Synthesia Sessions: talk to an AI avatar and practise a conversation

Give the avatar a conversation to practise

Roleplay Sessions lets you practise one side of a conversation. In Luke’s example, Alex Mercer is a simulated operations director at a fitness business. The task is to understand the client’s problem before pitching a solution.

1. Open Roleplay Sessions and choose a relevant starter or create your scenario.
2. Set the avatar persona, the situation and the skills to assess, such as listening and asking useful questions.
3. Check the time limit and retry settings, then generate the session.

Speak, inspect the feedback and retry

Open the generated player, play its introduction, enable the microphone and join the conversation. Luke’s video includes the actual exchange, coaching and a second attempt. The displayed scores are 45 and 48 out of 100; this small demonstration does not establish reliable training outcomes.

1. Ask an open question, listen to the reply and follow up on what was actually said.
2. After the call, ask the coach for a specific change and the moment in your conversation that supports it.
3. Retry the same situation with one change. Compare the conversation and feedback rather than assuming a higher score proves better performance.

Let the avatar ask the questions

Survey Sessions reverses the roles: the avatar interviews the participant. Luke builds a short workshop-feedback example and checks whether the follow-up responds to a brief answer. The workshop and client are demonstration scenarios.

1. Open Survey, describe the feedback you want and review the generated questions.
2. In the editor, adjust the questions, follow-up behaviour and session length. Luke changes his generated example to three minutes.
3. Generate the survey, play the introduction, enable the microphone and begin. Start with a short answer, then respond naturally to the follow-up.

Check what was saved

Open the survey’s response and reporting views to see what the system retained. The Survey Analytics screens in the finished video use real Synthesia output from a separate off-camera run with the same answers, rather than the exact on-camera interview.

1. Compare the saved response with the answer given: did it preserve the specific setup problem?
2. Check summaries for omissions before treating them as research findings.
3. Use an appropriate scenario and non-sensitive demonstration details when trying the workflow.

The takeaway

Give the avatar a specific job, then judge the conversation, coaching and saved response against what actually happened.