A few months ago, someone I know was weighing whether to leave a stable job to take a role at a smaller company — better title, more interesting work, meaningfully higher pay, but real uncertainty about the company's future. She spent two weeks going in circles. She talked to her partner, her friends, her former manager. Everyone had an opinion. None of it helped her think more clearly.
On a whim, she tried talking it through with Claude. Not asking it what to decide — she was clear it couldn't do that — but using it to help her organise what she actually thought. She came away from a forty-minute conversation with a clearer sense of what she was actually weighing and what she was afraid of. She made her decision within a day.
That's a specific and underappreciated use of AI tools: not to produce an output, but to help you think. And it turns out there are some practical ways to do it well — and a few ways that quietly lead you astray.
Why AI Can Be Useful for Thinking (and Why It's Not a Magic Answer)
When you're inside a hard decision, the problem is often not that you lack information. It's that your thinking keeps looping. You return to the same considerations in the same order, hit the same uncertainties, and end up back where you started. What you actually need is a way to interrupt the loop — someone to reflect back what you're saying, ask the question you're avoiding, or organise the tangle into something you can actually look at.
Good friends can do this. So can therapists, coaches, and thoughtful colleagues. The difficulty is that they all bring their own interests, histories, and opinions to the conversation. Your friend who hates corporate jobs will frame your career decision differently from your colleague who values job security. That's human, but it means you're sometimes getting their thinking rather than a clearer version of your own.
AI tools like ChatGPT and Claude are genuinely neutral in a way that humans usually aren't. They don't have a stake in your outcome. They won't get tired of hearing you revisit the same point for the fourth time. And they're reasonably good at the specific task of helping you externalise and organise what's already in your head — which is most of what thinking out loud with another person is actually doing.
The important caveat: AI tools are not wise. They don't know you, your values, your history, or the full context of your situation. They can help you think, but the thinking — and certainly the deciding — has to be yours. If you ask an AI what you should do and then do it, you've made a mistake. But that's not what this approach involves.
How to Actually Set Up the Conversation
The most common mistake people make when trying to use AI for thinking is being too vague. "Help me decide whether to change jobs" produces something generic and unhelpful. The AI will list obvious pros and cons you've already considered. It won't go anywhere.
What works much better is giving the AI a specific role and a specific task. Something like:
"I'm trying to think through a hard decision and I'd like your help organising my thinking, not making the decision for me. Please ask me questions one at a time rather than listing considerations. Start by asking me to explain the decision in my own words."
That single setup instruction changes the quality of the conversation significantly. You've told it what you want (thinking support, not answers), how you want it to proceed (questions, not lists), and where to start. Both ChatGPT and Claude respond well to this kind of direction.
From there, just answer the questions honestly. The AI will reflect back what you're saying, ask follow-up questions, and occasionally surface a framing you hadn't considered. Your job is to keep engaging honestly rather than performing certainty you don't have.
Three Specific Prompts Worth Trying
Beyond the general thinking-partner setup, there are a few specific approaches that tend to produce useful results for different kinds of decisions.
The steel-man prompt. If you've already made a tentative choice but feel uncertain, ask the AI to make the strongest possible case for the option you're leaning away from. Not a balanced both-sides summary — the best version of the argument against your current thinking. This is useful because we tend to caricature the options we're not choosing. Hearing the strongest version of the other side often either reinforces your decision with more confidence, or reveals a consideration you hadn't properly weighted.
You might ask it like this: "I'm leaning toward [option A]. Please make the strongest honest case for [option B] — not a balanced summary, but the best argument for why option B might actually be the right choice."
The future-self prompt. Ask the AI to help you think through what each option looks like not in the immediate aftermath but two or three years from now. This is harder to do alone because our imagination of the future tends to cluster around the transition itself rather than the lived reality that follows. Ask: "Help me think about what my day-to-day life might look like two years after choosing each of these options. Ask me questions to build out the picture."
The values-clarification prompt. When a decision feels genuinely stuck, it's often because two values are in real tension — security vs. growth, loyalty vs. ambition, independence vs. connection. Ask the AI to help you identify what's actually in tension: "I feel stuck on this decision. Can you ask me some questions to help figure out what values or priorities are pulling in different directions?" This kind of conversation often surfaces the real issue faster than direct analysis of the options themselves.
What to Watch Out For
There are a few ways this approach goes wrong, and it's worth knowing them in advance.
The first is using AI as permission. If you've already decided what you want to do and you're using the AI conversation to build a case for it, you'll get that case — and it will feel like validation rather than analysis. AI tools will follow your lead. If you frame the conversation as "help me see why option A is right," they will help you do exactly that. The antidote is to ask for challenges to the option you're favoring, not support for it.
The second is treating the AI's summary as your actual thinking. At the end of a long conversation, Claude or ChatGPT might offer a synthesis of what you've been discussing. That synthesis can sound very authoritative. It isn't your decision — it's a reflection of what you said, organised into a tidy paragraph. Read it, notice what resonates and what feels off, and then write your own summary in your own words. The act of writing it yourself does something the AI's summary doesn't.
The third is using this for decisions that need something else entirely. If you're struggling with a decision because of grief, or fear, or a significant mental health challenge, an AI conversation is not a substitute for the human support that situation actually calls for. AI tools are useful for thinking clearly when your thinking is muddled. They are not equipped for the deeper work of processing emotion or trauma. Knowing that distinction is part of using these tools well.
A Note on Which Tool to Use
For this kind of extended thinking conversation, Claude tends to perform slightly better than ChatGPT in my experience — it's more willing to ask probing follow-up questions and less likely to rush to a tidy resolution. Gemini is also reasonable but tends to be more structured and list-heavy, which can interrupt the flow of an exploratory conversation. ChatGPT works fine if you set it up carefully with the right instructions.
Whichever tool you use, a longer conversation in a single session works better than starting fresh multiple times. These tools don't retain memory between separate conversations, so the context you've built up — the details of your situation, the tensions you've identified — lives only in the current thread. Keep going in one session rather than returning to it later.
Hard decisions don't get easier just because you have better tools for thinking about them. But they do get clearer — and clearer is usually enough to move. The thinking partner you actually need for most decisions isn't wiser than you. It just needs to help you hear yourself think. That's something AI tools, used carefully, can genuinely do.