AI simulations in workplace learning: Voice vs Text?
When it comes to developing communication skills, AI simulations are opening up some exciting possibilities. But as the technology becomes increasingly sophisticated, there's an important question for learning designers: does a more realistic experience necessarily lead to better learning?
One of the most promising applications of artificial intelligence in workplace learning is the ability to practise important conversations. Whether it's a manager delivering difficult feedback, a customer service representative responding to a complaint, or a social worker navigating a sensitive discussion, AI simulations offer opportunities to rehearse real-world situations in a controlled environment. We’ve talked about this in previous blogs.
Unlike traditional branching scenarios, where learners select from predetermined responses, generative AI simulations can respond dynamically to what someone says or types. Learners can explore different approaches, experience the consequences of their choices, receive feedback and try again.
And now, with increasingly sophisticated voice-to-voice technology, those conversations can feel remarkably natural.
Now that we can provide voice simulations, should we be rushing to replace all text-based simulations with voice?
What does the research tell us?
Recent research provides some interesting insights into the different learning experiences offered by text and voice.
In a 2026 study, El Sawah and colleagues compared text-based AI simulations with voice-based avatar simulations involving 24 postgraduate clinical psychology students. Both approaches were positively received, but participants rated the voice-based experience more highly for perceived usefulness, skill application and improvement. They particularly valued the social and emotional cues available through the avatar.
However, the researchers also identified benefits of text-based interactions, including opportunities for reflection and lower-pressure practice (El Sawah et al., 2026).
Importantly, the study measured participants' perceptions rather than objectively demonstrating which approach produced better learning outcomes. And because the voice simulation also incorporated an avatar, the findings cannot be attributed to voice alone.
Another study involving 27 medical students found that voice-based AI simulations increased participants' self-reported confidence in handling challenging conversations, including breaking bad news and supporting anxious patients (Mukadam et al., 2025). While promising, the researchers acknowledged that further investigation is needed to determine whether increased confidence translates into improved real-world performance.
This distinction is important. An experience can feel more engaging, immersive or realistic without necessarily producing better learning outcomes.
Different experiences, different learning opportunities
Text and voice offer different opportunities for developing communication skills, and understanding those differences is an important part of effective learning design.
Text-based simulations give learners time to consider their responses. They can pause, reflect on their choice of words and experiment with different ways of approaching a situation. This may be particularly valuable when developing foundational conversational skills, exploring complex decisions or learning to respond thoughtfully to challenging circumstances.
Voice-based simulations introduce additional dimensions. Learners need to formulate responses in real time, listen carefully, manage the flow of conversation and consider how their words are delivered. Depending on the technology, they may also be able to practise tone, pacing and responding to emotional cues.
For someone preparing to deliver performance feedback, a text-based simulation might help them develop constructive language and explore different responses. A voice-based simulation could then help them practise delivering that feedback aloud, responding to hesitation or resistance as the conversation unfolds.
Neither approach is inherently better. The question is what we're trying to help people learn.
The important role of emotion and psychological safety
Of course, realistic conversations aren't always comfortable conversations.
In many professions, employees need to navigate emotionally demanding situations. Social workers, healthcare professionals, managers and frontline employees may regularly encounter distress, anger, conflict or vulnerability.
Practising these conversations before encountering them in the workplace makes sense. But how do we provide meaningful opportunities to practise without creating unnecessary distress for learners?
A 2025 review by Madsgaard and Svellingen explored psychological safety in simulation-based education. Their findings highlight an important tension: meaningful learning often involves uncertainty, challenge and the possibility of making mistakes, yet learners also need to feel sufficiently safe to engage with those challenges.
Psychological safety, therefore, isn't necessarily about removing discomfort. It's about creating an environment where learners can take appropriate risks, reflect on their experiences and develop their capabilities without feeling threatened or exposed.
This becomes especially important when designing simulations involving sensitive or potentially traumatic situations.
For example, professionals supporting survivors of abuse need opportunities to practise sensitive, compassionate communication. However, a highly immersive voice simulation could be distressing for some learners, particularly where the scenario touches on personal experiences.
A text-based simulation may provide a more appropriate starting point, allowing learners greater control over the pace and emotional intensity of the experience. Other learners may be ready to rehearse the conversation aloud.
The appropriate response isn't to avoid challenging scenarios altogether, nor to assume that greater realism is always desirable. It's to design experiences thoughtfully, considering the learners, the context and the potential emotional impact.
This might involve clear information about scenario content, opportunities to pause or step away, choices about how to participate, carefully calibrated levels of challenge and supportive debriefing. In particularly sensitive contexts, subject-matter and trauma-informed expertise should inform the design.
Learning design matters more than the technology
There's another consideration that's easy to overlook when exploring new AI capabilities: a sophisticated simulation is only as valuable as the learning experience built around it.
A 2025 study comparing voice-based AI simulations with actor-based role-play found that both approaches improved medical students' self-reported communication skills. Although the actor-based approach achieved somewhat higher ratings, AI offered a lower-cost way to provide additional practice opportunities (Tyrrell et al., 2025).
The implication for workplace learning is significant. AI simulations can make repeated, individualised practice more accessible, particularly where facilitating live role-plays for large or geographically dispersed teams would be difficult.
But making practice available is only part of the equation.
Effective simulations require realistic and relevant scenarios, clearly defined learning objectives, appropriate levels of challenge, meaningful feedback and opportunities to reflect and improve. We also need to consider how the learning will transfer into workplace performance, rather than relying solely on whether learners enjoyed the experience or felt more confident afterwards.
At Pukeko Learning Solutions, we've seen the potential of thoughtfully designed AI simulations through our award-winning text-based simulation work. As new technologies create opportunities for more natural spoken interactions, our focus remains on making purposeful learning design decisions.
Sometimes that will mean using text. Sometimes voice will offer important additional benefits. And sometimes a combination of both may provide the most valuable learning journey.
The opportunity isn't simply to make simulations feel more like real life. It's to create the right conditions for people to practise, reflect, build confidence and develop the capabilities they need when those conversations really matter.
Would you like to see some AI simulations in action?
Click here to receive demo links for a text-based AND a voice-based simulation, and experience for yourself what the difference is like.
Watch the AI Simulation: Text Demo
Watch AI Simulation: Audio Demo
References
El Sawah, V., Bhardwaj, A., Pryke, A., Gamaleldin, D., Nice, J., Ang, C. S., & Martin, A. K. (2026). Artificial intelligence as a training tool in clinical psychology: A comparison of text-based and avatar simulations. Computers in Human Behavior: Artificial Humans, 9, Article 100383. https://doi.org/10.1016/j.chbah.2026.100383
Madsgaard, A., & Svellingen, A. (2025). The benefits and boundaries of psychological safety in simulation-based education: An integrative review. BMC Nursing, 24, Article 922. https://doi.org/10.1186/s12912-025-03575-y
Mukadam, A., Suresh, S., & Jacobs, C. (2025). Beyond traditional simulation: An exploratory study on the effectiveness and acceptability of ChatGPT-4o advanced voice mode for communication skills practice among medical students. Cureus, 17(5), Article e84381. https://doi.org/10.7759/cureus.84381
Tyrrell, E. G., Sandhu, S. K., Berry, K., Ghannam, S. F., Lewis, S. A., Crowfoot, D., Sahota, G. S., Carson, J., Wilson, E. E., & Taggar, J. (2025). Web-based AI-driven virtual patient simulator versus actor-based simulation for teaching consultation skills: Multicenter randomized crossover study. JMIR Formative Research, 9, Article e71667. https://doi.org/10.2196/71667









