Every writing guide tells you ideas are cheap and execution is everything. That is true and it is also unhelpful, because the reason most short stories stall is not weak execution. It is that the idea was never load bearing in the first place. An idea that goes somewhere has a specific property: it contains a conflict that cannot be resolved without someone changing. “A woman inherits a house” is a situation. “A woman inherits the house she ran away from, and the only way to sell it is to contact the sister she has not spoken to in nine years” is an idea, because you cannot get to the end without something breaking. Here is how to generate ideas with that property on purpose rather than waiting for them.
Start with a want and an obstacle that are the same thing
The strongest premises are ones where the thing the character wants is the thing that will hurt them. A detective who needs the case solved and cannot afford who it points to. A man who wants his father’s approval and can only get it by becoming someone he does not respect. Try this as a mechanical exercise. Write a want. Then write the obstacle. Then ask whether you can rewrite the obstacle so it comes from inside the want rather than from outside it. Most first drafts of a premise have an external obstacle, and most of them get better with this single move.
Collide two contexts that do not usually meet
A large share of memorable stories are one setting plus one incongruity. A funeral and a job interview. A first date and a hostage negotiation. A monastery and a product launch. The collision does its own work. Both contexts have rules, and the story writes itself out of the places where the rules disagree. Make a list of twenty settings you know something about, then pair them at random. Most pairs are nothing. Two or three will produce an immediate scene in your head, and that reaction is the signal you are looking for.
Ask what happens the day after the obvious ending
Take a premise that feels tired and skip past its natural conclusion. The heist is over and the money is split; now what. The couple got together; now what. The prophecy was fulfilled and the world did not improve. This is one of the reliable ways to get an original story out of an unoriginal premise, because the familiarity of the setup does the exposition for free and the interest lives entirely in territory nobody has covered.
Use generation as a pressure test, not a source
This is where tools earn their place. The point of an AI story generator is not to hand you a premise you then write. Premises you did not choose are hard to care about, and caring is most of what carries a writer through a draft. The useful pattern is the opposite. Bring your own premise and generate several different directions out of it, quickly. Seeing four versions of what your idea could become tells you which one you actually want, and it usually tells you within a minute. That is a decision you would otherwise make by writing three thousand words and abandoning them. It is also a good way to find the obvious version so you can avoid it. Whatever a model produces first is very often the average take on your premise, which makes it a useful map of the ground already covered.
The test before you commit
Before you spend a week on a story, write one sentence that states what changes. Not what happens. What changes. If you cannot write that sentence, the idea is a situation, and no amount of good prose will rescue it. If you can, you have something with a shape, and the rest is the ordinary work of finding the words. Most stalled drafts I have seen fail this test, and almost all of them fail it at the outset rather than partway through. Five minutes of honesty about it saves a fortnight.
