Similarity can begin before the sentence
Two writers can use different words and produce similar stories. Both choose the expected reversal, give the protagonist the same reassuring realization, and close with an image explaining what the reader should feel. Replacing familiar phrases would change the surface while leaving the imaginative choices untouched.
Voice includes rhythm and vocabulary, but it also emerges from attention: what a narrator notices, what remains unexplained, what a character finds funny or shameful, and how readily a story resolves discomfort. Similarity in one dimension does not establish sameness in all of them.
What research actually found
In a 2024 online experiment, Anil Doshi and Oliver Hauser studied short stories written with or without access to AI-generated ideas. Assisted stories received better evaluator ratings on several dimensions, while becoming more similar to one another. This was a bounded short-story task, not a longitudinal test of professional novelists or proof that every use of AI erases voice. [1]
Something can help an individual writer while narrowing the range of choices across a group. For a novelist, the practical question is whether assistance repeatedly steers the project toward familiar possibilities the writer would otherwise question. That is an editorial inference from the research, not a measured claim about your manuscript.
Protect the observation
Imagine two versions of a retirement-party scene. One describes a crowded room, warm applause and a man reflecting on the years. Another notices that the replacement employee already knows where the spare mugs are kept, while the retiring man still carries the key he must surrender before leaving. Neither guarantees quality. The second supplies a particular relationship between belonging and replacement that a writer can develop.
A style guide asking for warm, literary and emotionally resonant prose leaves many choices unaddressed. A more useful guide describes habits of attention: this narrator notices practical arrangements before expressions; discomfort appears through an unnecessary task; reflection complicates action rather than announcing its lesson. These are directions to test, not a recipe for distinctive prose.
Use examples from your own writing and explain why they matter. Keep counterexamples too. A rejected paragraph can show where a narrator becomes more certain, sentimental or explanatory than the book permits. Make the distinction visible enough to question.
Choose ideas before choosing language
Write your own possibilities before asking for alternatives. Then change a specific assumption: what if the apology is sincere but insufficient? What if leaving is the kinder choice? What if the discovery increases uncertainty instead of solving the mystery?
Compare consequences. Unusualness alone can become another convention. A spectacular twist may undermine the ordinary pressure that made the premise interesting. Record why a direction fits and what it requires earlier in the book, rather than choosing it because it looks original in a list.
Inspect repeated resolutions across chapters. Do confrontations consistently end in mutual understanding? Do difficult decisions become affirmations of personal growth? These patterns may be intentional. If they are not, the manuscript needs a structural decision rather than a synonym pass.
A comparison you can use
Write a short scene from your own notes. If you want an assisted comparison, supply the same premise and a compact description of viewpoint, stakes and constraints. Save the instructions and output. Compare what each version notices, which choices it avoids, how sentences distribute emphasis, and what readers are allowed to infer.
Ask a reader familiar with your work which passages feel consistent with it, without identifying their origins. This is an editorial exercise, not a controlled study or an AI detector. One preference cannot establish a general effect. Its value lies in identifying a decision you want to retain or recover.
The assistant may suggest something better than your first approach. Keep it if it belongs and describe the contribution accurately. Protecting a voice does not require refusing outside influence. It requires attending to the choices that accumulate into a recognizable way of seeing.
Sources & editorial note
Prepared with AI research and drafting assistance. PennedCanon is the editorial publisher and has a commercial interest in the workspace. Examples are original illustrations, not recorded model experiments. Practical exercises are editorial recommendations, not guarantees of quality, authorship or copyright.