises in the moment.
Outdoo AI's scorecards are built from your team's own playbook, call transcripts, and battlecards, so a rep is graded against how your company actually sells, not a generic idea of a good call.
That grading also runs on both sides of the call.
The same scorecard grades roleplay practice and live customer calls, so a sales leader can point to a coaching change and trace it to a real shift in how calls go, not just a completion count.
Roleplays support up to three AI stakeholders at once, so a rep can practice a skeptical economic buyer and an enthusiastic champion pushing back on each other in the same session.
Managers assign practice through courses, batches, and call-blitz campaigns, and see who finished what on a dashboard, which is the exact piece "share a GPT with the workspace" can't do.
It connects to the CRM and LMS your team already runs, including SCORM export for existing training systems, and personas can be created without writing a single prompt by hand, straight from a call transcript, a template, or even a LinkedIn profile.
Embedding a live practice agent directly inside Salesforce takes the same idea one step further: the practice lives where the work already happens.
Build the GPT tonight if you want to feel how this works. Reach for something built for a team once you need to run it as a program.
See how Outdoo AI grades practice and real calls on one scorecard.
Frequently Asked Questions
.svg)
Use the Custom GPT builder's Configure tab (My GPTs > Create a GPT), not the conversational Create tab, and fill in a Name, Description, and an Instructions field that defines the buyer's title, company size, objections, and behavior rules such as not breaking character. Knowledge files (up to twenty) and Conversation Starters are optional additions, and Capabilities like Web Browsing, Code Interpreter, Canvas, and Image Generation should be left unchecked.
No. A Custom GPT handles one persona per conversation, so a scenario with a skeptical CFO and an eager end-user arguing with the rep at the same time isn't possible within a single GPT conversation.
It can be built for free or on the $20-a-month ChatGPT Plus plan, since both the Custom GPT builder and Advanced Voice Mode are available on those tiers.
Not directly. Team and Enterprise plans let a GPT be shared with the whole workspace or with specific people, which makes it visible to the team, but there's no way to assign it with a deadline or see who actually completed it.
Yes. Outdoo AI roleplays support up to three AI stakeholders in a single call, compared to the one persona per conversation that a Custom GPT is limited to.
Table of Contents
Talk to Sales
Have questions about training and enablement for your sales, CS, support, or leadership team? Let's talk.
Talk to Sales
