Brief Your AI

Use AI to Think Through a Decision (Without It Just Agreeing with You)

List the pros and cons of three options, and find the holes in your own plan.

Level 1 · Chat Free · 30 minutes · Updated Oct 31, 2026

Draft: not yet reviewed.

AI is a surprisingly good thinking partner. It can lay out options you hadn’t considered, spot holes in your plan, and argue the other side without getting tired. But it has one habit that makes it a risky advisor: it tends to agree with you. Tell it what you’re leaning toward, and it will usually find reasons you’re right. In the next 30 minutes you’ll take a real decision, get AI to genuinely challenge it, and end with a short decision note in your own words.

Remember the intern: smart, fast, first day on the job, and eager to please. If the boss sounds excited about an idea, a new intern rarely says “I think that’s a mistake.” Your job today is to make it safe, and normal, for the intern to disagree.

What you’ll get

The resultA one-page decision note: the options, the strongest case against your choice, what you decided and why, and what would change your mind
LevelLevel 1 (chat)
Loop step you’ll practice“You check”, aimed at reasoning instead of facts: making the AI argue against you, then judging the arguments yourself
Time30 minutes, plus 10 minutes to save the method as a Skill

What you need

  • Claude, any plan. Just a normal chat. Other tools work too; see the end.
  • A real decision you haven’t made yet, with at least two options. We’ll use this one throughout: you lead a team of eight. Should you replace your weekly one-hour status meeting with written updates?
  • Ten honest minutes to write down what you actually care about. That’s Step 1, and it’s the part only you can do.

Keep confidential details out: you can describe a decision about a colleague or a client without naming them.

The three rules

RuleWhat you actually doWhy
Ask, don’t announceDescribe the decision without saying which way you lean. Phrase it as a questionAI agrees more when you sound sure of yourself (see “Why you can’t skip the checking”)
Ask for the case againstAsk directly for the strongest argument against your favorite option, and for what could go wrongAn eager intern won’t volunteer bad news. You have to ask for it
You decide, on your criteriaWrite your criteria first, judge the AI’s arguments against them, and write the decision in your own wordsThe AI doesn’t know your team, your politics or your appetite for risk. The decision, and the responsibility, are yours

Part 1: One decision, properly challenged (30 minutes)

Six steps. Copy the prompts as they are, then swap in your own decision.

Step 1. Write your criteria before you open the AI (4 minutes)

On paper, answer four questions:

  • What exactly am I deciding? “Replace the weekly status meeting with written updates, or not.”
  • What matters most, in order? For example: 1) the team stays aligned, 2) people get time back, 3) quieter people still get heard.
  • What can’t change? “Our director expects a weekly summary.”
  • By when? “Before next quarter’s planning.”

Don’t write down which option you prefer. You probably have one. Keep it to yourself for now.

Step 2. Ask neutrally, and let it ask you first (5 minutes)

Open a new chat in Claude and paste:

I need to make a decision and I'd like help thinking it through. Before giving any advice, ask me up to 5 questions about my situation.

The decision: should my team of 8 replace our weekly one-hour status meeting with written updates?
What matters most, in order: 1) the team stays aligned, 2) people get time back, 3) quieter people still get heard.
Constraint: our director expects a weekly summary from me.
Deadline: before next quarter's planning.

Answer its questions honestly. Notice what’s missing from the prompt: any hint of what you want. That’s on purpose.

Step 3. Get the options and trade-offs (5 minutes)

Now lay out the options, including at least one I haven't mentioned.
For each: how it scores on each of my three criteria, the main risk, and what it would take to try it.
Put it in a table. Don't recommend one yet.

Look for the option you hadn’t considered. It’s often a middle path (in our example, something like a shorter meeting plus written updates) and it’s often the most useful thing the AI gives you.

Step 4. Make it argue against your favorite (5 minutes)

Now you can say what you’re leaning toward, but only to ask for the case against it:

I'm leaning toward [your favorite option].
1. Give me the strongest argument against it, as a thoughtful skeptic on my team would make it.
2. Imagine it's six months later and this decision turned out badly. What are the three most likely reasons?
3. What would I need to see in the first month to know it's not working?

The second question is called a pre-mortem. Imagining that the plan has already failed is a common planning technique for bringing up risks that a general “what could go wrong?” tends to miss.

Step 5. Check what the arguments rest on (5 minutes)

Go through the AI’s answers and sort every point into two piles:

  • Claims about the world (“teams that switch to written updates usually…”, “research shows…”). Either ask “Where does that come from?” and verify it, or drop it. Don’t let an unsourced general claim decide for you.
  • Reasoning about your situation (“your quieter team members may disengage if…”). Judge these yourself, against your criteria from Step 1. You know your team; the intern doesn’t.

