With AI’s potential to increase the quality of collaborative outputs, teams should incorporate it into their workshops to maximize success. That starts with thoughtful preparation — but once the workshop begins, facilitators need to guide participants in using AI effectively, creatively, and collaboratively. This article discusses how to do just that.   

1. Narrow AI-Generated Ideas Before Sharing 

To understand how to run effective AI workshops, we need to consider how human and AI ideation differ. 

For one, most humans tend to fixate on their first few ideas and must be pushed to think outside the box. They also tend to settle on one idea — whether it’s good or not — often because someone charismatic or authoritative has suggested it. AI does not struggle with these problems. 

This difference is a bit like that between sheep and rabbits: 

• People (the sheep) tend to follow a leader and hesitate to stray far from their starting point. Even smart people working together often move as one.  
• AI (the rabbits) isn’t as predictable. People who use AI can generate more divergent ideas than those who don’t, because AI isn’t swayed by groupthink, moves on quickly, and doesn’t get tired

This contrast is a great benefit and also a big challenge when involving AI in workshops and ideation. In just 5 minutes, a few people armed with an AI tool can come up with more information than they have time to fully consider or share. In a Procter & Gamble study (discussed in a previous article), both individuals and teams using AI submitted much longer solutions (measured by word count) than those working without AI. 

To solve this problem, workshop facilitators must first understand how AI enhancement alters the workshop structure. In human-only sessions, most ideation-based activities have a diverge–converge structure: 

1. Diverge and spend time coming up with as many ideas as possible. 
2. Converge and work with the group to cluster, discuss, and prioritize everyone’s outputs. 

When AI is involved, this second step is much harder because there can be hundreds of ideas to consider. You could have people limit the number of ideas the AI is generating, but this undermines its superpower and violates the proper ideation mindset. Personally, I rarely use AI to help ideate without asking it to provide at least 15–20 different options per prompt.  

Instead, use the following structure to benefit from diverging and converging while using AI:  

1. Diverge and spend time using the AI to come up with as many ideas as possible. 
2. Have the AI narrow down the ideas to top suggestions, based on provided criteria. 
3. Converge and work with the group to cluster, discuss, and prioritize everyone’s outputs. 

2. Prepare Evaluation Criteria for AI Outputs 

Now, the facilitator must decide what the criteria for narrowing should be.  

This question becomes a discussion about prioritization. There are many potential methods for this, but, in my experience, given that LLMs are text-based, a scorecard works best. That way, the AI can provide a rating for each idea based on different criteria. 

You can either provide these criteria ahead of time or collaboratively create them as part of the workshop. This is your choice. Here’s an example of a sample prompt I might give participants if I had decided the criteria before: 

We can’t implement all these ideas. We need to narrow in on around 5 of the best ones. Use the following criteria to select which of these ideas we should move forward with: 

• Feasibility: The degree to which the item can be technically built. Does the skillset and expertise exist to create this solution? 
• Desirability: How much users want the item. What unique value proposition does it provide? Is the solution fundamentally needed, or are users otherwise able to accomplish their goals? 
• Viability: If the item is functionally attainable for the business.  Does pursuing the item benefit the business? What are the costs to the business and is the solution sustainable over time? 

Follow these steps: 

1. ​​​​​​Rate each of the ideas you’ve provided on a scale of 1 – 10 for each of the criteria provided. For example, 1 = low feasibility, meaning it would be impossible to build, and 10 = high feasibility, meaning it would be very easy to build. Apply this same framing to each criterion. 
2. Provide a rationale for each of your ratings. 
3. Add the three scores from each idea together to create a total score out of a potential of 30. 
4. Provide all ideas and their respective ratings in a table, ranking them from the highest total scores at the top to the lowest total scores at the bottom. 

You can swap out feasibility, desirability, and viability with other criteria relevant to your context. The AI’s ratings will also become more meaningful if you provide more context (such as uploaded documents or even a custom AI — as discussed in another article in the series). 

3. Plan More Time for Reading and Prompt Iteration than You’d Think 

Because the AI generates ideas so quickly, it’s easy to assume that AI-assisted people won’t need as much time to work independently as they would when doing all the mental work themselves. This isn’t true. People working with AI generally need just as much time, if not more, as people working on their own. Now, they need to write thoughtful prompts, read through the responses, and repeat the cycle a few times. 

To individually generate ideas for one problem, I typically give people without AI around 5 minutes; with AI, I give them around 8–10 minutes. However, this duration varies based on the group’s familiarity with AI, the complexity of the problem, and how quickly people seem to be wrapping up. 

4. Encourage People to Cowrite Prompts 

When people diverge to use the AI in ideation, they don’t always need to do it alone. Sometimes it’s powerful to allow a group of 2–4 people to collaborate on a prompt, then work through its outputs together. I’d suggest this approach in a few different scenarios: 

• You are facilitating a group larger than 10, so everyone will have limited time to share what AI helped them develop. 
• Some people are not comfortable using AI and might learn from working with others who are. 
• You are looking for specific outputs, and prompts need to be carefully crafted to get there. 

This is not essential, but a powerful option. The discussion people have while drafting a prompt together is often as valuable as the outputs they receive. Cowriting the prompt aligns each participant’s thinking and helps them prepare to discuss the outputs. 

5. Document Things Digitally 

We love our physical sticky notes and Sharpies, but most teams increasingly run workshops remotely or in a hybrid format. Thus, I recommend using some form of digital documentation for the workshop to allow people to easily copy and paste insightful AI outputs. My favorite format is simply a shared document, but a shared slide deck or digital whiteboard tool can work too. Just remember that AI outputs will be much longer than typical human outputs, so they won’t always fit well on a digital sticky note. 

Conclusion 

AI use in workshops requires facilitators to rethink and retool their normal methods and strategies. But its benefits are worth it. By bringing these strategies into your next workshop, you’ll be one step closer to optimizing your team’s ideation.  

Reference

Fabrizio Dell’Acqua, Charles Ayoubi, Hila Lifshitz-Assaf, Raffaella Sadun, Ethan Mollick, E., Lilach Mollick Yi Han, Jeff Goldman, Hari Nair, Stew Taub, Karim R. Lakhani, 2025. The cybernetic teammate: A field experiment on generative AI reshaping teamwork and expertise (March 28, 2025). Harvard Business School Strategy Unit Working Paper No. 25-043, Harvard Business School Technology & Operations Mgt. Unit Working Paper No. 25-043, Harvard Business Working Paper No. No. 25-043, The Wharton School Research Paper, Available at SSRN: https://ssrn.com/abstract=5188231 or http://dx.doi.org/10.2139/ssrn.5188231