Konshus

Explainer · 7 min read

Why does ChatGPT forget me?

It's not personal. Forgetting is the default in ChatGPT — remembering is a small, opt-in exception layered on top of a fundamentally stateless system. Once you see the shape of it, the fix stops feeling like a hack.

A silhouette walks away as a shadow portrait of them fades into vapor

The default is amnesia

At its core, a large language model has no persistent state between turns. Every message you send is a fresh call with a window of context — the current chat, whatever fits, nothing else. If you close the tab and open a new chat tomorrow, the model has zero knowledge that yesterday's conversation ever happened. That's the baseline. Everything called "memory" is a small system bolted onto the outside that tries to reintroduce a bit of continuity.

Understanding this changes what you expect. It's not that ChatGPThad memory of you and lost it — it never had it in the first sense, and the little memory it does have is fragile by design.

What Memory actually keeps

When you have Memory turned on, ChatGPT keeps a small list of short notes about you — things like "prefers concise replies," "based in Denver," "working on a novel called X." Not transcripts. Not full conversations. Short, model-authored notes. There's a cap on how many; heavy users hit the wall around 200–400 entries. When the cap fills, older notes are silently overwritten with newer ones.

The second piece is "Reference chat history," which lets the model search across your past conversations when it thinks the current prompt is relevant enough. That decision is probabilistic — a question like "give me some ideas" often skips retrieval;"given what you know about me, give me some ideas"almost never does. Same underlying memory, different result.

Why characters disappear

A character or persona you built up over a long chat lives in that chat's context window. When you open a new chat, the context is gone — even if Memory is on, the model didn't save the character traits as memories unless you told it to. This is the single most common source of "wait, why does it feel like a different AI now?" The AI is the same. The context it's operating inside is empty.

Long chats have the same problem in slow motion. As the chat gets longer, old messages fall out of the effective window. The character you established in message 12 is gone by message 400. For a deeper look at where the line is between "still remembers" and "already forgot," see AI memory vs context window explained.

Why the model itself feels different sometimes

Every major model release — GPT-4 → 4o → 5 → whatever comes next — ships with subtle changes in how memory is retrieved, weighted, and expressed. The stored memories don't change, but the model's decisions about them do. This is what people mean when they say "ChatGPT feels different lately" — it usually is. We wrote about this specifically in AI personality changed after an update and in the broader piece on persona drift.

The Temporary Chat trap

Temporary Chat (the lightning-bolt icon) is essentially incognito mode: nothing gets written to memory, nothing added to history, nothing used for training. It's easy to accidentally start one. If you feel like ChatGPT has been unusually forgetful for the last few sessions and history looks sparse, this is worth checking before anything else.

What you can actually do

Inside ChatGPT: turn on both memory switches, keep the saved-memory panel pruned so nothing gets silently evicted, phrase prompts explicitly when you want the model to reach for context, and use Custom Instructions for the handful of things you want on every single turn. That's the ceiling of what ChatGPT alone can do.

Outside ChatGPT: keep a canonical copy of what you want the AI to know about you — preferences, projects, relationships, voice — somewhere you control. Paste the slice that matters into the top of any new chat, and the model behaves as if it remembered you all along. This is the pattern behind every "portable memory" tool, Konshus included: not a better ChatGPT, but a layer underChatGPT that survives when it forgets you.

Never lose your AI again

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 Konshus

Related reading

Frequently Asked Questions

Never lose your AI again

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 Konshus