The AI gives you a first draft. This article covers reviewing it, writing a buyer who behaves realistically, setting how reps are scored, and publishing.
ℹ️ The editor has its own guidance panels beside each field. This article adds the decisions those panels do not cover.
1. How the editor is organized
The sections appear in this order. Work down the page.
| # | Section | What it controls |
|---|---|---|
| 1 | Settings | The role play title, up to 70 characters |
| 2 | Personalize Your Simulation | What the buyer knows and how they behave |
| 3 | Presentation Settings | A deck, if reps are practicing with slides |
| 4 | AI Persona Setup | Who the rep is talking to |
| 5 | Conversation Opening | How every session starts |
| 6 | Topics to Cover | How reps are scored |
| 7 | Language Settings | Whether reps choose their practice language |
| 8 | On-Screen Action Analysis | Scoring on-screen actions from a demo video |
| 9 | Score Settings | The passing score and scoring weights |
| 10 | Advanced Settings | Time limits and session behavior |
2. Personalize Your Simulation
This section does more for realism than anything else. It has 3 fields.
| Field | What goes in it |
|---|---|
| The Conversation Context | The situation the buyer is walking into. Where the conversation happens, who is involved, the stage of the deal, and why it is happening |
| The Persona's Background Information | Who they are. Role, company, current situation, personality, communication style |
| Persona Concerns | 3 to 5 hesitations the rep has to work through. No more than 5 |
ℹ️ The buyer can share anything written in the background field. If you want something held back until the rep earns it, say so there. For example, "does not mention the failed rollout unless the trainee asks what has been tried before".
Five rules make the difference between a buyer who feels real and one who does not.
| Rule | Why |
|---|---|
| Keep the AI in the dark | Describe the situation only. Never include the rep's goals or plan, or the buyer will anticipate their moves |
| Write narratively, not as rules | "Jordan has little patience for generic pitches and probes for specifics" beats "Jordan will push back on anything vague". Try "[Persona] tends to [behavior] because [reason]" |
| Frame concerns as internal worries | "Jordan had a frustrating experience with a rigid contract and is cautious about long commitments" beats "Jordan will ask about contract length" |
| Tie concerns to rep actions | "When pricing is mentioned" fires too early. "If pricing comes up before the trainee has established value" fires at the right moment |
| Say "trainee" when the rep must act | "Jordan won't discuss budget unless a business case is presented by the trainee". Without it, the AI may think it is the one who should act |
💡 Pro Tip: Use the <persona name> token instead of typing a name. The text then still works if you swap the persona.
3. Example prompts
Conversation Context
Good: You are on a scheduled discovery call with a business stakeholder who agreed to speak after an initial introduction. The call was booked for 30 minutes.
Avoid: You are on a discovery call. The trainee will ask about your current process, quantify the pain, and try to book a demo.
The second hands the buyer the rep's plan.
Persona Background Information
Good: <persona name> is the Director of Operations at a regional logistics company. They own day-to-day execution across several sites. They are direct, practical and hard to impress. Numbers matter more than polished language. Past vendor conversations drifted into long feature tours, so <persona name> now listens for proof that a conversation is based on their actual workflow.
Avoid: <persona name> will push back on anything vague and will not answer budget questions.
Persona Concerns
Use the same 4 part shape each time. [Concern name]: <persona name> [the worry, in their terms]. When it comes up: [what the trainee says or does that triggers it] How to frame it: [tone and angle] Good enough to proceed when: [what the trainee must say or do]
Filled in: Generic value claims: <persona name> gets skeptical when a conversation relies on broad promises instead of concrete evidence. When it comes up: When the trainee stays high level or leads with features. How to frame it: Direct and brief, anchored to numbers or workflow facts. Good enough to proceed when: The trainee gives 1 specific example or metric tied to the problem.
4. AI Persona Setup
Choose who the rep will be talking to.
| Tab | What it holds |
|---|---|
| System Personas | Ready-made personas provided with the product |
| Custom Personas | Personas your team has built, showing who created each one |
| New Persona | Build your own |
Click Edit on a persona to change its name, image, voice and personality. Voices are set by gender, age and accent, and you can adjust the speaking rate and preview the voice before saving.
| Rule | Detail |
|---|---|
| System personas | Cannot be changed or deleted. Duplicate one to make your own version |
| Custom personas | You can only delete personas you created, and only if no live role play uses them |
💡 Pro Tip: Preparing reps for a specific upcoming call? Match the persona's name, role and company to the real buyer.
