How to use AI for workshop facilitation: my full workflow, with prompts

AI cut the time I spend on a workshop engagement in half. That covers the preparation, the facilitation and the delivery.

I used to rely on a full team for this. Facilitators, designers and analysts helped me at every step. Now I work alone, and AI is what makes that possible.

AI can do almost anything in your workshop if you let it. The result will be average, and your clients won’t come back. So I won’t list everything AI can do. I’ll show you how I use AI for workshop facilitation, step by step, with the prompts I use.

A small disclaimer: I’m not an AI expert. This is how I work today, after six years of facilitating workshops and ten years in consulting. Take what fits your own process.

The example I’ll use

To show every step, I’ll use a made-up client called Pure Computers. They make and sell computers. Their six-person customer experience team wants an offsite to define a vision for 2027, the objectives to get there and an action plan.

I use Claude for all of this. I switched from ChatGPT because everyone was talking about Claude and I had big FOMO. Most of what follows also works in ChatGPT.

Step 1: use AI to prepare the discovery call

Start by creating a project in Claude, with one project per client. I’d use the desktop app, because it gives you the full set of features. Everything you add to the project stays available for the whole engagement.

Then ask for help preparing the meeting.

Once it has the context, go deeper.

For my made-up client, it came back with warranty and repair logistics, the difficulty of telling a hardware problem from a software one, and consistency across channels.

That’s useful because clients often feel they need a specialist from their own industry. Walking into the call with a real picture of their world lets you ask better questions and challenge them. I keep this part short, though. It’s only a first call.

A tip that saves me hours: stop typing

I dictate my prompts with a tool called Wispr Flow. I press a key, I talk, and it writes. Most of the time it even understands when I correct myself mid-sentence. I think I’m about three times faster this way. Other tools do the same job.

Turn the call into a written confirmation

Most of my discovery calls are remote, so I can record them and get a transcript. With the client’s consent, I paste that into the project.

That last sentence matters. AI hallucinates, which means it invents information that was never there. Telling it to stick to the transcript reduces that risk, and you still have to check everything it gives you.

Why bother with this email? I’ve worked with clients who later said I hadn’t understood their need. I ended up redesigning the session a couple of times, and once the client wasn’t happy at the end of the workshop. Ten years of consulting taught me to validate the need on paper. If things go wrong, you have proof. It also makes you look very professional.

Never send the first version. Check the structure, cut what isn’t meant for a client, and remove the dashes, because people spot those as AI straight away.

Better still, give it your past emails with the sensitive data removed, and ask it to turn them into a reference document. In Claude that’s a Markdown file, an MD file. An example teaches AI far more than a long list of instructions, and it gives you better results.

Step 2: keep the workshop design for yourself

This is where I’ll disappoint some of you. I don’t use AI to design my workshops.

I enjoy this part, and it’s the riskiest part of the journey. Clients trust me to reach a certain outcome, and I don’t feel AI can carry that.

There’s a bigger reason. Designing a session is how you get to know the client’s context. You ask yourself “what if I do this?” and “what if I go that way?”. That’s a workout for your brain.

Your workshop won’t follow your agenda anyway. At some point you’ll have to change direction and improvise. If you’ve already explored those routes while designing, you’ll be more comfortable and faster when it happens in the room.

I’ve tested ChatGPT, Claude and Gemini on real briefs. They do an okay job, and most of the time I don’t like what I get.

If you still want AI to design for you, look at MCPs, the connectors that link Claude to your other tools, like Miro or Stormz. I’m also working on one for the Workshop Deck app, because my users keep asking for it. Please review whatever it gives you, line by line. This is the part you can’t get back once you’re in the room.

There are two design jobs I do hand to AI.

Test a new activity before you run it

When I try an activity for the first time, I’m not always sure it will give me what I need at that moment of the session. So I ask AI to play the participants.

For Pure Computers, I wanted to open with an expert interview, 20 minutes with the team leader to surface the biggest challenges.

