A few years ago, the idea of an AI avatar running your workshop sounded like science fiction. Now you can type a script, pick a presenter and have a realistic video of a ‘person’ explaining your content in dozens of languages before lunch. Some avatars can even hold a live conversation, answering questions and playing a role in real time.
So it’s a fair question, and one I’m asked more and more: can an AI avatar replace a workshop facilitator?
My short answer is no, not if you mean facilitation rather than presenting. But I don’t think that’s the end of the conversation. AI avatars are genuinely useful, and facilitators who ignore them will miss out. The interesting question isn’t whether they replace us, it’s what we hand over to them and what we keep.
What is an AI avatar, exactly?
An AI avatar is a computer-generated character, often a realistic human face and voice, that speaks content generated or scripted with AI. They come in roughly three kinds, and they’re good for quite different things.
| Type | What it does | Where it fits a workshop |
|---|---|---|
| Pre-recorded video avatars | Turns a script into a video of a presenter, in many languages | Pre-work, welcome videos and short explainers |
| Real-time conversational avatars | Listens and replies live, and can play a character | Role-play practice, Q&A and simple coaching conversations |
| Avatars that stand in for you | Represents you on a video call, copying your expressions | Not much, yet: a novelty in most workshops |
The tools have moved quickly. Synthesia, best known for scripted training videos, now also offers interactive avatars for role-play.1 HeyGen’s LiveAvatar promises live, two-way conversations aimed squarely at learning and development.2 Zoom announced photorealistic avatars in September 2025 that can appear on a call for you when you’re not camera-ready.3
What AI can already do surprisingly well
Let’s give credit where it’s due. There are things an AI avatar does better than I do.
- It never has an off day: it delivers the same explanation in the same way, at 9am or 9pm, to the first group and the five-hundredth.
- It speaks your participants’ languages: a welcome or explainer video in Welsh, Polish and Arabic costs a fraction of what it once did.
- It’s a patient practice partner: rehearsing a difficult conversation with an avatar playing an awkward customer is lower stakes than doing it in front of colleagues, and you can try again as many times as you like.
- It’s good at finding common ground in text: this is the finding that surprised me most, and it’s worth looking at more closely.
In 2024, Google DeepMind researchers published a study in Science on what they called the Habermas Machine. More than 5,000 people in the UK took part. The AI collected each person’s view in private, then drafted group statements that people critiqued and the AI revised. Participants preferred the AI’s statements to those written by human mediators 56% of the time, and the share of groups reaching unanimous agreement rose from 22.8% to 38.6%.4
That’s impressive. But read the small print. This was a ‘caucus’ process: the AI never ran a live conversation. The researchers themselves point out that the approach lacks nonverbal cues and the chance to build relationships, and that it can’t keep a discussion on track or moderate it. They suggest using it as one part of a wider, human-led process.5
Where an AI avatar falls short
Facilitation isn’t delivering content. It’s noticing what’s happening in a group and responding to it, often before anyone has said a word. That’s where avatars struggle.
Reading the room
When I facilitate, I’m constantly reading small signals: the person who goes quiet after a comment from their manager, the energy that drops after lunch, the laugh that’s slightly too loud. Most of it is never said out loud. An avatar responds to what’s typed or spoken to it. It doesn’t notice who isn’t speaking, and the quietest people are usually the ones a facilitator most needs to notice.
Making it safe to speak
Amy Edmondson’s research on psychological safety showed that teams learn and perform better when people believe it’s safe to take interpersonal risks: to ask a daft question, admit a mistake or disagree with the boss.6 A facilitator builds that safety through dozens of small choices, from how they react to the first awkward contribution to whether they protect someone who’s been talked over. I’ve yet to see an avatar earn that kind of trust from a group.
Handling conflict and judgement calls
Workshops go off-script. Someone dominates, two people clash, or the group realises halfway through that it’s been solving the wrong problem. Handling those moments well needs judgement, a sense of the organisation’s history and sometimes the courage to say something uncomfortable. If you’ve ever had to deal with difficult participants, you’ll know that no script covers it.
Accountability
When a workshop goes wrong, someone needs to own it, follow up and put it right. An avatar can’t take responsibility for a group’s experience. And if the decisions made in a workshop matter, people want to know a human was looking after the process.
