What is the best way to use ChatGPT for brainstorming?
Ask for volume first, then ask for strange. The strongest result we found comes from a Wharton working paper by Lennart Meincke, Ethan Mollick and Christian Terwiesch. They tested 35 prompts on GPT-4 and found that asking it to list 100 ideas, make them bold and different, then describe them gave the widest spread of ideas.

Here's a shortened version of that method as a prompt you can paste. Swap the brackets for your own topic.
List 100 ideas for [your topic], one line each, with no explanations. Then rewrite the list so every idea is bolder and different from the others. Finally, describe the ten strangest ideas in two sentences each.
Those numbers come from one task: new products for college students priced under $50. The authors counted an idea as unique when it stayed under a cosine similarity of 0.8 to every earlier one. The authors dropped about 15 percent of chain-of-thought runs because the model skipped the second step. The test used GPT-4, not GPT-6, and we have not rerun it. Treat the lift as direction, and test the method on your own topic before you trust it.
One more finding from the same group explains why brainstorming with ChatGPT feels easy and still disappoints. In an earlier test by some of the same authors, ChatGPT-4 ideas were "on average of higher quality (as measured by purchase-intent surveys)" than ideas from students at an elite university. The weak spot showed up in the later paper: plain GPT-4 pools were more alike than a student group's.
Which seven brainstorming methods run in ChatGPT?
Seven methods cover most brainstorming jobs, and six of them need only a text box. The prompts below are written for this page. Replace the brackets, paste, and read the whole reply before you pick anything. All run on Free as of 15 November 2026 except the Voice limits shown in the plan section.
| Method | Use it when | Prompt to paste |
|---|---|---|
| 100 then bolder | You need range fast | List 100 ideas for [topic], one line each. Then make every idea bolder and different from the rest. Describe the ten strangest. |
| Reverse brainstorming | You are stuck on a problem | List 15 ways to make [goal] fail as badly as possible. Then turn each failure into a rule for how to succeed. |
| SCAMPER | You have an idea or product to improve | Take [idea] and apply SCAMPER: substitute, combine, adapt, modify, put to another use, eliminate, reverse. Give two variations per verb and flag the three least obvious. |
| Role panel | You want views you would not think of | Act as a panel of a skeptical accountant, a ten-year-old and a hardware engineer. Each gives three ideas for [topic] and one objection to another panelist's idea. |
| Constraint list | Every idea sounds the same | Give 10 ideas for [topic] that cost under [amount], need no new tools, and could start this week. Then give 10 that break one of those limits on purpose. |
| Voice | You think better out loud | Open Voice and say: I want to think through [topic]. Ask me one question at a time and push back when I give an easy answer. |
| Diagram | You want to see how ideas connect | Draw a mind map of [topic] with five branches and three sub-ideas each, as a diagram I can preview. |
Reverse brainstorming works because failure is easier to imagine than success. Asking how to ruin a launch produces specifics, such as no price on the page or a signup form with ten fields. Flip each one into a rule.
SCAMPER is a checklist of seven verbs, and it's a good fit for ChatGPT because the model never gets bored of the fifth verb. Ask for two variations per verb so the weak verbs still produce something. Skip any that don't fit.
A role panel gives the model something to disagree with. It also adds a little variety, though less than people expect. In the Wharton paper's abstract, telling GPT-4 to think like Steve Jobs moved cosine similarity only from 0.377 to 0.368. The objection line in the prompt does more work than the costume.
A constraint list fights the obvious. Money, time and tool limits cut off the first ten ideas everyone has. The second half of the prompt, where the limits break on purpose, shows you which rule was blocking the good idea.
How do you get past ChatGPT's generic first ideas?
Don't use the first batch. Treat ideas one to ten as the sample everyone else got, and mine what comes after. The cause is measurable. In the Wharton paper, cosine similarity for a student group was about 0.243, while GPT-4 pools ranged from 0.255 for the 100-then-bolder prompt to 0.432 for an earlier plain GPT-4 baseline, with this paper's base prompt at 0.377. A higher number means the ideas sit closer together.
Five habits close most of that gap, in the order we'd try them.
- Ask for 100, not 10, and throw away the first 30. The paper notes that the most common ideas tend to appear early in a run.
- Name the ideas to avoid, such as no apps, no subscriptions and no marketplaces. Banning the obvious forces a detour.
- Borrow from another field. Ask how a hospital, a bakery and a ferry company would each solve your problem.
- Start a fresh chat and ask again. A new chat with the same prompt returns a different pool, and you can compare the two.
- Add your own constraint from real life: your budget, your customers, your calendar.
Keep one rule above all of those. ChatGPT supplies options, and you supply taste. The Wharton authors measured diversity with a math score, not with judgment about which idea is good. No score picks your best idea for you.
If two chats open with the same ideas, the prompt is the problem. If you want to see whether another model breaks that pattern, run the prompt in a second model and read both lists side by side. Whizi, which publishes this page, does that on Pro: side by side model chats run one prompt in two columns, and Pro carries Claude, GPT and Gemini in one app. The first week is $1.99. The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, no card. Whizi does not run ChatGPT's own Voice or diagram preview, so use ChatGPT for those.
How do you brainstorm out loud with ChatGPT Voice?
Open Voice, tell ChatGPT to interview you, and talk. OpenAI's Voice help page, updated two days before we read it on 15 November 2026, gives three steps for the web (the last two lines below are our additions):
- Go to chatgpt.com and select the Voice icon in the prompt window.
- Allow your browser to use the microphone if it asks.
- Begin speaking to start the conversation. Use the microphone control to mute or unmute.
- Our addition: say Ask me one question at a time about [topic] and challenge my first answer each time.
