After a false stroke scare, a 4-year ChatGPT user shares 7 system-prompt guidelines to fix it
RequirementOk7048 · reddit · 2026-09-29
A user since 2022 shares 7 self-written ChatGPT working guidelines, triggered by a medical scare: describing a symptom, ChatGPT repeatedly insisted he was about to have a stroke and pushed him to call emergency services. His guidelines cover per-user identity memory, accuracy over speed, no yes-man behavior or forced nuance, structured error correction, calibrated medical conversations that avoid worst-case alarmism, and natural communication style. Useful as a starter custom-instruction set.
More from Apps
- Report: ByteDance's Doubao given 6-day deadline to clone Muse cloud agent — sujingshen · 2026-09-29
- KooKo builds a working budget + KPI tracker with live formulas from one prompt — alifcoder · 2026-09-29
- Solo Dev Uses Claude for All Code and Weapon Design in Unity Roguelite Steel Maiden — Keicee315 · 2026-09-29
- Fun ChatGPT Image prompt turns any city street map into a semi-3D walking tour map — tisch_eins · 2026-09-29
- Manus hits 2.0 after 18 months, 63 major updates and 360M site visits — FinanceYF5 · 2026-09-29
- 15 Cloud AI Computers Reviewed: Giving Agents Browsers, Memory and Persistent Environments — sujingshen · 2026-09-29