The first answer missed the point, because it didn’t know how my expert interview works. So I pasted my activity card and asked for the “How Might We” notes the team would write.

How might we support the whole product, hardware, software and subscription, as one experience?

That’s what I needed to see. I knew the activity would produce material I could use in the next step. So I kept it and moved on.

If the answer is full of jargon, ask AI to talk to you as if you were 12.

Stress-test your full agenda

Once the agenda is designed, paste it in and ask for a tough review.

The 25% question is my favourite, because workshops always run over. On my test agenda, it told me the plan was rich in activities and poor in decisions. That’s the kind of sentence you want to read before you’re in the room.

Step 3: use AI to prepare everything for the room

This is where AI starts to save serious time.

Workshop slides

When I use slides, I give AI an example of one of mine and describe the deck I want. A context slide, a goal slide, a purpose slide that explains why we’re spending a day together, then the workshop principles and the instructions for each activity.

If your branding is fairly classic, Claude can build a clean PowerPoint right away, with elements you can edit. For something more specific, upload your templates to Claude Design. My own style is hard for AI, with hand-drawn visuals and a handwritten font, so I get a decent result, and I finish it myself.

Examples for your activities

Examples help participants understand what you’re asking for, and they can check their own work against them. Getting the level right is hard. If the example is too close to the answer, you steal it from them. If it’s too far away, it doesn’t help.

AI is good at finding that sweet spot.

I also ask for examples in the style of real companies, one direct competitor, one from a similar industry and one from far away. These are invented for inspiration, so present them that way. I use one to three.

Inspiration for lightning demos

In a lightning demo, people present companies that solved a similar challenge. Left alone, clients go straight to Google and Apple, which rarely relate to their problem. I used to spend hours researching better references, and I asked my team to do the same. AI now does that job very well. Just check that the examples are real.

A briefing in the car

I often drive to my workshops, so I use voice mode to prepare on the way. How is a customer experience team usually structured in a hardware company? What are the quick wins? It’s a good way to arrive with the context fresh in my head.

Step 4: why I keep AI (almost) out of the room

I barely use AI during facilitation. A laptop in a workshop quickly leads to emails and Teams messages.

I tried it when the hype started. My clients kept saying “let’s ask AI”. They got so focused on what it would say that they stopped trying to find answers themselves.

The magic of a workshop comes when participants share their own ideas. That’s what makes them motivated to implement the result. If the ideas come from AI, you lose it.

If you want AI in the room, my friend Alexandre Eisenchteter, who knows far more than me about this, has a principle called the sandwich method. You always start with a human layer, you add an AI layer that reacts and builds on the group’s ideas, and you finish with a human layer that votes or decides.

What I do recommend is recording the session, with consent. It’s easy for remote workshops. In the room, you can use your phone or an AI recorder. I bought a Plaud and I’m very happy with it.

Someone I coach recorded a session, gave the transcript to Copilot, and had it send Outlook invites for every next step, with deadlines, to the right owners. That’s impressive. In my sessions a lot of the work happens silently on sticky notes, so a transcript alone isn’t enough for me.

Update: the one AI tool I now bring into the room

Since I recorded the video, one thing has changed. I now use an AI note taker during my workshops.

Its role is very limited. It sits on the table and records, with the group’s consent.

That small change made a big difference for me. I used to scribble down everything people said, afraid I’d miss an important insight I wouldn’t remember later. Now I know it’s all being captured, so I can stay fully present with the group.

Of course, I still write down anything we’ll use later in the session, because the group needs it in front of them.

It also makes my reports stronger. When I build the deck, I have the photos of the sticky notes and the conversation that happened around them, so the nuance doesn’t get lost.

I explain which one I use and why in The best workshop tool I’ve bought.

Step 5: build the workshop report with AI

The deliverable is the most time-consuming part of the journey, and it’s where AI helps me most.

