How to brief an AI whiteboard so the first draft is usable

The first time most people try an AI whiteboard, they type "brainstorm ideas for our launch" and get a board full of stickies that could belong to any company on earth. "Run a social media campaign." "Partner with influencers." "Collect customer feedback." They delete the lot and decide the AI is useless.

The model did exactly what it was asked. The prompt named no product, no audience, no format and no constraints, so the draft came back as an average of every launch ever written about. You'd get the same result from a freelancer you briefed in one sentence on their first day.

A good prompt for a whiteboard takes about thirty seconds longer to write. Here's what goes into one, with examples from Prompt Board's AI whiteboard.

What the AI does and what the code does

It helps to know what happens after you press enter. On Prompt Board the model never draws anything. It returns a plan. In board mode the plan is a list of sections, each with a title, a colour and some sticky-note text. In diagram mode it's a list of nodes, each marked as a start, end, step, decision or note, plus the edges that join them.

Code then does the layout. Sections are packed into a grid of frames, and diagrams get a layered layout where each step sits a row below the step that leads to it. The model never picks a coordinate, so a sticky can't land on top of another one or spill out of its frame.

That has two practical consequences for your prompt. First, don't spend words on layout. "Put the risks on the left" does nothing, because the model doesn't control position. Spend those words on content instead. Second, there are size limits. A board-mode draft holds up to six sections of up to eight notes each, and a diagram holds up to 24 nodes. Ask for fifteen sections and the draft gets cut. Short notes work best, and the model is told to write each one as a phrase or a short sentence.

Pick board mode or diagram mode first

The mode decides the shape of the output, so choose it before you write anything else.

Use board mode when the content is a set of ideas sorted into groups. Retros, brainstorms, SWOT analyses, journey maps and lean canvases all fit here. You get frames full of stickies.

Use diagram mode when order matters and there are decisions. A support process, a sign-up flow or an approval chain belongs here. You get shapes joined by arrows, with yes and no branches on decisions. The flowchart template shows what a finished one looks like.

If you can't decide, ask yourself whether someone will read it top to bottom or scan it in any order. Top to bottom means diagram.

Five things a strong prompt names

The format. Say "retro", "SWOT", "user journey map" or "flowchart". A format word carries a whole structure with it, and the model knows what a SWOT has in it without being told.

The audience. "For a six-person mobile team" or "for the board of a regional bakery chain" changes the vocabulary and the level of detail in every note.

The sections you want. If you already know the columns, list them. "Sections: went well, to improve, ideas, action items" gets you those four frames in that order. Leave sections out only when you want the model to propose them.

Constraints. Note length, the number of sections, the time period, things to leave out. "Keep notes under eight words" works. So does "no generic advice, only things specific to a B2B tool".

Real nouns. This one makes the most difference. Name the product, the city, the tool, the bug, the deadline. "The App Store review delay" and "the flaky login tests" give the model something to work with. "Some technical issues" doesn't.

Weak and strong prompts side by side

A retro:

Weak:   retro

Strong: Sprint retro for a six-person mobile team after a release
        that slipped a week. Sections: went well, to improve, ideas,
        action items. Mention the App Store review delay and the
        flaky login tests. Action items as "Name: task by day".

A journey map:

Weak:   customer journey

Strong: User journey map for a first-time buyer at a bike shop with
        an online store, from searching "commuter bike Leipzig" to the
        first free service. Sections: search, compare, buy, pick up,
        first service. Notes under ten words, written from the
        buyer's point of view.

A flowchart, in diagram mode:

Weak:   support flowchart

Strong: How a support ticket is handled at a 10-person SaaS company.
        Start: ticket arrives. Steps: tag product area, reproduce the
        bug. Decisions: is it a known issue, can we fix it in under a
        day. End: ticket closed. Label decision branches yes and no.

None of the strong versions runs past five lines. Each one names the format, the people involved, the structure and two or three facts only your team knows. That's enough for a draft you edit instead of deleting.

One more detail. The model answers in the language you write in, so a German prompt gets you German stickies.

Edit after you generate

Even a strong prompt gives you a first draft. Give it five minutes before anyone else sees the board:

1. Delete the notes that are true of every team. You'll spot them at once.
2. Rewrite the rest in your own words, especially anything a colleague will read aloud.
3. Add what only you know. The model doesn't know that the design review moved to Mondays.
4. Rename frames whose titles are too broad.

Then share the link and let people add their own stickies. The AI draft is there to get a blank board moving. Once your team starts writing, their notes matter more than the model's.

You get 20 AI generations a day, which is plenty if you rewrite the prompt between attempts. Pressing generate on the same weak prompt three times gives you three versions of the same generic board.

Cluster and summarise once people have written

The two other AI actions come in after the humans have had their turn.

Clustering takes the stickies you select and groups them into themed frames. It's most useful after a brainstorm or a silent-writing round, when the board holds forty notes in no order. Check the groups afterwards. The model sorts by meaning, and sometimes it puts a sticky in a group your team wouldn't.

Summarising turns a selection into one sticky. Select the action items at the end of a retro and you get a short summary to paste into a chat or a ticket. It's quicker than writing the recap yourself, and you can edit the result like any other note.

When a template beats a prompt

If the format is fixed and the content comes entirely from the people in the room, skip the AI. A retro where the team writes every sticky is better started from the retrospective template than from a generated draft full of guesses. The same goes for a Kanban board you'll fill with real tickets.

Prompts win when you need content to react to. A first-pass SWOT for a client call starts faster from a sentence than from four empty frames, and so does a flowchart of a process nobody has drawn yet.

The model is only as specific as you are. Brief it the way you'd brief a sharp new colleague on their first morning, and the first draft becomes one you keep. Try it on the AI whiteboard.