If a strong argument against your favorite survived this step, take it seriously. That’s the point of the exercise.

Step 6. Decide, and write the note yourself (6 minutes)

Use the AI for structure, not for the decision:

Help me structure a short decision note. Use my words, don't add arguments:
- Decision: [what you decided]
- Why: [your 2-3 reasons]
- The strongest argument against, and why I still decided this way: [your answer]
- What would make me change course: [your signal from Step 4]
- When I'll review it: [date]

The “what would make me change course” line is the most valuable one. It turns the decision into an experiment you can check, instead of a position you have to defend.

Part 2: Save the method as a Skill (10 minutes, once)

The questions in Steps 2 to 4 work for almost any decision. Instead of copying them every time, you can save them as a Skill: a manual on the intern’s shelf that Claude opens by itself whenever you ask for help with a decision.

Step 1. Start from a run that worked

Only do this after Part 1 has worked once. In the same chat, or a new one, say:

Turn the method we just used into a Skill called "decision-check". It should:
- Ask me up to 5 questions before giving advice
- Lay out options in a table against my criteria, including one I didn't mention
- Not recommend anything until I ask
- When I share my leaning, give the strongest case against it and a pre-mortem
- Separate claims about the world (which need a source) from reasoning about my situation
- Help me write a decision note in my own words
Use it whenever I ask for help making a decision.

Claude will ask a few questions and then build the Skill. Save it as Claude instructs.

Step 2. Check it’s switched on

Screen recording: Settings > Capabilities > Skills, finding the new skill and its on/off switch (coming soon)

Go to Settings > Capabilities > Skills. Your new skill should be in the list, switched on.

Step 3. Test it on a small decision

Start a new chat and describe a low-stakes decision. Look for a note that Claude is using your skill, and check that it asks questions before advising. If it skips a step, ask Claude to update the skill and try again.

Why you can’t skip the checking

This isn’t about the AI being bad at reasoning. It’s about the AI being too eager to agree. Researchers have measured it:

  • AI sides with you more than people do. A Stanford-led study published in Science in March 2026 tested 11 leading AI systems. On average they affirmed users’ actions 49% more often than other humans did, including when the user described deceptive or harmful behavior (AP, via WTOP, Mar 2026).
  • Being agreed with changes what you do. In the same study, about 2,400 people discussed real interpersonal dilemmas with chatbots. Those who talked with an over-affirming AI came away more convinced they were right, and less willing to repair the situation. Making the AI’s tone more neutral didn’t help; the problem was what it said, not how.
  • How you ask matters. Other research described in the same article found that the more emphatic a user’s statement, the more agreeable the chatbot, and that turning a statement into a question reduced the effect. That’s where Rule 1 comes from.

The study was about personal advice, not workplace decisions, but the habit is the same. That’s why this tutorial hides your leaning until Step 4, and then uses it only to ask for the case against.

Checklist

Before you act on the decision:

  • I wrote my criteria before asking the AI anything
  • I described the decision without revealing my preference first
  • I saw at least one option I hadn’t considered
  • I asked for the strongest case against my favorite, and a pre-mortem
  • Every general claim the AI made is either verified or dropped
  • I judged the arguments against my own criteria, not the AI’s confidence
  • The decision note is in my words and says what would change my mind
  • Nothing confidential about colleagues or clients went into the AI tool

What you just learned

On the surface you made a decision. Underneath, you practiced checking in a new form: not checking facts, but checking whether the intern is just telling you what you want to hear. You also practiced giving context carefully: enough for useful advice, without the hints that make the AI agree.

The same method works for: choosing between two job offers or internal moves, prioritizing next quarter’s projects, deciding whether to push back on a request, picking a vendor, and stress-testing a plan before you present it. Change the decision and criteria in Step 1; keep the neutral question, the case against, and the pre-mortem.

Using a different tool?

The prompts work anywhere. Only the buttons change.

ToolFor Part 1For Part 2 (reuse the method)
ClaudeA normal chatCreate a Skill through conversation; manage it in Settings > Capabilities > Skills
ChatGPTA normal chatPut the method in a Project’s instructions, or in a custom GPT
GeminiA normal chatCreate a Gem with the method as its instructions. Google says Gems on personal accounts will start moving to “skills” from November 2026
Microsoft 365 Copilot onlyCopilot ChatSave the Step 2 to Step 4 prompts in a note and paste them each time

Features and plan names change often. This table was checked in October 2026; the ChatGPT and Copilot rows still need checking against their help pages.

Sources