The short answer
The best AI prompts for brainstorming force variety by naming directions in advance, ban the obvious answers, and feed the model context only you have, such as customer quotes and hard constraints. Claude from Anthropic and ChatGPT from OpenAI are the strongest brainstorming partners, and both have free plans. Use Perplexity to check whether an idea already exists, because it cites every answer. The 13 prompts below cover generating, pushing further, naming, testing and group sessions.
What AI is genuinely good at in brainstorming
AI assistants are tireless at quantity and fast at variety, which are the two things human brainstorms run out of first. They are weaker at knowing what is new to your market, what your customers actually care about and what your company would never approve. So the best brainstorming prompts do two things: they force the model away from its most predictable answers, and they feed it the context only you have, such as the customer's own words, your constraints and your criteria for a good idea. Without that, you get the same 20 ideas everyone else using ChatGPT, Claude or Gemini gets.
The 13 prompts below follow a sensible order: generate options, push past the obvious, work on names and messages, test the ideas, and run a session with real people. Replace everything in [square brackets] before you run them.
Prompts for generating lots of options
Generate ideas in deliberately different directions names the directions in advance, because left alone a model tends to produce twenty versions of one idea. Brainstorm from the customer's problem, not our product starts from a frustration, ideally quoted from a call, survey or review, which keeps ideas useful. Get quantity first, then cluster separates generating from judging: forty rough ideas, then themes. The clustering step is where you spot a direction worth checking: the model cannot see your market, so confirm the least-explored theme with a quick cited search before you build on it. Ask for one line per idea; long explanations slow you down and make weak ideas sound convincing.
Prompts for pushing past the obvious
Kill the obvious answers asks the model to list the predictable ideas and then ban them, which is the most reliable way to get something new out of any assistant. Borrow from another industry works because the model knows how airlines, games or hotels solved similar problems, and transferring those patterns is a well-known creative technique. Flip the constraints forces unusual limits on the problem; the ideas that survive once the limit is removed are often the best ones. If you want truly different perspectives, run the same prompt in two assistants: models from different makers have different habits. Perplexity's Model Council can run one question through several models side by side. Perplexity's help centre lists it in Search for the Max and Enterprise Max plans (its pricing page also lists it under Pro), and inside Perplexity Computer it is on all paid plans and uses credits.
Prompts for names and messaging
Brainstorm names with reasons mixes naming styles and asks for the reasoning, so you can steer the next round. It tells the model not to claim availability, because AI cannot reliably check trademarks or domains; do that yourself through official registries before you fall in love with a name. Find the angle for a message gives you ten distinct ways into the same news, each with a headline and first sentence, which is enough to test with colleagues or a small audience. For campaign and copy work, see the best AI for marketing and the best AI for copywriting.
Prompts for testing ideas before you commit
Score ideas against our criteria makes the judgement explicit and then rescues one low scorer for a small test, because scoring systems punish unusual ideas. Find the fatal flaw asks for the cheapest test that would kill each idea, which keeps experiments small. Build the idea out one level turns a one-liner into a page people can react to. For any idea that depends on facts about the market, check those facts in a tool that cites sources, such as Perplexity, before you present it. When the testing turns into a real choice between options, our decision-making prompts take over.
Prompts for running a brainstorm with a group
Design a 45-minute brainstorm session builds in a silent round and a voting method, because open discussion tends to follow the most senior or loudest person. Synthesise a messy session turns sticky notes and scribbles into themes while keeping the original wording of the most distinctive ideas, and it deliberately saves the odd idea nobody followed up. If the session ran on a call, transcribe it first; our meeting notes prompts show how to clean up a transcript.
How to adapt these brainstorming prompts
- Give real context. Paste customer quotes, your strategy in two sentences and your hard constraints. Generic input produces generic ideas.
- Ask for one line per idea first. Explanations come later, for the shortlist only.
- Iterate on the best five. "Take ideas 3, 7 and 12 further, and combine 7 with 12" gets better results than starting again.
- Change the persona to change the output. "Answer as a sceptical CFO" or "as a first-time customer" shifts which ideas the model favours.
- Keep unannounced plans out of consumer plans. Personal plans of ChatGPT and Claude can use chats for training unless you switch it off, and Gemini does while Keep Activity is on; use temporary or incognito chats, or your company's work plan, for confidential launches.
