You ask AI to help you choose something. It gives you a numbered list. You read through it, none of the options feel right — but you can’t explain why, and you’re not sure if the problem is the list or you.
This happens more than people admit, and it’s worth understanding what’s actually going on when it does.
The list is doing something, just not what you think
When you ask “what are my options here,” the model pattern-matches against everything it’s seen on the topic and returns the most statistically plausible answers. For well-defined problems, those answers are usually solid. But for decisions with personal stakes, it doesn’t know the thing you haven’t said yet: your actual constraint.
Not your stated constraint. Your actual one. The freelancer rate you ask about “for someone at your experience level” doesn’t factor in that you’re also negotiating rent next month. The ten business name ideas don’t know that three of them would embarrass you at family dinners. The training plan options don’t know your knee.
The list looks complete. It has the right shape. But it was generated against a version of your problem that’s missing the load-bearing detail, so none of the answers land cleanly.
Why more options make this worse, not better
The instinct is to ask for more. “Give me 20 instead of 10.” This is almost always the wrong move.
More options generated from the same underspecified prompt are just more instances of the same miss. You’re not increasing coverage of your actual decision space — you’re increasing the pile of things to sort through before concluding that none of them fit. There’s a real cognitive cost to that, and it tends to make the decision feel harder even though you technically have more material to work with.
Barry Schwartz documented this in The Paradox of Choice — more consumer options producing paralysis and regret rather than satisfaction — and the mechanism here is similar: more options without better criteria doesn’t help, it just pushes the unsolved problem one step further down.
What to do instead
Stop generating and start interrogating. Specifically: interrogate the option that got closest to right, even if it still wasn’t right.
Pick the one from the list that got the furthest before it broke down. Then ask the model to pressure-test just that one. “What would go wrong with this? What am I assuming if I pick this? What would I need to believe for this to be the right call?” That’s a different kind of prompt — you’re not asking for more options, you’re asking for resistance. Resistance is more useful than variety when the problem is clarity, not coverage.
The second move is to tell the model why the list didn’t work. Not in vague terms — specifically. “Three of these assume I have more time than I do. Two require a tool I don’t want to learn right now. The rest are fine but feel generic.” Feed that back verbatim. The model can work with it; it just didn’t have it the first time.
This feels slower. It is slower, by a few minutes. But it’s faster than iterating on ten more lists with the same structural problem.
The version of this that’s actually a you-problem
Sometimes the list is fine and the issue is that you don’t actually know what you want yet. The model can’t fix that, and getting it to generate more options is a way of deferring the moment you have to figure it out.
I’ve caught myself doing this — running the same decision through a prompt three different ways, getting back three reasonable lists, and realizing at some point that I’m using the tool to avoid thinking rather than to support it. The outputs were all defensible. None of them resolved anything because what I needed to do was commit, not generate.
If you’re on your third list and nothing feels right, there’s a decent chance the block isn’t informational. It’s motivational or emotional, and option number 31 isn’t going to fix that. Worth knowing before you spend another 20 minutes prompting.
One reframe that actually helps
Treat the list as a first draft of the decision space, not a menu to pick from. Your job when you read it isn’t “which of these” — it’s “what does this list reveal about how I’m framing the problem.” Sometimes a bad list is diagnostic. It shows you what you actually care about by process of elimination, or it surfaces an assumption you were making without realizing it was shaping the question.
The options that obviously don’t work are often more informative than the ones that almost do. If you immediately rule out seven of ten, write down why. That’s your actual criteria list, which is what you should have fed the model in the first place.
Go back in with that. The next list will be shorter and more useful — not because the model got better, but because you got more specific about what “right” actually means here.
The hard part of any decision is usually knowing what you’re actually deciding. Once you know that, AI is fast. Getting there is still your job.
