There's a particular kind of decision that resists the usual approaches. Not the small stuff โ€” those you just make. And not decisions where you can gather more data until the answer becomes obvious. The hard category is somewhere in the middle: you have some information, you have strong feelings, and the stakes are high enough that you keep circling without landing anywhere. Moving across the country for a job. Leaving a career you've spent twenty years building. Whether to make a major financial commitment right now or wait.

Most people in that situation talk to a few people they trust, make a list of pros and cons, and then continue circling. The list helps some. The conversations help more. But there's a gap that both tend to leave: the pros and cons list can't push back on your reasoning, and the people you trust often have opinions of their own that color their responses.

AI tools โ€” Claude and ChatGPT in particular โ€” turn out to be genuinely useful here, but not in the way most people expect. They're not useful because they can analyze your situation and tell you what to do. They're useful because they can help you think more clearly before you decide. That's a different thing, and it's worth understanding how it actually works.

What AI Is Actually Doing When You Use It This Way

When you describe a complex personal situation to Claude or ChatGPT and start working through it in conversation, you're doing something that has a long history in psychology: externalized thinking. Putting a problem into words, in a form that can be read back to you, does something to your understanding of it. Therapists use this. Coaches use it. It's why journaling works for a lot of people even when nothing writes back.

What AI adds is the "writes back" part. It can reflect your situation back to you, ask follow-up questions, surface assumptions in your reasoning, and offer framings you hadn't considered โ€” without the complication of having its own stake in your decision. Your friends and family usually have opinions. Sometimes those opinions are wise. Sometimes they're based on their own fears or their own experiences, which are not your fears or your experiences. The AI has no agenda about what you do. That turns out to be useful.

There is a limitation worth naming right at the start: AI has no real knowledge of your life. It knows only what you tell it. This means the quality of what you get back is directly related to how much you put in. A vague question gets a generic answer. A specific, honest description of your actual situation gets something much more useful. We'll come back to this.

How to Set Up the Conversation

The single most important thing is context. AI tools like Claude or ChatGPT work within what's called a context window โ€” essentially, everything you've told them in the current conversation. They can't access your history, your values, your track record of decisions. You have to supply that yourself, upfront.

A useful opening for this kind of conversation looks something like this:

I'm trying to think through a major decision and I'd find it helpful to work through it in conversation rather than just getting advice. I'll describe the situation โ€” what I know, what I'm uncertain about, and what I'm feeling. I want you to ask me questions, push back on my reasoning where it seems shaky, and help me surface things I might be avoiding. Don't tell me what to decide. Help me think.

That framing does two things. It tells the AI what mode you want it to be in โ€” questioning rather than advising โ€” and it signals the kind of engagement you're looking for. Without that setup, AI tools tend to default to helpful-advice mode, where they summarize your situation and offer options. That can be useful, but it's not the same as being helped to think.

After that opening, describe your situation in as much honest detail as you can. Include the parts you're embarrassed about or that you suspect might not be entirely rational. The parts you're avoiding in other conversations are often exactly what's driving the difficulty.

What Good AI Questioning Actually Looks Like

Once you've set up the conversation and described your situation, a good AI exchange starts to feel something like a slow, patient interview. Claude is particularly good at this โ€” it tends to pick up on things you've said in passing and return to them, which can surface assumptions you didn't notice you were making.

For example, if you describe a job offer in another city and mention that you'd need to sell your house, Claude might ask: "You mentioned selling your house relatively quickly โ€” how do you actually feel about that part of the transition, separate from the job itself?" That question seems simple. But if you hadn't separated the housing piece from the job piece in your own thinking, it can open something up.

Other questions that tend to be generative in these conversations:

You can ask the AI to ask you these kinds of questions explicitly, or you can let the conversation develop and see where it goes. Both approaches work. The directed version is faster. The open version sometimes finds things the directed version misses.

The Part Where You Push Back on Yourself

One of the more useful things AI can do in this kind of conversation is play devil's advocate โ€” not randomly, but specifically against the direction you seem to be leaning.

You can ask for this directly:

Based on what I've told you, it sounds like I'm leaning toward [option]. I want you to make the strongest possible case against that choice โ€” not to talk me out of it, but to make sure I've genuinely considered the other side.

The response you get will sometimes include things you've already thought through, and that's fine โ€” confirming that you've considered them and have an answer is itself useful. Occasionally it will surface something you hadn't genuinely sat with, and that's the more valuable outcome.

The important thing is to read these counterarguments as genuinely as you can. There's a temptation to dismiss them quickly โ€” "I already thought of that, it doesn't apply to my situation." Sometimes that's true. Sometimes it's the part of the decision you've been moving past without fully examining.

Where AI Can't Help โ€” and What to Do Instead

This kind of conversation has real limits, and being honest about them is part of using it well.

AI doesn't know what you actually value, even if you describe your values in the conversation. It can work with what you tell it, but values are often revealed in behavior rather than stated in words. If you say you value stability but have spent your career making risky moves, that gap is meaningful โ€” and only you can see it clearly.

AI also doesn't know the specific people in your life: your partner, your kids, your parents, the colleagues who would be affected by your decision. It can work with descriptions, but the texture of those relationships โ€” and how they factor into the decision โ€” is something it can only approximate.

And AI tools do sometimes confidently state things that aren't true. If the conversation touches factual territory โ€” tax implications of a financial decision, specific legal requirements, medical information โ€” verify separately before acting on it. That applies especially to anything where being wrong has real consequences.

This kind of AI conversation works best as a complement to other sources of clarity, not a replacement. After spending an hour working through something with Claude, the conversation you have with your partner or a trusted friend often gets sharper โ€” because you've already done some of the preliminary work of understanding what you actually think.

A Practical Way to End the Conversation

Once you feel like you've worked through the main dimensions of the decision, it's worth asking the AI to reflect back what it heard โ€” not to tell you what to do, but to summarize the key tensions and the questions that seem most unresolved. Something like:

Based on our whole conversation, can you summarize: what are the two or three things that seem most unresolved for me? Not what I should decide โ€” just where the real uncertainty still lives.

That summary is often worth reading twice. It's a reflection of your own thinking, organized by something that was paying close attention to the conversation. It won't make the decision for you. But it tends to clarify where the actual work still needs to happen โ€” and where you can stop circling because you've genuinely thought it through.

Big decisions are still yours to make, in all their difficulty. But thinking poorly about them doesn't make them less difficult โ€” it just makes the outcome less considered. AI, used this way, is one more tool for thinking better. That's enough of a reason to try it.