Most delayed decisions aren’t delayed because you lack information. You’ve usually got enough. What you’re missing is a way to make the actual shape of the problem visible — to yourself.
This is the one place where talking to an AI is more useful than it looks. Not because it’ll tell you what to do. It won’t, and you shouldn’t want it to. But it’s genuinely good at reflecting your own reasoning back at you in a form you can actually argue with.
Why Avoidance Persists Even When You Have the Facts
A decision tends to get avoided for one of a few reasons: the options feel genuinely close, there’s a cost you haven’t admitted out loud yet, or the real blocker isn’t the decision at all — it’s something adjacent that you’d have to face if you committed. These aren’t the same problem, and they don’t respond to the same prompting.
The mistake people make when they bring a stuck decision to an AI is treating it like a research query. “Should I leave my job?” or “Is it worth moving cities for this role?” framed that way gets you a balanced pro/con list. Balanced pro/con lists are useless precisely because the decision was already balanced in your head — that’s why you’re stuck.
The Prompt That Actually Moves Things
Instead of asking for a recommendation, describe the decision in full — both options, what you’ve already considered — then add this at the end: “What question am I not asking that I probably should be?”
That reframing forces the model away from summarizing your inputs and toward finding the gap in your reasoning. It’s not magic, but it works more often than you’d expect, because most stuck decisions have an obvious-in-retrospect thing you’ve been skirting. A model with no emotional stake in the outcome will usually name it directly.
A follow-up worth running immediately after: “If I had already made this decision six months ago and it went badly, what would I say the reason was?” Pre-mortem prompting has been around in decision-theory circles for a while — Gary Klein formalized it in the 1990s — but it’s tedious to run on yourself. Asking an AI to do it using your actual stated reasoning as input is a better use of the tool than asking it to research options you already know.
What to Feed It
The output quality here is almost entirely determined by what you put in. A thin prompt gets a generic response. A rich one gets something you can actually react to.
Write down:
- The two (or three) options as you currently see them — not idealized versions, the actual ones
- What you’ve already decided you don’t want, even if you haven’t admitted why
- Who else is affected and what they seem to want
- The thing you keep coming back to when you think about it at 11pm
That last one matters. What occupies your head when you’re not actively reasoning about it is usually closer to your actual priority than anything you’d write in a structured list. Feed it all in. Don’t clean it up.
Where This Breaks Down
A few honest caveats, because this approach has real limits.
First: the model doesn’t push back on false premises you state confidently. If you describe an option as “basically off the table” because you’ve half-convinced yourself of something that isn’t quite true, it’ll take that at face value. Catching motivated reasoning buried in your own setup text is your job, not the model’s.
Second: for decisions with significant financial or legal stakes, this is a thinking aid, not a substitute for a professional who knows your actual situation. The model doesn’t know what it doesn’t know about your context, and it’ll generate plausible-sounding considerations that may not apply to you specifically.
Third: it’s easy to use this as another form of avoidance. Running the same stuck decision through three different prompt framings across two sessions is procrastination wearing the costume of productivity. If you’ve done one serious session and still haven’t moved, the problem probably isn’t that you need more information. You likely already know what you’re going to do and don’t want to say it yet.
The Actual Useful Version of This
What works: one focused session, one rich input, and a concrete question about what you’re missing. Ask it to steelman the option you’re leaning against. Ask it what someone who made the opposite choice would say two years later. Then stop prompting and sit with what came up.
What doesn’t work: fishing for the answer you want by tweaking the framing until the model agrees with you. It’ll do that. It’s very agreeable. That’s the failure mode.
Before you open a chat at all, try writing your decision out properly — like you’re explaining it to someone who needs to understand it completely. Most of the value from this kind of session comes from the act of articulating clearly, not from the response. The AI’s reply is just a forcing function for doing the hard part, which is making your own thinking legible to yourself. If writing it out already shook something loose, that’s your answer.
