How to use these prompts
The trap is stopping at the first suggestions: the AI starts from the most common associations, the ones anyone would think of. Good ideas arrive when you force it to change angle or impose an absurd constraint on it. All the prompts below aim there: they don't ask "give me ideas," they ask "give me ideas from a perspective I wouldn't consider."
A practical rule: the AI generates, you select. Don't expect it to hand you the right idea; expect it to give you the raw material your right idea is born from, often by combining two mediocre suggestions.
The prompt library
Generate ideas, then dig beneath the obvious
Give me 20 ideas on this topic. The first 5 will be obvious, I know:
include them but then label them "obvious" and go further. The last 5
have to be strange, uncomfortable, or counterintuitive, even if risky.
For each idea in the last group, one line on why it might work precisely
because no one does it.
Topic:
Find a name
Suggest 15 names for this. Split them into three groups: descriptive
(you can tell what it does), evocative (they create an image or an
emotion), invented (new words). For each, one line on why. Avoid names
already overused in the field and tell me if any risks being already
taken or hard to pronounce.
What I need to name, who it's for, and the impression I want to give:
Change perspective to get unstuck
I'm stuck on this problem and I keep circling the same ideas.
Tackle it from five completely different perspectives: how a child
would solve it, an engineer, an artist, an accountant, a con artist.
One concrete suggestion for each. I want to get out of my rut, not
get reasonable answers.
The problem I'm stuck on:
Combine distant elements
Generate ideas for me by crossing these two worlds that usually don't
touch. Force the pairing even if it seems absurd: the best ideas are
born where two distant things collide. Give me 10, from the most
sensible to the most insane.
World 1 / World 2:
Stress-test an idea you already have
I already have an idea and I want to see if it holds up. Play devil's
advocate: list the three most serious reasons it could fail, who might
oppose me and why, and what I'm taking for granted. Then, only after
that, tell me how to strengthen it. Don't praise it: put it to the test.
My idea:
A real example
Sofia is opening a small gluten-free bakery and needs a name for it. She starts by asking for generic ideas and gets the usual things with "free" and "nature" in them, all already seen. She switches to the names prompt, specifying "gluten-free bakery, young clientele, I want a name that doesn't scream 'diet' but makes you hungry."
The AI gives her three groups. Among the evocative ones "Rise" pops up, playing on rising and growing without naming the absence of gluten. Sofia crosses it with the stress-test prompt: the devil's advocate points out that the name doesn't communicate "gluten-free," a risk for anyone looking for exactly that. She solves it by adding a tagline beneath the sign, not by changing the name. The AI didn't hand her the name ready-made: it helped her find it by showing her twenty paths and then attacking the one she'd chosen.
When it does NOT work (and how to fix it)
If all the ideas resemble each other
When the suggestions are all variations of the same thing, the prompt was too neutral. Force diversity: "each idea has to use a different mechanism from the others, no variations of the same concept." Or use the perspectives prompt, which forces the AI to change its point of observation.
If the ideas are original but unworkable
Sometimes it pushes originality to the point of uselessness. Bring them back down to earth: "now take the three strangest ones and make them feasible with the resources I have, tell me the first concrete step of each." A wild idea becomes useful when you force it to face reality.
If the proposed names are already used by others
The AI doesn't have an up-to-date register of existing trademarks: it can suggest a name that's already taken. Before you get attached, check yourself with a search and an availability check (domain, any registered trademark). The AI gives you the candidates; the legal and availability check is on you, before you print the signs.
A tip from someone who actually uses it
Keep your discarded ideas. When you brainstorm with the AI, save the suggestions you don't use too, in a document: today's fourth useless idea becomes the key to a problem three months from now. And never ask for just one round: generate, pick the two that intrigue you, and go again asking for ten variations of just those. Good brainstorming is a funnel, not a single shot.
Frequently asked questions
Are the AI's ideas really original or just reshuffled?
They're recombinations of existing things, like most human ideas, for that matter. The originality isn't in the single suggestion, it's in the pairing you make by choosing and combining. The AI widens the field of possibilities; the creative leap, recognizing which combination is worth it, stays an act of yours.
Can I use AI for brand and product names to register?
To generate candidates yes, to decide no, not without checks. The AI doesn't know which trademarks are already registered nor what a name means in another language. Take its suggestions, then verify the trademark's availability, the domain, and that the name has no unpleasant meanings where you sell. The creativity is its, the due diligence is yours.
The more ideas I ask for, the better?
Up to a point, then no. Asking for a hundred ideas produces ninety variations of the same one and buries you. Better twenty ideas with the constraint "the last five must be uncomfortable" than a hundred neutral ones. Quantity is only useful if you use it to go beyond the obvious, not to pile up. After the first twenty, it's better to dig deep into two than to widen to a hundred.