An AI idea can help your story. Is it also making your options more alike?

Our view: evaluate the differences between your options as well as the quality of each one. Several polished drafts may still share one basic premise.

A Science Advances study published July 12, 2024 tested eight-sentence stories. Access to AI ideas improved outside evaluators’ ratings on measures including novelty and usefulness, while AI-assisted stories were more similar to one another. This is evidence from a specific writing experiment, not a finding about every creative task or current model.

Quality and variety can matter together

If you need one clear draft, improving that draft can be useful. If you need several directions to choose from, ask whether the options differ in their central idea. Counting drafts alone does not answer that question. This is our editorial interpretation of the distinction.

Compare premises before polishing

In an invented planning example, three stories all start with a lonely keeper in a haunted lighthouse. Changing the weather or the keeper’s name does not create three substantially different premises. A story about a crowded train and another about an unreliable family recipe offer different starting points; their quality still needs judgment.

Keep an independent starting point

Our suggested practice is to write a few rough ideas before asking AI for alternatives, then compare the premises without immediately rewriting them into the same style. Ask for different constraints or viewpoints when options cluster. The cited study did not validate this specific checklist as an intervention.

Avoid the sweeping conclusion

The research does not establish that AI inevitably destroys creativity, or that using it guarantees original work. Treat the results as a reason to inspect your option set. Keep the useful assistance and still decide which differences matter to your audience.

Published October 14, 2026 as a column using a dated 2024 study, not new research news. We read the published paper via UCL; scenarios and recommendations are editorial, with no model trial or new creativity experiment.

Sources: Science Advances · UCL.