How to Use LLMs as Copyeditors Without Losing Your Voice

To improve writing with LLMs without losing personal voice, authors must treat the technology as a mechanical copyeditor rather than a creative partner. By strictly refusing to use LLM-suggested phrasing and ignoring the model's tendency to praise early drafts, writers can focus on fixing objective flaws. This approach leverages the model's tireless ability to spot errors while keeping the author's unique perspective intact.

Key Points

• Use LLMs strictly as copyeditors to find flaws rather than as ghostwriters to generate text.
• Rule One: Never adopt specific words or turns of phrase suggested by the model to avoid the 'uncanny valley' of AI prose.
• Rule Two: Disregard the model's inherent praise and encouragement, as it prevents the deep revision necessary for good writing.
• LLMs are exceptionally good at spotting mechanical errors like passive voice, nominalization, and overused filler words.
• Maintain your personal voice by manually rewriting sections based on the model's critiques instead of letting the model rewrite them for you.

Sentiment

The overall sentiment is cautiously aligned with the article but not uniformly supportive. Hacker News largely accepts the core point that LLMs can be helpful as copyeditors, critics, and structure checkers while human authors keep control of the final wording. At the same time, the thread contains a strong anti-LLM current that distrusts AI involvement in prose, worries about contamination of voice, and mocks recognizable model diction.

In Agreement

• LLMs are useful for spotting filler, passive constructions, unclear flow, inconsistencies, and missing context, as long as the author rewrites the text manually.
• Writing one's own commit messages, pull request descriptions, and explanations preserves understanding of agent-generated work and prevents shallow review.
• Good AI-assisted writing workflows constrain the model to critique, organize raw notes, or suggest a few targeted clarity improvements rather than generating finished prose.
• The article's warning about bland AI phrasing matches commenters' experience with verbose, over-polished, and generic model output.
• Human editing remains essential because LLMs can over-explain, hallucinate, or produce plausible summaries that miss the real reason behind a change.

Opposed

• Some commenters reject LLM use for writing altogether, arguing that prose for humans should convey a human mind rather than outsourced expression.
• Several participants argue that even reviewing model suggestions can contaminate a writer's taste and make their own language drift toward AI-like phrasing.
• Critics say LLM-generated commit messages often restate the diff instead of explaining the purpose, making them redundant or misleading.
• Some commenters think repeated model-driven editing loops make writing worse, turning reasonable text into generic or incoherent prose.
• A few accuse the article itself of using AI-coded metaphors and phrasing, treating that as evidence that the proposed discipline may be harder than it sounds.