The roleplay and feedback scenes are the core of a Roleplay Session. The roleplay scene defines the conversation your learner practices. The feedback scene closes the loop with an automated score and coaching summary. This article explains each configuration option and how to use it to build a Roleplay Session that meets your training goal.
How to configure the roleplay scene
To configure the roleplay scene:
Select Roleplay as the scene type.
Choose your Activity Type: Cold call, Field sales, Manager feedback, Customer conversation, Discovery call, or Media training.
Activity Type
Fill in each field described below.
AI agent persona
The persona is the person your learner speaks to. Choose or define a persona that matches the scenario, such as a customer, prospect, manager, teammate, or direct report.
Define these fields:
Name — The persona's name.
Base personality — The persona's demeanor.
Job title — The persona's role.
Company — Where the persona works.
Background — Stable facts about who the persona is that hold true in any conversation.
Think of Background as the persona's character sheet — who they are before this specific situation. The more grounded and specific it is, the more realistic the conversation feels. Background can include how long the persona has been in their role, how they feel about feedback or conflict, and relevant personal context they would carry into the conversation.
✍️ Keep the situation out of Background. What is happening right now belongs in Instructions, not Background. If you use AI to draft the persona, check each sentence afterward — if it describes what is happening right now, move it to Instructions.
AI agent persona fields
Instructions
Instructions describe the specific situation the learner walks into — what is happening right now in this conversation. The AI uses this to stay in character. The learner does not see this text, so they have to work out the situation through the conversation itself.
Instructions field
Write 3–5 sentences, as if you are briefing an actor before they go on stage. Include:
What prompted the conversation
What the persona knows
What the persona does not know
How the persona feels at the start
Any conditional behavior you want the AI to follow
For example:
"You are a prospect who requested a demo after researching Synthesia. You are interested, but you are worried about cost and whether your team will actually use the product. You have 20 minutes before your next meeting, so you want clear answers quickly. If the learner asks thoughtful discovery questions, share more detail about your team's training goals."
💡 Instructions have the biggest impact on roleplay quality. If the conversation is not going the way you want, this is the first field to adjust.
Objections
Objections are the pushbacks and challenges the persona raises during the conversation. They create the friction that makes the learner use the skill. Write 3–5 objections from the persona's point of view, grounded in the scenario.
Objections field
The AI weaves these objections into the conversation naturally. They are not read verbatim, but they guide how the persona responds.
✍️ Write the objections, not the answers. Include only what the persona says, not how the learner should respond. The learner's job is to work out the response.
Examples for a sales roleplay:
Examples for a sales roleplay:
"We're not budgeted for this this year."
"Your competitor's pricing is lower."
"We need to get buy-in from our CFO first."
Examples for a customer support roleplay:
Examples for a customer support roleplay:
"This is the second time this has happened to me."
"I don't have time for a long process to fix this."
"I want a refund, not a replacement."
Examples for a manager feedback roleplay:
Examples for a manager feedback roleplay:
"I don't understand why I got a lower rating than my peer."
"I'm not sure this feedback is actionable."
"I feel like this is the first time I'm hearing about this."
Skills
Skills are the observable behaviors the AI scores the learner on after the roleplay. They drive the feedback scene — what the learner gets credit for, what the coaching highlights, and whether they pass.
Where skills live
Skills are organized under two tabs:
Synthesia — a library of prebuilt skills maintained by Synthesia, ready to add to any persona.
Workspace — custom skills created by your team.
To create your own, select + New skill, define a skill specific to your team's methodology, and it will appear under the Workspace tab.
Writing effective skills
How many: Add 3–5 skills. More than 5 dilutes the signal and makes feedback harder to act on.
Weighting: Each skill carries a weight, and the weights add up to 100%. To stop a weight from redistributing when you adjust others, select the lock next to that skill.
What makes a good skill: Write skills as observable behaviors you could watch happen in a conversation, not abstract qualities. The AI scores based on whether the behavior shows up, so vague skills produce inconsistent scoring. For example, use "Acknowledges the employee's perspective before introducing their own point of view" rather than "Shows empathy".
Examples:
❌ Too abstract | ✅ Observable |
Shows empathy | Acknowledges the employee's perspective before introducing their own point of view |
Communicates clearly | Uses specific examples rather than general statements when describing the issue |
Stays calm | Doesn't interrupt or escalate when the employee pushes back |
The rubric: Each skill has a rubric — a description of what each score level looks like. The AI uses this to evaluate the conversation. Write rubric criteria in terms of what the learner did or didn't do, not how they felt or what they intended.
✍️ The rubric field is for the AI's evaluation criteria, not learner-facing feedback copy. Think of it as instructions to the scorer, not a message to the learner.
💡 If you're creating skills from scratch, use AI to draft them. Give it your scenario context and ask it to write 4 observable skills with a 0–3 rubric for each. Then review and adjust — it's much faster than starting from a blank field.
Time limit
The time limit sets the maximum conversation length. When the limit is reached, the roleplay scene ends and the learner moves to the feedback scene.
Recommended ranges are:
5–7 minutes for practice
8–12 minutes for assessments
If a learner finishes before time runs out, they can end the call early. If they reach the time limit, the roleplay scene ends automatically. Either way, they are scored on the conversation they had.
💡 Tell learners the time limit in your scripted intro scene. Knowing there is a limit changes how learners pace the conversation. If a learner ends the call early, they still get a feedback scene scored on the conversation so far.
How to configure the feedback scene
The feedback scene is auto-generated — you set the logic, and the AI writes the feedback.
To configure it:
Add the feedback scene after the roleplay scene.
Set the passing score, for example 75%.
Turn Retries on or off, depending on whether the roleplay is practice or an assessment.