5. Conversation Opening
Two fields control how every session starts. Both cap at 500 characters and both accept the <trainee name> and <persona name> tokens.
| Field | What it is | Example |
|---|---|---|
| AI Trainer Opener | What the rep hears before the conversation begins. It is not part of the conversation, so it can include the rep's goal for the call | You're about to join a discovery call with the VP of Revenue Operations. They attended your webinar. Your goal is to understand how their team builds its forecast today |
| AI Persona Opener | The first thing the buyer says. Keep it to 1 or 2 sentences | Scheduled call: "Thanks for making time. I've got about thirty minutes." Cold call: "Hello? Who is this?" |
⚠️ Do not reveal the buyer's concerns in the opener. Reps should have to uncover them.
ℹ️ The AI trainer's name can be changed.
6. Topics to Cover
Topics are how reps are scored. Each topic holds evaluation guidelines, which are the specific things a rep must do to get credit. Guidelines are generated when you add a topic. You can edit them, or use Regenerate on a guideline cell for alternatives.
ℹ️ Add your parameters in Personalize Your Simulation first. Topics are built from them.
Set your topics
- Click Edit on Topics to Cover to open the topics table.
- Use Add topic to add one, or Custom to type your own.
- Hover over the last column of a row to Duplicate or Delete a topic.
- Click Done.
⚠️ Every topic must start with this exact line, or grading does not work properly: To successfully cover the topic, the trainee needs to: [xyz].
Write guidelines that score consistently
The AI reads the transcript only. It cannot judge tone, confidence or body language. Ask yourself whether someone could highlight the exact line that meets the guideline. If not, rewrite it.
| Too vague | Can be scored |
|---|---|
| Show empathy | Acknowledge the customer's concern before responding |
| Build rapport | Ask 1 question about the customer's current situation before explaining the purpose of the call |
| Sound confident | Introduce themselves by name, company and purpose in the opening |
| Active listening | Reference at least 1 challenge or goal the customer mentioned earlier |
| Uncover the budget concern | Ask whether budget is a factor in the decision |
Three habits that make the difference:
| Habit | Why |
|---|---|
| Start every guideline with a strong action verb | Ask, Explain, Confirm, Summarize, State, Describe. Weak verbs like "try to", "avoid" and "ensure" cannot be graded |
| Grade the rep's action, not the buyer's reaction | "Ask whether budget is a factor" works. "Make the customer feel heard" does not |
| Keep each guideline in 1 topic only | If the same behavior appears in 2 topics, the AI cannot score them independently |
Limits and structure
The AI scores more reliably inside these limits.
| Rule | Detail |
|---|---|
| Guidelines per topic | 3 to 6 is the target. 7 is the maximum. Over 7, split the topic |
| Minimum | 2 guidelines, except for the compliance case below |
| Topics per role play | Up to 10 |
| Weights | Equal by default. You can change them, but they must total 100 |
| Every guideline counts | None are optional or bonus. A perfect score means covering all of them |
Split a topic when it covers more than 1 skill or phase of the conversation, when guidelines need different kinds of evidence, or when 2 guidelines overlap. The quickest way is to duplicate the topic, then delete the guidelines you do not want from each copy.
Make or Break
Missing a Make or Break topic sets the entire score to 0. You can mark more than 1.
Reserve it for binary, non-negotiable statements, such as telling the customer the call is being recorded. This is the one case where a single-guideline topic is the right structure.
Give reps a choice where more than 1 answer is valid
Every topic is mandatory, but you can build flexibility inside one.
To successfully cover the topic, the trainee needs to:
- Introduce the product by name and explain its primary purpose.
- Mention any 2 of the following 5 features: [list them]
- Confirm whether the customer has any questions before moving on.
Use this when the skill is consistent but the exact content is not. Do not use it to make a topic effectively optional, and do not list so many options that the threshold stops meaning anything.
Make 1 guideline matter more than the others
Two ways:
| Option | How |
|---|---|
| Split it into its own topic | The most reliable. It gets an independent score and you can set its weight |
| Add a scoring cap | Below the guidelines, add: "The following guideline is critical for this topic: [restate it]. If this guideline is not clearly demonstrated in the transcript, cap the score at 30%." |
Use the cap sparingly. Only when skipping that guideline is a true failure, not just an incomplete performance.
💡 Pro Tip: Control difficulty with detail, not structure. "Mention that the solution improves efficiency" is beginner level. "State that the solution improved efficiency by 20%, referencing the three-month study" is advanced. The AI can score both. Only the depth changes.
