A banned.md prompt workflow that strips AI-sounding patterns from your writing
VibeMarketer_ · x · 2026-11-14
A practical prompt workflow for making AI text read like a human: encode rules into a banned.md file, have the model read it in full before drafting, then check the draft line by line against the rules — rewriting whole sentences rather than swapping synonyms when a rule breaks. Before sending, three more checks: count em dashes and exclamation marks (both must be zero), delete the last line if it summarizes, and reread it like a text to a friend, rewriting anything that sounds like a model. The banned.md rules target structures over words: negative parallelism ("it's not x, it's y") becomes just y; fake build-ups ("no x. no y. just z.") become just z; self-answered questions ("the result?") become plain statements.
Related event: Developers use banned.md files to strip the AI flavor from generated text(2 posts)→
More from Apps
- YC startup paused Demo Day, hit mid six-figure ARR in 4 weeks growing 50% WoW — ycombinator · 2026-11-14
- Google AI Mode local answers now show offer labels, powered by Gemini — gaganghotra_ · 2026-11-14
- Openside launches open-source, self-hostable Riverside and Loom alternative — damnGruz · 2026-11-14
- Hugging Face's New Llama Windows App Hits Download Bugs on Day One — Timalakeseinai · 2026-11-14
- Snowpaw demo shows multi-agent teamwork: engineer agent fixes parser, marketing agent updates changelog — gnukeith · 2026-11-14
- Google AI Mode now shows local store inventory in answers, no extra work needed for merchants — gaganghotra_ · 2026-11-14