Most people use AI to think the same way: describe a problem, ask "what should I do?", and accept whatever comes back. It feels productive. But you've usually just outsourced the one part that mattered — the thinking — and gotten a confident answer to a question you never checked was the right one.

The problem isn't the model. It's that thinking help, like writing help, follows an order — and skipping to the answer skips the steps that actually change the outcome. This library is built around that order: you frame the problem before you solve it, challenge the reasoning before you commit, and check your confidence before you act.

Here's how to actually use it.

Don't ask for an answer. Ask for better thinking.

The fastest upgrade to any thinking prompt is to stop asking the model to decide for you and start asking it to sharpen how you decide. Compare these:

Should I take the job or not?

I'm deciding between staying and taking the new job. Ask me up to 5 questions that would change your recommendation, then make the case for each side and tell me what you'd pick — and what I seem to be avoiding.

The first gets you a generic pros-and-cons list built on facts you never gave it. The second makes you do the work that only you can do, then pressure-tests it. The model isn't the decider here. It's the thinking partner that catches what you can't see from inside your own head.

Frame the problem before you solve it

This is the single most underused move, and the reason the library leads with Frame the Problem.

When something's wrong, the instinct is to start solving. But most bad outcomes come from solving the wrong problem well. Before anything else, try:

Help me state the real problem in one sentence, then check whether I'm solving that or a symptom of it.

What am I treating as fixed that might actually be changeable?

What's the smallest, sharpest version of this problem I could work on first?

If you can't state the problem in one clean sentence, you're not ready to solve it — and an hour spent here saves a week spent building the wrong thing.

Fix big problems before small ones

When you're reasoning through a real decision, work in this order:

Frame — what's the real problem, underneath the obvious one?
Clarify — what do I actually want, and what am I optimizing for?
Generate — what are the real options, including the ones I'm avoiding?
Challenge — where is my reasoning weak, motivated, or assumption-heavy?
Consequences — what happens at the second and third order?
Risk — what's the median case, and where does it fail?
Decide — with criteria set before I fell in love with an option.
Act — what's the next concrete step today?

The order matters because there's no point stress-testing an option you'd never pick, or running a pre-mortem on a plan that solves the wrong problem. Most people invert this — they agonize over the final 10% of a decision while the framing quietly stays broken underneath.

Argue with yourself on purpose

AI is very good at agreeing with you. Left alone, it will polish your reasoning into something that sounds airtight while leaving every weak assumption intact. The Challenge My Thinking and Bias & Blind Spots prompts exist to stop that.

What am I assuming that, if false, would change everything?

Red-team this plan. How would someone trying to make it fail attack it?

Where am I being motivated rather than objective?

The goal of an argument with yourself is to lose it cheaply, now, instead of expensively, later. Run these before you've committed, while changing your mind is still free.

Calibrate your confidence

A conclusion isn't done when it sounds right. It's done when you know how much to trust it. The Evidence & Confidence prompts force the question most thinking skips:

How confident should I actually be in this conclusion from 0–100%, and why?

What evidence would prove me wrong?

What do I know, what do I think, and what am I guessing?

Putting a number on it is the trick. "I'm about 60% sure" leads to a different plan than "I'm certain" — and saying it out loud usually reveals you're guessing more than you admitted.

A 30-second router

Not sure where to start? Match your situation:

Stuck, going in circles → Reframe & Get Unstuck, then Frame the Problem.
A real decision to make → Frame the Problem, then Decision-Making.
Big, irreversible, or expensive call → Decision-Making, then Risk & Pre-Mortem.
Too many options, can't prioritize → Compare & Prioritize.
A belief or plan you feel sure about → Challenge My Thinking, then Evidence & Confidence.
The same problem keeps coming back → Systems Thinking & Leverage.
Torn, and it's really about what you want → Clarify Goals & Values.
Just need to reason it out loud → Reason Through It Together.

A quick example

Say you're convinced: "We need to rebuild the whole onboarding flow — that's why people are churning."

Don't ask the model how to rebuild it. Frame first — "is this the real problem or a symptom?" — and challenge it: "what am I assuming that, if false, would change everything?" You might surface the assumption you never tested: that people are churning because of onboarding, rather than churning later for a reason onboarding can't fix.

Now ask for the cheap test: "what's the riskiest assumption here, and how can I test it this week?" Maybe you talk to five churned users before you rebuild anything. The value didn't come from a smarter "how do I fix this" prompt — it came from refusing to solve the problem until you'd checked it was the problem.

The bottom line

Prompts are starting points; your honesty does the heavy lifting. The best results come from three habits, every time:

Frame the problem before you reach for a solution.
Argue against yourself before you commit, not after.
Say how confident you actually are — and what would change your mind.

If you only ever use ten of these, use these:

What am I treating as fixed that might actually be changeable?
What am I assuming that, if false, would change everything?
What's the real problem versus the symptom?
What would a smart skeptic poke holes in first?
What information would actually change my decision?
What's the smallest version of this problem I can solve first?
What are the second-order consequences?
Run a pre-mortem: why did this fail?
What am I optimizing for — and is that the right target?
What single question, if answered, would make this obvious?

Bookmark the workflow, work it in order, and these prompts will do more than help you think faster — they'll help you think better, and catch the mistakes you'd otherwise only find in hindsight.