Which AI tool works best for brainstorming, and why
Claude and ChatGPT are the strongest all-round brainstorming partners: both are good at varied, specific ideas and at building on your feedback in a long conversation, and both have usable free plans. Gemini is a good choice if your source material lives in Google Docs and Drive. Use Perplexity alongside them to check whether an idea already exists and what competitors are doing, because it searches the web and cites every answer. For a picture of how these combine, see ChatGPT vs Claude, Claude vs Gemini and our tool stacks, or start with the free lessons at /learn?tool=claude and /learn?tool=chatgpt. Tool guides: Claude, ChatGPT and Gemini.
Prompts to copy
Generate ideas in deliberately different directions
We need ideas for [challenge] for [audience]. Constraints: [budget, time, brand or legal limits]. Give me 20 ideas in five clearly different directions (for example cheap and fast, premium, partnership, digital-only, something we would normally never do), four per direction. One line each. No explanations yet.
You get: Twenty ideas spread across directions instead of twenty variations of one
Brainstorm from the customer's problem, not our product
Our customer is [describe customer and role]. Their biggest frustration with [area] is [problem, ideally in their own words]. List 15 ways we could reduce that frustration, including ideas that do not involve our current product. Mark the three that a customer would notice within a week.
You get: Ideas anchored in a real customer problem rather than in what you already sell
Get quantity first, then cluster
Give me 40 rough ideas for [topic], one line each, numbered. Include obvious ones and strange ones. Then group all 40 into five to seven themes, name each theme, and say which theme looks least explored by competitors in [market], with two searches I could run to check that.
You get: A long list sorted into themes so the patterns become visible
Kill the obvious answers
What are the ten ideas most people in [industry] would come up with for [challenge]? List them briefly. Now ban all ten and give me ten more that are clearly different, and explain in one line what makes each one non-obvious.
You get: A second list that starts where everyone else stops
Borrow from another industry
How have companies in [three unrelated industries, e.g. airlines, gaming, hospitality] solved a problem similar to [our problem]? Give two examples from each with what they did, then adapt each example into a concrete idea for [our company or team].
You get: Transferable ideas borrowed from fields your competitors are not watching
Flip the constraints
Our challenge is [challenge]. Generate ideas under each of these extreme constraints: zero budget; it must launch in 48 hours; it must work without any software; the budget is ten times bigger; our biggest competitor must benefit too. Three ideas per constraint, then tell me which ones still work without the constraint.
You get: Ideas that come from forcing unusual limits on the problem
Brainstorm names with reasons
Suggest 25 names for [product, project or event] that is [one-sentence description] for [audience]. Mix descriptive, invented, metaphor and compound names. For each give the reasoning in a few words. Avoid names that are hard to spell after hearing them once. Do not claim any name is available; I will check trademarks and domains myself.
You get: A varied shortlist of names with the thinking behind each
Find the angle for a message
We need to announce [news] to [audience]. Give me ten different angles (for example the problem it solves, the number behind it, the story of one customer, what changes on Monday, the myth it busts), each with a headline of under ten words and a first sentence.
You get: Ten distinct ways into the same story, ready to test
Score ideas against our criteria
Here are our candidate ideas: [paste list]. Score each from 1 to 5 on [criteria, e.g. customer impact, cost, speed to launch, fit with strategy], explain each score in a few words, and show the results as a table sorted by total. Then tell me which low-scoring idea might still be worth a small test and why.
You get: A scored, sorted shortlist with a reasoned wildcard
Find the fatal flaw
For each of these ideas [paste top three to five ideas], describe the single most likely reason it would fail, the cheapest test that would reveal that flaw within two weeks, and what result would make us drop it.
You get: A quick test plan that kills weak ideas cheaply
Build the idea out one level
Take this idea: [idea]. Describe what it looks like in practice: who does what in the first 30 days, what it costs in time and money (as ranges), what we would measure, and the three questions we would need answered before committing. Keep it to one page.
You get: A one-page sketch that turns an idea into something people can judge
Design a 45-minute brainstorm session
Design a 45-minute brainstorming session for [number] people from [teams] on [challenge]. Include a warm-up, a silent idea-generation round, a round where people build on each other's ideas, a voting method that avoids the loudest voice winning, and the output we leave with. Give exact timings and the prompts the facilitator should say aloud.
You get: A ready-to-run session plan with timings and facilitator lines
Synthesise a messy session
Here are the raw notes and sticky notes from our brainstorm [paste the notes, or attach photos of the whiteboard and sticky notes]. Remove duplicates, group the ideas into themes, keep the original wording of the most distinctive ideas, and list the three ideas with the most support plus the one unusual idea nobody followed up.
You get: A clean record of the session that keeps the best odd idea alive