Before you publish, check the rubric
Run through these 7 checks. They catch the problems that are hardest to spot once reps are using the role play.
| # | Check |
|---|---|
| 1 | Every topic starts with the required preface line |
| 2 | Every guideline uses a strong action verb |
| 3 | Every guideline describes something said out loud, not a tone or attitude |
| 4 | 3 to 6 guidelines per topic, 7 maximum |
| 5 | No guideline appears in more than 1 topic |
| 6 | Acronyms and framework names are spelled out in full |
| 7 | Nothing is graded on whether the buyer agreed or reacted well |
⚠️ Uploaded files and selected Spekit content are not used to score reps. Neither are example videos or helpful links attached to a topic, which are there for the rep's own learning. If a behavior is not written into the evaluation guidelines, it is not scored.
7. Score Settings
These settings decide what counts as a pass and what reps see afterwards.
| Setting | What it does | Default |
|---|---|---|
| Passing Score | The minimum score out of 100 a rep needs. The Learning Path step completes only when they pass | 80 |
| AI Scoring Weights | How much comes from knowledge, what they said, versus style, how they said it | Knowledge 70%, Style 30% |
| Enable Practice Mode for learners | Lets reps take an unscored practice attempt with full feedback before a scored one | Off |
| Max Hints Per Session | How many hints a rep can use in a practice attempt. Appears only when Practice Mode is on | 5 hints |
| Hide score | Hides the score from reps. Managers and admins still see it | Off |
💡 Pro Tip: Because every guideline is required, lower the passing score if reps should be able to pass without covering everything perfectly. Practice Mode is a good way to give new hires a low pressure first try.
When Practice Mode is on, reps choose between a practice attempt and a scored attempt each time they start. Practice attempts give full feedback with no number, can be repeated freely, and are excluded from averages and completion.
Turning Practice Mode on also turns on hints, and adds 2 hint settings below it.
| Setting | What it does |
|---|---|
| Hint visibility duration | How long each hint stays on screen. Fixed at 30 seconds |
| Max Hints Per Session | How many hints a rep can use in 1 practice attempt. 5 by default, and you can change it |
ℹ️ Hints are generated by the AI from the live conversation and your scoring criteria. You cannot write or edit them. They appear in practice attempts only, never in a scored attempt.
⚠️ A practice attempt does not produce a score, so it cannot complete the Learning Path step. Reps still have to take a scored attempt and pass.
8. Other settings
Four more sections, each optional. Most role plays only need the language and time limit settings.
| Section | What it offers |
|---|---|
| Presentation Settings | Upload a PDF or PPTX, up to 100MB, for reps practicing with slides. You can also let reps upload their own. Make sure the deck is accessible |
| Language Settings | A toggle to let reps choose their practice language. 30+ languages are supported, and the conversation, analysis and email notifications all run in the chosen language. If you build in a language other than English, write everything in that language, including the persona background, concerns and guidelines |
| On-Screen Action Analysis | Upload a video showing how the AI should score on-screen actions during the role play |
| Advanced Settings | Allow the AI to end the role play, with instructions on when. Session time limit, 1 to 59 minutes, 20 by default. Short session penalty. Screen sharing. AI persona subtitles |
💡 Pro Tip: The session time limit is a real constraint for reps. 20 minutes is tight for a full discovery conversation. Set it to match the call you are simulating.
9. Edit with AI
Edit with AI opens an assistant panel. Describe the change in plain language.
| Goal | What to type |
|---|---|
| Harder buyer | Make the persona more price sensitive and slower to share numbers |
| New objection | Add a concern about implementation time, triggered when the trainee discusses rollout |
| New information | Update the persona background so their company has just merged with a competitor |
| Fix the opening | Shorten the persona opener and remove the hint about their budget concern |
⚠️ The buyer only knows what is written in the role play. Updating a Spek, adding files to the Learning Path, or releasing new product details does not update an existing role play. Change it by hand or with Edit with AI.
10. Save, preview and publish
Do these in order. The order matters.
- Save. Your role play is stored and appears in the Learning Path marked Draft.
- Preview. The session opens in Testing mode so you can run it yourself.
- Publish. The role play becomes visible to reps.
⚠️ Save before you preview. If you preview unsaved work, ending that session can lose it.
ℹ️ A role play left as Draft is hidden from reps and does not count towards completion. If reps cannot see your role play, check that it is published and that the Learning Path itself is published and assigned.
Test it properly before you assign it
Run it twice. Once doing it well, once doing it badly on purpose.
| What happened | What to do |
|---|---|
| The good attempt failed | Loosen the guidelines or lower the passing score. A bar strong reps cannot clear makes people stop trusting the exercise |
| The bad attempt passed | Tighten the guidelines or make the buyer more challenging |
Then pilot with a few reps, including 1 strong and 1 new, before assigning it to the full team.