If you start writing today, there's no way to know if you can write without AI

This is a practical, opinionated guide to using generative AI for writing that distinguishes two outcomes: AI-extended, elaborated copy that becomes bloated and homogenized, and raw, prompt-like notes that are leaner, more useful, and more human. It argues that when content is consumed by people it should be written by people, because AI outputs tend to look polished yet feel uniform and exhausting to read. AI is useful for brainstorming, research, and processing personal notes - examples include using local RAG workflows in Obsidian with DuckDB to synthesize markdown - but relying on models to craft final prose often leaves the writer more drained and produces less characterful work.

The note also addresses credibility and detection: newcomers who start publishing during the AI era have no easy way to prove they can write without model assistance, while established writers benefit from prior, human-authored track records. Large-scale detection work (e.g., Pangram Labs) shows AI content is widespread on platforms like LinkedIn and Medium, rarer on Substack and many Reddit communities, but detection is imperfect and non-deterministic. Practical tool mentions include ChatGPT, Claude, Lex, Byword, Spiral, River Docs and Wordsmith, and the conclusion favors manual writing for human audiences while recommending selective AI use for ideation and personal knowledge management.