How to Run a Brainstorm on an Online Whiteboard: A Step-by-Step Guide

What you walk away with

• A board that is already structured when people arrive, so nobody faces a blank grid
• A generation round where quiet people contribute as much as loud people
• A clustering pass that produces themes rather than a tidier mess
• A decision that is written down in the session, not "circulated afterward"
• A board someone can open in three months and still understand

The one rule that decides whether it works

Separate generating from judging, and use the board to enforce the separation.

Almost every failed brainstorm collapses those two activities into one. Someone proposes, someone else evaluates, and the group starts optimizing the first idea instead of producing a wide field to choose from. In a room you fight this with facilitation. On a board you can build it into the layout: a zone where ideas go without comment, and a separate zone, revealed later, where they get judged. The physical separation does work that verbal instruction does not.

Everything below is in service of that one rule.

How AI fits across the six steps

Worth setting expectations before the steps, because the honest answer is that AI helps in exactly two of them.

AI is genuinely useful in step 4 (clustering) and step 6 (capture). Reading eighty notes and proposing groupings is mechanical, tedious, and something a model does well. Drafting the summary of what was decided is the same kind of work.

AI is actively harmful in step 3 (generation). Asking a model to add ideas to a live brainstorm floods the board with plausible, generic material that buries the two or three genuinely odd contributions that made the session worth running. The value of a brainstorm is the weird idea from the person who knows the customer, and that idea gets lost in a pile of AI-generated adjacent phrasing.

The practical constraint: most tools cannot see the board. If the AI only reads what you paste into a side panel, it is not clustering your session, it is clustering a summary you already made, which is the hard part done by hand. A canvas AI that reads the actual board, including where things sit, is doing something categorically different. That is the specific thing Storyflow was built for, and it is also the thing to check before believing any tool's clustering claim.

1. Write the question before you open the board

The single highest-leverage thing you do happens before the session exists.

A brainstorm needs one question, narrow enough to answer and open enough to have many answers. "How do we improve onboarding" is too broad to produce anything but platitudes. "What could we remove from the first five minutes of onboarding" is answerable and still has fifty answers.

Write the question at the top of the board, large, so it is visible at every zoom level. It is the thing people drift from, and a visible question pulls them back without the facilitator having to say anything.

Then write, directly under it, two more things:

• What we will decide by the end. For example: "three candidates to prototype next sprint". Sessions without a stated output produce notes; sessions with one produce decisions.
• What is out of scope. One line. This prevents the most common derailment, which is the session sliding into an adjacent problem that happens to be more fun.

If you cannot write the question in one sentence, the session is not ready. That is not a formality, it is the most reliable predictor of whether the hour will be worth it.

2. Build the board before anyone arrives

Never open a blank canvas to a group. Build four zones, labeled, sized, and positioned, at least ten minutes before the session.

Two details that matter more than they look.

Give each participant their own labeled generation lane. A single shared pile produces pile-up anxiety, where people read what exists and unconsciously conform. Individual lanes mean everyone writes from their own head first. You merge the lanes in step 4.

Put real context on the board, not in a pre-read. Three customer quotes, the actual funnel numbers, a screenshot of the current flow. People generate from what they can see. A pre-read nobody opened is not context, and asking a group to brainstorm from memory produces the ideas they already had.

Keep the convergence zone off-screen or visually collapsed. Visible evaluation criteria change what people are willing to write down.

3. Generate in silence, on a timer

This is the step people skip and it is the one that makes the difference.

Run the first generation round completely silently. No talking, no reacting, no reading other lanes. Everyone writes into their own lane for a fixed period, seven or eight minutes is usually right. Put a visible timer on the board.

Silent generation is not a gimmick. It fixes three specific failures at once:

• Anchoring. In a spoken brainstorm, the first idea framed aloud shapes everything after it. Silence means fourteen independent starting points instead of one.
• Production blocking. In a spoken session only one person can talk at a time, so ideas are lost while waiting. Parallel writing removes the queue entirely.
• Status effects. The most senior or most confident voice dominates a spoken room. Nobody knows who wrote what during a silent round, so contributions compete on content.

One instruction to give explicitly: quantity, not quality. Tell people you want the bad ideas too, and that nobody will be judged on their lane. The wide field is the point, and the odd idea that turns out to matter almost never arrives in the first three notes someone writes.

Then run a second, shorter round after a break where people read everyone else's lane. This one is for building on what others wrote. Keep it silent as well. The two-round structure gets you independence first and cross-pollination second, in that order, which is the order that works.

4. Cluster together, out loud

Now open the floor.

Move all notes into the shared clustering area and group them by similarity, as a group, while talking. This is the first spoken step and it is genuinely collaborative: the discussion about whether two notes belong together is where the actual thinking happens.

Rules that keep it productive:

• Name every cluster. An unnamed cluster is a pile. The act of naming forces the group to articulate what the theme actually is, and it frequently splits one cluster into two.
• Keep an "orphans" area. Some notes belong to nothing. Do not force them. Orphans are disproportionately where the interesting ideas live, precisely because they did not fit the existing frame.
• Do not delete anything yet. Deletion is judging, and judging is step 5.

This is where an AI that can read the board earns its place. Asking it to propose an initial grouping of eighty notes gives the group something to react to, which is far faster than starting from scratch, and reacting to a wrong grouping ("no, those two are not the same thing") produces sharper articulation than building up from nothing. Treat its output as a first draft to argue with, never as the answer.

