I think you should almost never use AI to write

Original Article ↗ Hacker News Discussion ↗

Writers and developers are wrestling with when, if ever, it makes sense to let large language models draft prose for them. Many argue that outsourcing writing to AI undermines thinking, produces subtly wrong or bloated text, and misleads readers who expect a human voice, while others find AI invaluable as an editor, explainer, or starting point—especially for non‑native speakers or tedious workplace writing. The exchange surfaces broader concerns about responsibility, authenticity, and skill atrophy, as well as comparisons to AI-written code, where errors are easier to detect but long-term maintenance and clarity still suffer from machine‑generated output.

AI as Writing Tool: Assistant vs Author

  • Many distinguish between using AI to generate full text vs. using it to assist (critique, edit, rephrase, structure).
  • Strong support for AI as “copy editor” or “rubber duck” that highlights weak spots, overused phrases, and structure issues, while the human writes the actual prose.
  • Others happily let AI draft emails, tickets, or proposals, then lightly edit, prioritizing speed over craftsmanship.

Writing as Thinking and Skill-Building

  • Repeated theme: writing is part of thinking; outsourcing prose risks shallow understanding.
  • Several report that reviewing AI-generated documents took longer and yielded worse results than writing from scratch, especially for nuanced or technical work.
  • Some fear becoming mere editors instead of authors, which feels less creative and more like janitorial work.

Quality Issues: Vagueness, Errors, and Style

  • Many find AI prose verbose, vague, and subtly wrong, especially with logic, quantifiers, negation, and domain-specific nuance.
  • “LLM-isms” (generic tone, overconfident phrasing, formulaic structures) are widely described as grating and mentally fatiguing to read.
  • Others counter that AI often produces clearer, more concise versions of their drafts and that not using such tools is inefficient.

Reader Expectations, Authenticity, and Responsibility

  • Strong sentiment that when you write for others, they expect your words and understanding, not a machine’s.
  • Using AI to author text can dilute intellectual responsibility: the writer may not fully grasp or be able to defend what’s said.
  • Some see unlabeled AI-written text as misleading or disrespectful; others suggest disclaimers or “co-written” labels.

Use Cases People Like

  • Summaries for personal consumption, note cleanup, spelling/grammar, simplifying expert material for different audiences.
  • Helping non-native speakers with idioms and correctness.
  • Using AI to propose multiple phrasings, alternative outlines, or to critique and debate specific passages.

Code and Other Domains

  • Several note that the arguments also apply to code: code is communication, and AI-generated code can be subtly wrong and hard to maintain.
  • Others are comfortable using AI for code because correctness can be tested more objectively than prose.

Social and Economic Tensions

  • Some label opposition as Luddite; argue productivity and job pressure will force widespread AI use.
  • Others describe resentment, overload from AI-slop at work, and a growing tendency to ignore or penalize obviously AI-written text.