Why does ChatGPT forget previous conversations?
Most AI chats start fresh every time. Here's why that happens, what the built-in features do and don't fix, and what people actually do about it.
The short answer
Most AI chats start over. By default they don't carry what you talked about last time. This isn't unique to ChatGPT — Claude, Gemini, and most assistants work the same way.
So the project background you spent thirty minutes explaining last week is gone the moment you open a new chat, and you explain it again.
Does any of this sound familiar?
Almost everyone works around it the same way — summarise the chat, keep the summary somewhere, paste it back in next time. You might already be doing some of this:
- When a chat gets long, you ask the AI to summarize it and paste that somewhere — a notes app, a doc, a workspace tool.
- Every time you start a new chat, you go find that summary and paste it back in.
- You want to continue in Claude what you started in ChatGPT, so you paste the same thing again.
- You can't remember where you put it, so you just explain it from scratch.
If that's you, you already know what the problem is. You're solving it by hand.
Why the routine wears you down
Doing it by hand works — it just costs you three things: you have to remember you saved it, the filing takes time, and switching AI starts you over.
You're the one who has to remember. Saving it is the easy part. The hard part is that you have to remember you saved it at the exact moment it would be useful. If that doesn't cross your mind, the saved copy might as well not exist.
Filing takes time. Which page, what title, which folder. If you enjoy organizing, that's a hobby. For most people it's a tax you pay before you get to the actual work.
Switching AI means starting over. Wherever you filed it, moving to a different assistant means putting that context back in front of it again.
What's actually happening
Inside a single chat, the AI answers based on what's been said in that chat. Close it, open a new one, and none of that comes along.
Even within one long conversation, the early details get fuzzy. How much an assistant can hold at once is its context window, and when a chat outgrows it the oldest parts drop out first.
None of this is a malfunction — it's how these tools are built. But from where you're sitting, it feels like "I told you this yesterday" doesn't work.
What about the built-in features?
"Doesn't ChatGPT have a built-in memory?" It does, and it keeps getting broader. You just can't see all of it.
ChatGPT's memory. It picks up context from your chats on its own and carries it into later ones. What it keeps isn't entirely up to you, though. It decides what's worth holding on to and what to leave out — and what it does keep stays inside ChatGPT.
Claude's Projects. Put material into a project space and the chats inside that project can draw on it. It's a genuinely good feature. But it's scoped to that project, and you decide what goes in and put it there. Move to a different project and it doesn't follow.
And both of them work only inside their own service. What ChatGPT remembers, Claude has no idea about.
What actually helps
The idea is simple: save it once, and let the AI do the retrieving.
- When something's worth keeping, you save it once, right then. No folder to pick, no tags to invent.
- Later, when it matters, you just ask, the way you'd normally ask. The AI pulls up whatever you saved that's closest in meaning.
- You don't need to remember the exact title. "The migration plan we wrote up" is enough.
The point is that remembering stops being your job. You decide what's worth keeping; the AI decides when to go get it.
A store kept outside any single assistant like this is called a knowledge layer, and Contextick is one. Connect it to the AI you already use, and saving and recalling happen inside your normal conversation. There's no app to learn and nothing to file.
If you're on Claude, it takes about a minute — here's how.
And if you use more than one AI
Two assistants means keeping the same context in two places, by hand, indefinitely.
If you use ChatGPT and Claude, you're almost certainly pasting the same context into both. Neither one's built-in memory helps here, and neither one's project feature does either — each of them only works inside its own service.
Keep it in one place that any of them can reach, and that whole chore disappears.
How do you carry context between ChatGPT and Claude? — that one's worth its own answer.
FAQ
Do other AI assistants forget too, or is it just ChatGPT?
They all work this way by default. Claude, Gemini and most assistants start each conversation without the specifics of the last one. What differs is how much each keeps inside its own service, not whether a new chat starts fresh.
Why do the earliest parts of a long chat get vague?
Because an assistant can only look at so much at once — its context window. When a conversation grows past that, the oldest turns fall out of view first, so the last few exchanges stay sharp while the opening blurs.
Is keeping summaries in a notes app enough?
It works, but you become the index. The file only helps if you remember it exists at the moment it matters, go find it, and paste it in — and you repeat that for every assistant you use.
How do I stop re-explaining the same background every time?
Save it once somewhere the assistant can reach on its own, then ask the way you normally would. You decide what's worth keeping; going to get it stops being your job.