You know the moment. Someone in the workshop says “our clients clearly want more flexibility” and participants nod, and nobody asks where that came from. Maybe it’s a hunch from one sales call. Maybe it’s something someone read six months ago and half-remembered. Either way, the group builds the next hour of the session on top of it, and nobody stops to check if it’s true.

That’s workshop theatre: the session looks productive, people are talking, sticky notes are going up, but underneath it there’s no data holding any of it together. A conversation with a colleague last week clarified something for me: AI might be able to do the one thing most workshops skip, checking whether the assumptions participants raise actually hold up.

People rarely bring real data into a workshop. Pulling it up mid-session breaks the flow everyone’s protecting, so the group defaults to whoever sounds most confident, and confidence gets mistaken for evidence. This is the specific gap I think AI can close, not by designing the session or reading participants, but by finding the actual research behind an assumption someone just stated out loud, before three more decisions get built on top of it. You might have seen Microsoft recently announce a “Facilitator” feature in Teams; it fills knowledge gaps on the fly during a meeting. What I’m describing is narrower, and it only works if a participant is checking that what it finds is real, current, and relevant. Skip that check, and you’ve just swapped one kind of theatre for another, confident AI output instead of confident gut feeling.

Three ways to actually use it in a session

1. Invite AI as a participant. Let it run in the background of the workshop, searching in real time as topics come up, and feed what it finds directly into the discussion and activities. It’s not leading anything, it’s contributing the way a well-prepared colleague would, except it can look things up faster than anyone can type a search query mid-conversation.
2. Have participants do their own lookups. When an assumption or idea surfaces, ask the person who raised it, or anyone in the group, to use AI to find sources or data supporting or challenging it, then share what they found with everyone. This keeps the responsibility for verification with participants, which is exactly where it should stay.
3. Ask AI to answer the question before you run the activity, and use its answer as a straw dog. Before the group works through a problem themselves, have AI take a first pass at solving it. Don’t hand that answer over as the solution. Hand it over as something to argue with. Groups are often sharper at critiquing an existing answer than generating one from a blank page, and a flawed AI-generated answer gives them something concrete to push against instead of staring at an empty whiteboard.

Where this actually leaves us

None of this makes AI a workshop designer or a facilitator. It doesn’t decide what activity comes next, it doesn’t read what participants are feeling, it doesn’t know when to slow down or when to push. What it can do, if you build in the discipline to check its work, is close the exact gap that produces workshop theatre in the first place: the gap between what people assume and what’s actually true. That’s a smaller job than “run my workshop for me,” and it’s also a real one. The question “wait, is that actually true” is easy to skip mid-session; an AI that’s willing to go find out, with someone checking its homework, might be the thing that finally makes it easy enough to ask.