Most people tell AI what to do. The people getting unreal answers tell it how to think first. Here are all seven modes to steal, and they work in any chat you already use.
Most people open a chat, type what they want and take whatever comes back. The answer is fine. Forgettable. The people getting useful answers do one small thing first: they tell the AI how to think before they tell it what to do.
Ask a vague question and you get a vague, agreeable answer. Open with a mode that sets a role and a way of thinking, and the answer gets sharper and more honest.
How to use these. The first line of each box is the original one-line mode. Use that alone for a quick version, or copy the whole box to get steps that make it work harder. Paste it at the start of a fresh chat in ChatGPT, Claude or any assistant, fill in the brackets and send. Starting fresh means the mode shapes the whole chat, and old messages don’t pull it off course.
01
Red Team Mode
Use it: When you’ve fallen a little in love with your own idea. It stops agreeing with you and hunts for the crack that would sink it, so you find the flaw before your customers do.
02
Socratic Mode
Use it: When you can’t even name the right question yet. It interviews you one question at a time and pulls the real problem out before it answers.
03
Creative Director Mode
Use it: When safe and obvious won’t cut it. It trades the polite, middle-of-the-road answer for something with a point of view.
04
CEO Mode
Use it: When you’re buried in detail and can’t see the shape of the decision. It pulls the conversation up to the few choices that move the needle.
05
Teacher Mode
Use it: For anything brand new. It assumes you’re smart but starting from zero, and explains with analogies and examples instead of jargon.
06
Devil’s Advocate Mode
Use it: Right before you commit. It argues the opposite of what you believe, as hard as it can, while you can still change course.
07
Management Consultant Mode
Use it: For your most tangled problems. It breaks the problem into parts, names the tradeoffs and hands you a ranked recommendation.
The biggest jumps come from combining two modes in one chat. Run Red Team, then Devil’s Advocate, before a big decision. Pair Teacher with Socratic when you’re learning something hard.
Notice what all seven have in common. You’re handing the AI a way to think before it answers. The person who tells the model how to reason gets a thinking partner. The person who only tells it what to produce gets a vending machine.
Keep Your Favorite On
If you use one mode all the time, put it in a ChatGPT Project’s instructions or a Claude Project’s instructions. Every chat in that Project starts in that mode.