The rules you can’t ignore
If you’re thinking of using AI avatars or AI tools in workshops, there are some legal points to bear in mind, especially if any of your participants are in the EU.
- Tell people they’re talking to an AI: from 2 August 2026, the EU AI Act requires AI systems that interact with people to make that clear, unless it’s obvious, and requires AI-generated or manipulated video to be disclosed.7 Even where the law doesn’t require it, I’d always say so up front.
- Don’t use AI to read people’s emotions at work: since 2 February 2025, the EU AI Act has banned AI systems that infer people’s emotions in workplaces and educational settings, except for medical or safety reasons.8 Tools that promise to score how engaged or happy your participants look are risky territory.
- Remember UK GDPR: recording participants, or analysing their faces and voices, means processing personal data. Biometric data used to identify people is special category data, which needs extra care and a clear lawful basis.9
So where would I use an AI avatar?
I think of AI avatars as a capable assistant, not a co-lead. Here’s where I’d happily bring one in:
- Before the workshop: short explainer videos as pre-work, so the live session can be spent on discussion rather than slides.
- Practice and rehearsal: role-play partners for difficult conversations, sales calls or interviews, with a human debrief afterwards.
- Reaching more people: welcome messages and instructions in several languages for international or multilingual groups.
- Between sessions: quick reminders and recaps that keep momentum going on a longer programme.
- Drafting, not deciding: summarising contributions or drafting a group statement for people to critique and change, as in the Habermas Machine study.
And where I wouldn’t: opening and closing a session, anything involving conflict or sensitive personal topics, and any workshop where the group is making a decision that matters to them. Those moments need a human who can be held to account and who actually cares how it goes.
My verdict
AI avatars will replace some of what’s currently sold as facilitation, and honestly, that’s no bad thing. If a session is really a presentation with a few polls, an avatar can probably deliver it more consistently and more cheaply than a tired human on their third run of the day.
But good facilitation was never about delivering content. It’s about helping a particular group of people, on a particular day, think and decide together. That needs someone who can read the room, make it safe to speak, cope when things go sideways and take responsibility for the result. For now, that someone is human, and I’d argue that the more AI we use around a workshop, the more the human part matters.
My advice? Learn the tools, use them for the parts of your work they’re genuinely good at, and spend the time you save on the part only you can do.
Can an AI avatar run a workshop on its own?
An AI avatar can deliver scripted content, answer questions and play a role in real time, so it can run a simple, structured session. It can’t yet read a room, build trust or handle conflict and judgement calls the way a skilled human facilitator can, so it isn’t a substitute for facilitation.
What are AI avatars good for in workshops?
They work well for pre-work videos, multilingual welcome messages and instructions, role-play practice for difficult conversations, and recaps between sessions. They’re best used as an assistant to a human facilitator rather than as the lead.
Do I have to tell participants they are talking to an AI?
If your participants are in the EU, the EU AI Act requires AI systems that interact with people to make that clear from 2 August 2026, unless it’s obvious. Wherever you are, being open about it is good practice and helps build trust.
Can AI tools track how engaged participants are?
Some tools claim to, but be careful. The EU AI Act bans AI that infers people’s emotions in workplaces and education, and analysing faces or voices raises UK GDPR issues. Simple, open methods such as polls and pulse checks are a safer way to see how your group is doing.
Sources
- Synthesia: AI video and avatar platform. An example of a pre-recorded AI avatar tool. ↩︎
- HeyGen: LiveAvatar. An example of a real-time AI avatar tool. ↩︎
- TechCrunch (17 September 2025): Zoom launches a cross-application AI notetaker, AI avatars and more. ↩︎
- Harvard Program on Negotiation: Can AI Mediation Help Bridge Political Divides? A summary of the Habermas Machine study (Tessler and colleagues, Science, 2024), its results and its limits. ↩︎
- Tessler, Evans and colleagues (2026): on AI-mediated deliberation. The research team’s own assessment of what AI mediation can and cannot yet do. ↩︎
- Edmondson, A. (1999): Psychological Safety and Learning Behavior in Work Teams. Administrative Science Quarterly, 44(2), 350 to 383. ↩︎
- EU AI Act: Article 50, transparency obligations. ↩︎
- EU AI Act: Article 5, prohibited AI practices. ↩︎
- Information Commissioner’s Office: Biometric recognition guidance. How UK GDPR treats biometric data. ↩︎



