The hardest part of ideation isn’t “coming up with a good one” — it’s coming up with any at all, staring at a blank while the same two or three clichés loop in your head. Human brains aren’t built to diverge: the moment an idea appears, we start picking at its flaws, killing it before it can spread.
AI is the opposite — it’s great at churning out volume. The trick is to split diverging from filtering: first have it dump 20 directions with zero judgment, even if half are junk — that’s the point, because a good idea is often sparked by a bad one. Then you play the critic, pulling out the few that work, combining and reshaping them.
When to use it
Naming a project, picking a campaign theme, finding content angles, brainstorming titles or taglines — any time you need a pile of options to choose from.
How to do it
- Frame the topic clearly: what you need ideas for, who they’re for, and any hard limits (budget, platform, style)
- Paste the prompt and ask for 20 at once, “as divergent as possible, no repeating angles, don’t judge them yet”
- Skim quickly and mark the few that catch your eye (go with gut, don’t overthink)
- Have it “expand each in two lines / combine them / generate 5 more in this direction” to narrow toward a usable plan
Weak vs strong
The left likely returns three to five obvious ones; the right pins down “20 at once + categorized + no judging yet,” giving you a full menu to choose from.
Copy-paste prompt
I need ideas for【what you’re ideating on, e.g. a theme for a product launch】, audience:【who】, constraints:【budget / platform / style / must-haves】. Give me 20 ideas from different angles in one go, as divergent and non-overlapping as possible, each with a one-line rationale. Don’t filter them for me yet — just give me the volume.
Worked examples
You get：This two-step — pile up many, then expand a few — beats demanding “the single best one” upfront: you get both breadth and depth.
You get：Out of 20 titles you can usually use 3–4 outright, then remix their patterns — titles solved.
Level up
- Force a new angle: if ideas feel safe, add “these are too ordinary — give 10 bolder / counterintuitive / unexpected ones”
- Constrain the form: ask it to “frame each as a ‘what if…’ ” or “start each from a user pain point” for directed divergence
- Categorize then diverge: have it “list the big directions to attack this,” pick one, then ask for 10 within that direction
Common mistakes
- Judging while generating — in the diverge phase don’t nitpick; get all 20 first, filtering is the next step
- Too-vague topic — “give me ideas” yields generic ones; spell out audience, limits and tone you want
- Stopping at the pick — AI gives half-baked directions; always rework them and add your real details before using
FAQ
A lot of its ideas are pretty plain — is this useless?
Could the ideas have factual errors or be impractical?
Pro tip：Make “no judging, 20 at once” your fixed opener — let AI spread the options and you do the choosing; with the split right, inspiration stops being mysterious.