- Our addition: switch to typing if you need to paste something. The help page says Live accepts text and images in the same chat.
On a phone, select the Voice icon in the message bar. The setting lives under Settings > Voice. Live, the current option, powers the conversation. Advanced and Standard voice are retiring on 11 December 2026, per the same page, and ChatGPT will move you to Live automatically on that date.
| Plan | Voice in Chat limit, rolling 24 hours |
|---|---|
| Free | Limited access to GPT-Live-1.1 Quip; limits may change |
| Go, $8 a month | 3 hours with GPT-Live-1.1 Quip |
| Plus, $20 a month | 3 hours with GPT-Live-1.1 |
| Pro, from $100 a month | 15 hours with GPT-Live-1.1 at the $100 tier |
| Pro, $200 a month | Unlimited, subject to abuse guardrails |
Three hours is plenty for brainstorming, since few people talk through ideas for that long. The catch is video. Live doesn't support video or screen sharing, so if you want to point your phone at a whiteboard, switch to Advanced before it retires.
Can ChatGPT still draw a mind map now that canvas is gone?
Yes, as a diagram, though not through canvas. OpenAI's release notes say canvas was removed from GPT-5.5 Instant and GPT-5.5 Thinking on 28 May 2026, and paid users could keep it only on legacy models for a limited time. Guides that tell you to open canvas and ask for a mind map describe the old setup, and the GPT-6 notes do not bring canvas back. Our ChatGPT canvas page covers what replaced it.
Two newer features do the job. On 19 February 2026 OpenAI said Code Blocks let you "Preview diagrams and mini apps directly in chat," and on 27 February it said flowcharts and diagrams made in Code Blocks can be exported as images. On 7 October 2026 OpenAI launched GPT-6 with Intelligent UI, which can answer with "side-by-side comparisons and diagrams." Plus, Pro, Business and Enterprise get GPT-6 Sol, while Free and Go get GPT-6 Luna.
We did not drive a signed-in ChatGPT seat for this page, so we can't promise the diagram looks right on your first try. If no diagram appears, use this fallback prompt: Give me the mind map of [topic] as an indented outline, five branches with three sub-ideas each, then turn it into a diagram. The release notes also say Pro's reasoning option runs on GPT-6 Astra, which doesn't support Intelligent UI.
A paper notebook still beats a mind map for some people. If you only need structure, the outline alone is enough, and you can paste it into any notes app.
Which ChatGPT plan do you need for brainstorming?
Free runs all seven methods, with Voice time limited, and Go at $8 adds more voice time and memory. OpenAI's pricing page, read on 15 November 2026, lists the rows that matter here. US prices, before tax:
| What brainstorming uses | Free, $0 | Go, $8 | Plus, $20 | Pro, from $100 |
|---|---|---|---|---|
| Text chats | Unlimited* on GPT-6 | Unlimited* | Unlimited* | Unlimited* |
| Model for chat | GPT-6 Luna | GPT-6 Luna | GPT-6 Sol | GPT-6 Sol, plus Astra reasoning |
| Voice | Yes, limited | Expanded | Expanded | Unlimited* at the top tier |
| Projects | Yes | Yes | Yes | Yes |
| Memory with past chats | Limited | Yes | Expanded | Expanded |
| Deep research | Limited | Limited | Yes | Yes |
| Ads | Not stated | "This plan may include ads" | Not stated | Not stated |
The asterisk means "subject to abuse guardrails," in OpenAI's words, so unlimited isn't literal. Projects are listed as Yes on all four plans. A Project can hold your standing brief, such as audience, tone and ideas you have already rejected, so each new chat starts with the ban list. ChatGPT Projects explains how to set one up.
None of the seven methods needs Plus. Pay for it if you want the stronger model on every chat or deep research on a market. For the cost of keeping Claude or Gemini beside ChatGPT, see one subscription for every AI model.
- Paste the 100-then-bolder prompt and throw away the first 30 ideas before you read the rest
- Run reverse brainstorming when you're stuck: ask for 15 ways to fail, then flip each into a rule
- Add a ban list (no apps, no subscriptions) to any prompt that returns the same ideas twice
- Start a fresh chat with the same prompt and compare the two pools
- Open Voice and ask ChatGPT to interview you one question at a time when typing stalls
- Ask for a diagram or an indented outline for a mind map, since canvas left GPT-5.5 Instant and Thinking
- Keep your standing brief in a Project so every new chat starts with the ideas you already rejected
Frequently asked questions
Is ChatGPT good for brainstorming?
Yes for volume, with a catch on variety. A Wharton working paper found ChatGPT-4 ideas were on average of higher quality than students' ideas, but plain GPT-4 pools were more alike than a student group's (0.377 against 0.243, lower is more varied). Asking for 100 ideas, then bolder ones, cut that to 0.255.
Can you use ChatGPT for brainstorming for free?
Yes. OpenAI's pricing page lists unlimited text chats with GPT-6 on the $0 Free plan as of November 2026, subject to abuse guardrails. Free also lists Projects and limited Voice. Go at $8 adds 3 hours of Voice a day, and Plus at $20 moves chat to GPT-6 Sol.
What is the best ChatGPT prompt for brainstorming?
Ask for 100 one-line ideas, make them bolder and different, then describe the ten strangest. That sequence gave the widest spread in a 35-prompt Wharton test on GPT-4. For a stuck problem, use reverse brainstorming: list 15 ways to fail, then flip each into a rule.
Does ChatGPT remember my brainstorming sessions?
Partly. The pricing page lists memory with past chats as Limited on Free, Yes on Go and Expanded on Plus and Pro. Projects are Yes on every plan, so the surest way to keep a brief is a Project that holds your audience, tone and rejected ideas.