5. Converge with a visible method

Groups are bad at converging without a mechanism, and the default mechanism, whoever argues longest, is the worst one.

Use dot voting. Give everyone a fixed small number of votes, three or five, and have them place dots on clusters, not on individual notes. Voting on clusters keeps the conversation at the level of themes.

Two refinements worth the extra two minutes:

• Vote silently and simultaneously. If people vote in sequence, the early dots bias the later ones exactly the way early ideas did in step 3.
• Give the decision-maker a separate marker. Most real teams are not democracies. Pretending the vote is binding and then overriding it afterward is worse than being honest that the vote is input. A distinct marker for "the person accountable picks this" keeps the process legible.

Then take the top clusters and write, for each, a single sentence: what it is, and what the next concrete action would be. If a cluster cannot survive being written as one sentence with an action, it was a mood rather than an idea.

6. Capture the decision before the call ends

The board is not the output. The decision is the output, and it has to be written while people are still present.

In the last five minutes, write directly onto the convergence zone:

• The chosen items, one line each
• The owner for each
• The date something happens
• What was explicitly rejected, and why

That last one is the most valuable and the most skipped. Six weeks later somebody will propose the rejected idea again, and a one-line record of why it was rejected saves the whole discussion from being re-run. It is also the part a good AI summary can genuinely draft from the board, since the reasoning is right there in the clusters.

Then do two things to stop the board rotting:

Freeze it. Rename the board with the date and the decision, and stop editing. A board that keeps being edited stops being a record of what was decided.

Move the actions off the board. Tasks belong in whatever tool the team actually uses. A canvas is a terrible task tracker, and actions left as sticky notes are actions nobody does.

Common mistakes I see

• Opening a blank board. The single most common failure, and it is entirely preventable with ten minutes of prep. Structure is not decoration, it is what makes contribution possible.
• Starting with discussion. Once the framing has been said aloud, independent ideas are gone. Silent generation must come first.
• Inviting too many people. Above about twelve, generation still works but clustering collapses into chaos. Split into two sessions and merge the clusters rather than running one unmanageable room.
• No timer. Untimed generation produces a slow drift into conversation. The timer is what makes silence feel purposeful rather than awkward.
• Treating the vote as the decision. If someone is accountable, say so before the vote instead of quietly overriding it afterward.
• Letting AI generate the ideas. It buries the few genuinely original contributions under a mass of plausible filler. Keep it out of step 3.
• Leaving the board as the record. Nobody re-opens a canvas of eighty notes. The four-line decision block is what people actually read.

How this went for a positioning session

A team of nine, spread over three time zones, trying to decide how to describe a product that had drifted from its original pitch. Previous attempt: a two-hour call that produced a document nobody agreed with.

The rerun took seventy minutes. The question on the board was "what does someone believe after thirty seconds on our homepage, and what should they believe". Context zone had six real session recordings and the current homepage. Nine labeled lanes.

Silent round one produced 104 notes in eight minutes, which is roughly twelve per person and far more than the spoken session had produced in two hours. Round two, after reading, added another 40, and those were noticeably sharper because they were reacting to real material.

Clustering out loud took twenty-five minutes and produced eleven named clusters plus a fourteen-note orphan pile. The orphan pile is where the session paid for itself: four of those notes were the same observation from different people, that the product was being described by what it replaced rather than by what it did. That became the twelfth cluster and won the vote.

Decision block, written in the last four minutes: one positioning line, an owner, a date, and three rejected directions with one-line reasons. The rejected list has been pointed at twice since.

The tools you will actually use

You need one canvas and a timer, and honestly that is it. Where the tools differ:

• For live facilitation with a large group, Miro is the strongest option and has the deepest workshop feature set: built-in voting, timers, and facilitation controls that keep a big room in sync. If you run workshops as a job, this is the default.
• For a team already inside Figma, FigJam sits next to the design work and the voting and timer tools are good enough for most sessions.
• For a session whose output feeds ongoing creative work, a canvas that holds the board alongside the moodboards, references, and project plans is more useful than a dedicated workshop tool, because the decision does not have to be exported to be used. This is where Storyflow fits: it is free to start with unlimited boards and unlimited collaborators, paid plans begin at 7.99 dollars a month on annual billing, and it is priced per account rather than per seat, so inviting nine people to one session costs nothing. Its AI reads the whole board, which is what makes the step 4 clustering pass and the step 6 summary work on the actual session rather than on a pasted excerpt.
• For a quick three-person session, any of them, or a shared document with headings. Do not over-tool a small conversation.

The honest summary: for pure real-time workshop facilitation at scale, Miro is better than we are. For a brainstorm whose output has to live on inside a larger creative project, keeping it on the same canvas as the rest of the work is worth more than the facilitation features.

You are ready

The structure is the whole thing. A brainstorm on an online whiteboard is not a call with a canvas open, it is a designed sequence: one question, a board built in advance, independent silent generation, collaborative clustering, a visible convergence method, and a decision written down before anyone leaves.

Build the board before the session. Keep generating and judging apart. Name every cluster. Write the rejections down.

Do that and the format stops being a way to fill an hour and becomes the fastest way a distributed team can go from a vague problem to three things worth trying.

Author

Justkay is a documentary filmmaker and the founder of Storyflow. He has run planning and positioning sessions for distributed creative teams for years, and built Storyflow after watching the output of good sessions die in boards nobody reopened.