All methods
IdeationIntermediate

Diverge-Narrow-Converge with AI

Diverge-Narrow-Converge with AI is an ideation structure for workshops in which participants generate ideas with an AI chat assistant, have the assistant score and shortlist those ideas against agreed criteria, and only then discuss and prioritise as a group. The structure was described by Tanner Kohler of Nielsen Norman Group in 2025 as an extension of the familiar diverge-and-converge technique. The added middle step exists because an AI assistant produces far more ideas, and far longer ones, than a group can read and discuss in a workshop.

Duration
30m–45m
Group size
3–30 people
Materials
An AI chat assistant per person or per small group, Shared document or digital whiteboard, Agreed evaluation criteria and a scorecard prompt
Origin
Tanner Kohler, Nielsen Norman Group

Facilitation script

  1. 1

    Frame the challenge, agree the evaluation criteria and share the scorecard prompt. Form groups of two to four if people are sharing an assistant.

    7 min
  2. 2

    Diverge: participants prompt the assistant for many ideas, add their own and push for variety.

    10 min
  3. 3

    Narrow: participants run the scorecard prompt so that the assistant rates, explains and ranks all ideas.

    5 min
  4. 4

    Participants review the scorecard, adjust what they disagree with and post their top ideas, one line each, on the shared board.

    6 min
  5. 5

    Converge: cluster the posted ideas, discuss the clusters and prioritise as a group.

    13 min
  6. 6

    Record the selected ideas with reasons and agree next steps.

    4 min

Tips

  • Give everyone the same scorecard prompt so that rankings from different people or groups are comparable.

  • Be explicit that the AI ranking is a filter for reading time and not a decision: its ratings are plausible-sounding estimates with no knowledge of your users, budget or politics beyond what was typed in.

  • Ask people to shorten ideas to one line before posting, since raw AI output does not fit on a sticky note.

  • Check before the session what participants are allowed to enter into the tool.

Common pitfalls

  • Accepting the AI ranking as the result, so the group adopts ideas nobody has examined and loses the unusual ones the scorecard marked down

  • Setting criteria only after the ideas are on the table, which lets people pick criteria that favour their preferred idea

  • Giving the AI-supported divergent phase the same five minutes as an unaided one; participants need longer because they spend much of the time reading output

  • Pasting raw AI output onto the board, which produces walls of text that nobody reads and stalls the converge step

  • Letting each person use a different scoring prompt, so the shortlists cannot be compared and the group argues about method

Variations

With more than about ten participants, with people who are new to AI tools, or when the prompt needs careful wording, have groups of two to four share one assistant and one screen. In a mixed room, let some participants ideate without AI for about five minutes while others work with it, then compare the two pools. If the organisation has its own decision criteria, use those in place of feasibility, desirability and viability. The structure runs equally well in person, remotely or hybrid, as long as everyone can reach the shared board.

Where it fits

Ideation sessions in which participants use AIShortlisting a very large pool of ideasRemote and hybrid workshopsFeature and concept generation in product teamsIntroducing a team to working with AI in a structured way

When to use it

  • Participants have access to an AI chat assistant and are permitted to use it for the topic at hand

  • A previous AI-supported brainstorm drowned the group in output and the discussion never got past reading

  • The challenge is well enough defined that clear evaluation criteria can be agreed before ideation starts

  • The group is large or distributed and needs a quick way to bring comparable shortlists into one conversation

  • You want quantity from the divergent phase without spending the session on sorting

When not to use it

  • The topic involves confidential, personal or client data that may not be entered into the available tool; run a classic Brainstorming or Brainwriting round instead

  • The value of the session lies in people hearing and building on each other's thinking; use 1-2-4-All or Brainwriting so that ideas come from the room

  • The criteria are disputed; settle them first, because a scorecard built on contested criteria only moves the argument to the ranking

  • The decision depends on knowledge the assistant cannot have, such as internal politics or unpublished research; let the people who hold that knowledge do the narrowing, for example with Dot Voting

Related methods

IdeationBrainstorming
IdeationBrainwriting
IdeationAffinity Map
DecisionDot Voting

Frequently asked questions

What is Diverge-Narrow-Converge with AI?▾

It is a three-step ideation structure for workshops in which participants use an AI chat assistant. People first generate a large number of ideas with the assistant, then ask it to rate and rank those ideas against agreed criteria, and finally discuss and prioritise the shortlist as a group. Tanner Kohler of Nielsen Norman Group described the structure in 2025.

Why add a narrowing step to diverge and converge?▾

An AI assistant can produce dozens of ideas in minutes, each longer than a human sticky note. Without a filter the group spends its convergence time reading. Having the assistant score the ideas first cuts the pool to a size people can discuss.

Can the AI's scores be trusted?▾

Only as a first sort. The assistant rates ideas from general knowledge and from whatever context was typed in, and it gives confident reasons either way. Participants should read the reasons, change ratings they disagree with and feel free to rescue an idea the scorecard ranked low.

Should participants work alone or in groups with the AI?▾

Both work. Sharing one assistant in groups of two to four helps when the workshop has more than about ten people, when some participants are not comfortable with AI tools, or when the prompt needs careful wording. Individual work gives more variety when everyone is fluent.

How long does the exercise take?▾

About 30 to 45 minutes. The divergent step with AI needs around eight to ten minutes, the scoring and review a similar amount, and the rest goes to group discussion and prioritisation.

🪡

Plan your next workshop with AI

Workshop Weaver helps you combine methods like Diverge-Narrow-Converge with AI into a complete, timed agenda in minutes.

Try it free

Method descriptions on Workshop Weaver are original content written by our team, based on established facilitation practices. This method was inspired by work from Tanner Kohler, Nielsen Norman Group.