A short-story experiment found that AI ideas improved individual ratings while making the stories more similar to one another.
Writers do not need another tool that makes every draft sound acceptable. They need something worth saying and a way of saying it that belongs to the subject. A family argument, a book review, and a report about a broken bus route should not all arrive in the same smooth, agreeable voice.
That is why a study of AI-assisted short stories is more interesting than a simple claim that AI makes writing better or worse. Readers gave better ratings to stories written with AI ideas, particularly when the writer had scored lower on an earlier creativity measure. Yet the assisted stories also became more alike. An improvement for one writer came with a loss of variety across the collection.
For anyone publishing a blog, that is a real editorial problem. A page can be grammatically clean and still feel interchangeable with a hundred other pages. If the tool chooses the premise, the examples, and the ending before the writer has formed a view, polishing the sentences afterward will not restore much individuality.
Anil Doshi and Oliver Hauser ran a short-story experiment. Writers could work alone or get ideas from generative AI. Readers rated the resulting stories. AI ideas helped the stories get better ratings. Writers with lower scores on a first test of creative skill gained more, but the AI-assisted stories were more like one another. The study appeared in Science Advances in 2024. The authors also share an open paper version. It tested stories made in a set task. It did not show that a writer loses creative ability for good. The risk it found was a loss of variety across outputs.
That is a useful distinction. Better work on one measure can still leave a collection poorer on another.
Try this invented story opening: A girl finds a strange key in her grandfather's desk. You can already guess some likely next steps. There is a locked room. There is a family secret. There is a choice that brings the family together. Any of those could become a good story, but once the key is on the page, you may stop thinking about all the other ways to begin. The grandfather could have sold the desk. The girl could hate old things. The key could fit a bus station locker full of unpaid bills.
A suggested opening can shape all the decisions that follow. Once a writer has accepted the key and the locked room, they may spend the session developing that premise rather than considering a different story. Starting with some independent ideas preserves alternatives before that commitment forms.
For a personal essay or an observational piece, begin with details from the subject you actually know. A shopkeeper repairing a sign or a passenger counting change may be more useful than a general instruction to write something moving. For a review, begin with a scene or argument from the work itself. Do not worry yet about a theme. A detail can be useful before you know what it means. The point is to give the draft a source other than the tool's most likely opening.
Then choose two details that do not seem to belong together. Who would care about both? What might they want? What would get in the way? You now have a start that grew from your attention. This exercise was not tested in the short-story experiment. It is a way of bringing material of your own to a drafting tool, so its suggestions have something specific to respond to rather than setting the whole direction.
If you want AI help, show it the rough draft and ask where the story becomes easy to predict. Ask which part is hard to follow. Ask what a reader needs to know sooner. Those questions leave the main choices with you. You can accept a useful complaint without accepting the tool's rewrite. A friend can say a scene is dull; that does not mean the friend should write the scene.
Review suggested changes for meaning and voice, not just smoothness. An unusual sentence may capture a particular observation better than a conventional rewrite. Reading aloud can help you notice whether the draft sounds like someone describing this subject or using language that could fit almost anything.
You can swap a castle for a space station and still keep the same story. A young hero finds a secret power, faces a test, and wins through belief. New scenery does not always make a new idea. Change the question at the heart of it. What if the hero does not want the prize? What if the problem cannot be solved by one brave person? What if the kind act makes the day harder rather than easier?
Keep the answer honest to the people in the story. Strange for its own sake can feel just as empty as a stock ending. The goal is to find a choice you mean, not to confuse the reader.
I would write the argument or scene before asking a tool to improve it. Decide what you think, choose the details that support it, and leave the assistant a specific job: identify a confusing passage, point out a missing explanation, or flag an unsupported claim. You can use that feedback without adopting its voice. The study does not show that assisted writers permanently lose creativity. It shows why a collection needs more than well-rated individual pieces. If every article makes the same moves and arrives at the same mild conclusion, readers have little reason to keep reading, however clean the prose is.