I always send a slide deck, because clarity on what happened and what comes next is what gets things done after the workshop. It’s also a big time sink, like the one I wrote about in when workshop design becomes procrastination. This is where you get the hours back.

Build a template from your best decks

I created a Claude project and uploaded about ten of my best workshop decks, cleaned of sensitive data. AI cares about the structure and the level of detail, so you don’t need the real content.

From these, AI builds a reference file with my slide structure: a cover, the context and objectives, the takeaways from the activities where we made decisions, and the action plan. I leave out activities that only moved us to the next milestone.

My three rules for client data

Please don’t skip this part. You can do a great job and still lose a client over data, especially in Europe with GDPR.

1. Anonymise everything. Remove the client’s name and logo, people’s names and sensitive numbers. If a sticky note is very specific to that client, remove it. If it’s generic, like “we never know who’s in charge of what”, I may keep it to show the level of detail I want.

2. Get consent. Only use a recording or transcript if the client has agreed that you can record it and put it in your AI tool.

3. Check your settings. In Claude or ChatGPT, turn off the option that lets them use your data to train their models.

Claude also has an incognito mode. The chat doesn’t stay in your history or memory, and you can’t go back to it afterwards. Check what your client’s contract allows before you rely on it.

From sticky notes to slide deck

At the end of the workshop, I photograph every exercise. Transcribing those notes used to be the slowest part.

The Miro and Post-it apps can do this too. I clean up the result, remove anything sensitive, and paste it into the project with my template.

Then validate the text before you ask for the slides. Changing finished slides consumes far more tokens, so get the words right first.

AI gets you from 0 to 80%. Spend your time on the last 20%, the part that makes the document special and makes your client use it.

Go further with a tool

A deck tells your client what to do. A tool shows them how, which makes it much easier to act. Claude can create a Miro board with the objectives, the key results and empty spaces for the actions, and Claude Code can build a simple OKR tracker. Big companies can be sensitive about this, so ask about privacy and security first.

The tools I use

• Claude for almost everything in this article.

• Wispr Flow to dictate my prompts.

• OpenArt when I want to surprise a client with a short video or visual. It gives you access to several image and video models with one set of credits, so you don’t need a subscription for each.

• ChatGPT for a couple of things, including voice mode, which I still prefer. It isn’t a must.

4 tips to get better results from AI

1. Treat AI like a junior consultant. It can do amazing work, but it can’t guess what you need. Tell it the goal of the workshop, the type of client and your facilitation style. You can even ask what information it needs, because too much context can also make results worse.

2. Ask it to be honest. AI models tend to flatter you, and that can be dangerous for your project. I use this prompt:

3. Let it ask the questions. When you don’t know where to start, give it your goal and add:

4. Show, don’t just tell. One to three real examples of your past reports will do more than any explanation you type in the chat.

Get all my prompts

If you want to see me do all of this on screen, the full tutorial is on YouTube. You can also find it on Spotify and Apple Podcasts.

FAQ

Can AI replace a workshop facilitator?

In my experience, no. AI can prepare material and save you hours, but it can’t read a room or earn a group’s trust. The ideas also need to come from participants, because that’s what makes them want to act on the result.

Can AI design a workshop for me?

It can, and the result is okay at best. I’d keep the design yourself and use AI to test your activities and stress-test your agenda. You’ll also be far better prepared when you need to improvise.

Which AI tool is best for workshop facilitators?

I use Claude, mainly for its projects and the quality of the slide decks it builds. Most of this workflow also works in ChatGPT, so start with the one you already have.

Is it safe to put client data into AI?

Only with care. Anonymise names and numbers, get the client’s consent before using a recording, and turn off model training in your settings. When in doubt, ask your client what their policy allows.

Should participants use AI during a workshop?

I avoid it, because it distracts people and they stop producing their own ideas. If you want to try it, use the sandwich method, starting and finishing with a human layer.

So, how are you using AI in your workshops? And which step would you never hand over?

Mehdi