How to Use Claude for Brainstorming

Most guides on how to use Claude for brainstorming are wrong about what actually makes it useful. They treat volume as the goal — "generate 50 ideas" — when the real bottleneck in brainstorming was never idea count. It's genuine divergence. Fifty ideas that are all variations on the same underlying concept aren't more useful than five, they just take longer to read through and reject.

The Problem the Product Team Faced

A five-person product team at an early-stage habit-tracking app needed new feature ideas for their Q3 roadmap. Their product lead asked Claude for "50 feature ideas for a habit tracking app" and got back a list that, once she actually read through it, broke down into maybe six distinct concepts repeated with minor variations — streak counters described eight different ways, reminder notifications described six different ways, and so on. The volume was real. The genuine variety wasn't.

The Wrong Approach

Here's roughly what she'd asked for:

Generate 50 feature ideas for a habit tracking app aimed at people 
trying to build better daily routines.

What this does: it asks for quantity without giving Claude any axis along which to diverge, so it fills the request by rephrasing a smaller set of default concepts rather than genuinely exploring different directions. Without constraints forcing variety, the path of least resistance is repetition dressed up as variation.

⚠️ Common mistake: assuming a bigger number in the prompt produces broader thinking. It usually just produces more restatements of the same handful of ideas, padded out to hit the count.

The Correct Prompt

Here's what the team rebuilt it into:

I need feature ideas for a habit tracking app. Instead of one long 
list, generate ideas across five distinct approaches, 4 ideas each:

1. Social accountability (features involving other people)
2. Financial stakes (features tying money to habit completion)
3. Gamification (features using points, levels, or competition)
4. Data/insight-driven (features that surface patterns in the user's 
   own behavior)
5. Friction reduction (features that make the desired habit easier, 
   not features that track it better)

For each idea, name one habit-tracking app that already does something 
similar, if one exists, and one way to differentiate from it.

What this does: naming five genuinely different mechanisms up front forces Claude to generate within each lane separately, rather than drawing repeatedly from one default well of ideas, and the competitor-naming requirement filters out ideas too generic to differentiate.

⚡ Pro tip: define your divergence axes before asking for ideas, not after. If you don't know what "different" should mean for your specific brainstorm, Claude can't diverge along a dimension you haven't specified — it'll default to surface-level rephrasing instead.

Results and What Changed

The rebuilt prompt produced 20 ideas instead of 50, but her team kept meaningfully more of them for further discussion — 11 out of 20, versus roughly 6 out of 50 from the original list once duplicates were filtered out. The financial-stakes category surfaced an idea (a small refundable deposit tied to a weekly habit streak) that became one of two features actually built that quarter.

Honestly, the team was skeptical about asking for fewer total ideas at first — it felt like a step backward from "more options." But the filtering time saved more than made up for the smaller raw count, and the ideas that survived filtering were stronger because they'd been forced to differentiate against a named category and a named competitor from the start.

How to Apply This to Your Situation

1. Name your divergence axes before generating anything. What are the genuinely different directions this idea could go? If you can't name three to five, spend time on that first — it's the most valuable part of the exercise.
2. Cap ideas per axis rather than total ideas. Four ideas within five distinct categories beats twenty ideas in one undifferentiated pile.
3. Require differentiation from an existing example, where relevant. This filters out ideas too generic to be useful before you've spent time reviewing them.
4. Resist the urge to ask for a bigger number. If a brainstorm feels thin, the fix is usually adding a new axis, not asking for more ideas along the existing ones.

⚠️ Common mistake: brainstorming without ruling anything out first. If certain approaches are off-limits (budget, timeline, technical constraints), say so up front — otherwise you'll spend review time filtering out ideas that were never viable rather than comparing ideas that actually could have shipped.

It's also worth revisiting your axes mid-session if the first pass feels thin in one category and rich in another. That's a signal the axis itself might be poorly defined, not that the category has nothing to offer. The habit-tracking team found their "gamification" axis kept producing weak ideas until they split it into two narrower ideas — competitive gamification (versus other users) and personal gamification (versus your own past performance) — at which point both halves produced ideas worth keeping. A thin category is usually a badly drawn line, not an empty well.

Next Steps

Before your next brainstorming session, spend five minutes defining three to five genuinely distinct directions the idea could take, instead of jumping straight to "generate ideas about X." That five minutes of framing work does more for output quality than doubling the requested idea count ever will.

If you land on a set of divergence axes that works well for a recurring type of brainstorm — feature ideas, marketing angles, event themes — save that structure in PromptABCD so your next session starts from a framework that's already proven to surface real variety, not a blank "give me ideas" prompt.