ChatGPT silently mines your old chats to shape new answers — here's the 2-minute fix
Ok_Negotiation_2587 · reddit · 2026-09-13
A Reddit user found ChatGPT kept steering a law-firm pitch review toward a "friendly, casual" tone — caused by a months-old coffee-brand copywriting chat it had quietly absorbed. ChatGPT memory has two layers: editable saved memories, and an invisible mechanism that reads your past conversations to infer job, preferences and style, then applies them to new chats without any disclosure. The post lists telltale symptoms and a free fix: audit Settings → Personalization → Memory, use Temporary Chats for unbiased reads, and prompt the model to state its assumptions about you and where the answer would differ without them.
More from Apps
- Meta's Muse agent calculates a creator is 4.6 years from financial independence — armand_ruiz · 2026-09-14
- MIT spinout Atlas turns plastic waste into building materials with AI-driven recycling and 3D printing — nordicinst · 2026-09-14
- MIT spinout Atlas turns plastic waste into building materials with AI robotics — MIT News AI · 2026-09-14
- Perplexity launches Hybrid Compute to split AI tasks between cloud and local Mac — Aiden_Tech_Ai · 2026-09-14
- He screenshotted a tweet and had Muse build a kids' bank with 3% weekly interest — armand_ruiz · 2026-09-14
- Open-source cloud-native GIS platform GeoLibre ships v3.0 with Cesium 3D as core engine — giswqs · 2026-09-14