On AI writing

Dima Bolkunov examines how AI-generated prose - defined here as text with at least 50% of words produced by AI - alters authorship, style, and communication. Drawing on a bachelor’s thesis that used word-usage statistics and pattern analysis to attribute short paragraphs to authors, he shows that human writers leave detectable signatures in word choice, sentence structure and narrative patterns. Large language models, by averaging styles and resisting idiosyncrasy even with prompt engineering, erase those signatures and strip away the contextual cues that create empathy and reveal something about a writer’s background.

AI writing also degrades precision and forces different cognitive habits. Models make implicit assumptions to stitch text together, so they favor safer, more ambiguous phrasing to avoid logical errors; authors who rely on AI lose the self-critical process of drafting that exposes gaps in thinking. The result is slower, costlier reading: ambiguous wording and loss of stylistic context require readers to reinterpret text, and model verbosity increases total reading time - the author reports at least a one-third slowdown. Those reader costs erase much of the writer’s time savings and generate second-order harms such as more follow-up questions, longer collaborative cycles, and diminished trust.