Troubleshooting · 6 min read
ChatGPT's Memory feature breaks more often than OpenAI lets on. Here are the five most common causes — in order of how often they're the actual culprit — and the structural reason memory will keep breaking until you stop renting it from a provider.
Short answer
If ChatGPT has stopped remembering things, check (1) Memory is on in Settings → Personalization → Memory, (2) you're not in a Temporary Chat, (3) the saved-memory cap isn't full, (4) you're signed into the right account, and (5) a recent model update hasn't reset behavior. Most "broken memory" complaints are one of these five.
The most common one. Go to Settings → Personalization → Memory and check that both Reference saved memories and Reference chat historyare on. They're separate switches and either can be off without warning — especially after an app update or account change.
Temporary Chat is the lightning-bolt icon next to the model name. Anything sent in that mode is excluded from memory, history, and training — and it's easy to accidentally start one. If your last few chats are missing from history, this is almost certainly why.
OpenAI caps saved memories at a few hundred entries. When the slot fills, the system silently overwrites older ones with newer ones. Open the Memory panel and prune ruthlessly — every entry kept is one less that gets dropped. (See our deeper piece on ChatGPT memory limits.)
Memory is tied to the account, not the device. If you signed into work SSO last week or have multiple accounts, you might be talking to a different ChatGPT than the one that remembers you. Sign-out, sign back in, confirm the email in Settings.
Every major model release (GPT-4 → 4o → 5) has shipped with subtle changes to how memory is surfaced and weighted. The data is usually still there — but the model decides what to retrieve, and a new model can make different decisions. (More on this in why model updates wipe AI memory.)
Even when ChatGPT memory is working perfectly, it is structurally fragile: opt-in, capped, opaque, and entirely under OpenAI's control. Every fix above is patching a symptom. The cause is that the memory of youlives on someone else's server, under their terms, subject to their schedule. The next model release can change recall behavior overnight. A deprecated version takes its personalization with it. A policy update can quietly re-flag your training preferences.
If memory matters — for work continuity, long projects, evolving voice — the durable fix isn't a setting. It's owning the layer.
Konshus is one way to solve this — a persistent memory vault and portable persona that follows you across ChatGPT, Claude, Gemini, and whatever ships next.
Meet KonshusKonshus is one way to solve this — a persistent memory vault and portable persona that follows you across ChatGPT, Claude, Gemini, and whatever ships next.
Meet Konshus