Try this before you read any further: open ChatGPT, paste in something you're working on, and ask for ten ideas. A few might be usable, several will likely overlap, and most will feel familiar, because a cold request tends to produce the same safe ideas everyone else gets. The fix isn't a cleverer prompt formula. It's two moves: make ChatGPT interview you before it generates anything, and treat its first list as raw material instead of a deliverable.
Why the First List Is So Generic
Ideas are only as specific as the information behind them. Ask for "ideas for a newsletter" and ChatGPT has to guess at your audience, your voice, what you've already tried. It fills those gaps with the average of everything it has read — and the average of everything is what you get when anyone asks the same question.
In our first video we argued that AI output comes down to two ingredients: memory (what the AI knows about your situation) and intelligence (how well the model reasons). A cold brainstorm leaves the memory side empty, and more intelligence can't fill a blank page about you.
Research points the same way: a 2024 Science Advances study found AI-assisted story ideas rated as more creative — with the biggest gains for less creative writers — but also more similar to each other (Doshi & Hauser, 2024). AI raises the floor and narrows the range. Both moves below steer away from that center.
The Interview Method
Reverse the direction of the conversation: instead of you prompting ChatGPT, it interviews you. We demonstrated this reverse-prompting technique in our first video for brand images; it works the same for ideas. Paste this:
I want to brainstorm [topic] with you. Don't give me any ideas yet. Interview me first, one question at a time, until you understand my goal, my constraints, what I've already tried, and what a good outcome looks like. Tell me when you have enough context, and only then start suggesting ideas.
Expect five to eight questions — constraints you forgot to mention, which past attempts you secretly disliked. Answer out loud (microphone icon): talking surfaces specifics that typing flattens. The interview works because your answers contain the material that moves ideas off-center. "We tried that in March and it flopped" can't be inferred from a one-line prompt — someone has to say it.
"I'll know it when I see it" isn't context, either. Say which past attempts you liked and why; vague standards produce vague ideas.
The difference shows in the output. A cold newsletter brainstorm produces "weekly roundup, founder story, five tools worth trying." After an interview that surfaced "readers are experienced freelancers who skip generic productivity advice," it produces a monthly teardown of one expensive mistake from a past project. That's not a magic phrase — it's the context the interview surfaced.
Work the List
Don't pick from the first list — work it. Each move adds memory:
- Ask for its own ranking: "Which of these is the most surprising, and why?"
- Push one further: "Take idea 4 and give me three variations, each one stranger."
- Attack one: "Which of these would fail first, and what would that teach us?"
- Combine two: "Merge ideas 2 and 5 into one option."
- Argue. Your objections are context — give them, then ask it to regenerate with that in mind.
Fifty ideas in thirty seconds feels productive. The value arrives in the two or three you push on, so pick them before you close the chat. By round three, ideas are coming out of a conversation the two of you built rather than a sentence you typed.
Bring Your Worst Idea
The interview pulls context out of you; this move pushes it in. Hand over a half-formed idea — even one you suspect is weak — and have ChatGPT stress-test it:
My half-formed idea: [idea]. Stress-test it. Tell me what's interesting here, what's weak, and what a stronger version might look like. Don't be polite.
Asking "what's interesting here" directs ChatGPT at the part that makes the idea yours, instead of grading the whole thing against safe alternatives. One honest warning: ChatGPT will help you develop an idea whether or not it deserves it. "Don't be polite" helps and doesn't cure — the judgment about which ideas are worth building stays with you.
FAQ
What is reverse prompting?
Asking the AI to interview you about a task until it has enough context to do the task well — instead of writing one long prompt. It works anywhere output quality depends on context only you have: brainstorming, briefs, planning, images.
Is brainstorming with ChatGPT cheating?
For your own work — product names, content, business plans — it's a tool, the same way asking colleagues is. For a class assignment or contest where ideas must be your own, having ChatGPT generate the core idea may break the rules, so check them. Using it to stress-test and develop an idea you brought yourself is the use we recommend in any case.
Want to go further? How to write better ChatGPT prompts without prompt engineering covers the conversation-first approach, and what ChatGPT can do for beginners covers more uses like this one.