Why does AI forget what you told it earlier?

Short answer

A language model only knows what you told it if that text is in its context window, the text it sees while answering. A new chat gets your instructions, project context and a selection from memory, but not your earlier conversations in full. In a long chat, accuracy also drops as the window fills, and older messages can be dropped or summarized.

The model has no memory of its own

A language model does not store your conversations in itself. It has general knowledge from training, but what you told it only comes along through the context window: your instructions, the conversation so far and whatever the application adds, such as project files and memories. Anthropic’s documentation describes it as the model’s working memory and stresses that it is not the same as the data the model was trained on.

So forgetting takes two forms. Either the text was not in the window. Or it was, but the model did not use it well, for example because the window was full and cluttered. There are five common causes. Check them in this order.

Five reasons the AI forgot

Situation What happens How to check What helps
You started a new chat Earlier conversations do not come along in full, only instructions and a selection from memory Ask: “what do you know about X?” Project instructions, asking it to remember explicitly
The chat got very long Accuracy drops, older messages can fall out or be summarized Ask about something from the start of the chat Start a fresh chat with a short summary
Memory selected something else Memory stores and retrieves a selection Look in the memory settings at what is stored Ask explicitly to remember
A setting limited memory Temporary or incognito chats create no new memories; a project with project-only memory sees only what is in the project Check which kind of chat or project you have open Pick the right chat or project before you start
You switched tool or account Each tool keeps its own memory Ask the same question in the other tool A memory layer that sits outside the tools

1. A new chat does not bring the old conversation along. Without a memory feature or project instructions, nothing from yesterday is there. ChatGPT and Claude both have memory features, but they work by putting a selection back into the window, not by replaying your history.

2. Long chats get worse. Anthropic calls it context rot: as the number of tokens grows, the model’s accuracy and recall go down. It is a gradual decline, not a hard cut-off. On top of that, chat interfaces like claude.ai can manage the window first in, first out, so in a very long chat the start disappears first. Agents such as Claude Code use compaction: they summarize the conversation when the window fills up, and details that do not make it into the summary are gone after that. Current Claude models have a window of 1 million tokens, but a large window does not prevent this. More on that in What is context rot?

3. Memory is selective by design. OpenAI’s help page says memory “does not retain every detail from every conversation” and that ChatGPT decides which available information is relevant to a response. ChatGPT also does not search your history for every answer, only when it is likely to help. Claude saves memory as separate topics while you chat and you can see them under Settings > Your Claude > Manage.

4. A setting limited memory. In Claude you can turn memory off for a single chat in the “+” menu, and incognito chats are not saved at all. ChatGPT’s temporary chats do not create or update memories, though they can use existing memories if you choose that before starting. Projects can be separated too: Claude uses a project’s own memory plus your account memory, but never memory from other projects. In ChatGPT, a project set to project-only memory cannot reference chats outside it, and shared projects always use project-only memory.

5. You changed tools. What ChatGPT remembers is not available in Claude, Cursor or Copilot. Claude has an experimental import and export for memory, but there is no live connection between assistants.

The misunderstanding

“Memory full” can mean two different things, and the fix differs:

  1. The conversation is full. A very long chat hits the limit of the context window. Deleting memories does not help. Start a new chat and paste in a short summary of what matters.
  2. Your saved memories. Under Settings > Personalization > Memory > Manage you see what is stored and can clear out outdated or duplicate entries. For what ChatGPT can reference through chat history, OpenAI states there is no separate storage limit.

Even when nothing is full, memory is narrow: it holds what one tool selected from one person’s chats. The lasting fix is putting the important things where the AI will reliably load them, such as custom instructions or project instructions.

What this means for a team

Each person builds up their own memory, separate from their colleagues. When you explain to ChatGPT how your company writes quotes, your colleague’s ChatGPT still does not know. When a developer switches from Claude Code to Cursor, the context stays behind. And when someone leaves, the memory built in their personal account leaves with them. See How can a team share one AI memory? and Does Claude have memory across chats?.

Where Bombos fits

Bombos keeps what your team tells its AI tools outside any single chat, account or tool, and loads it into whichever assistant someone is using: Claude, ChatGPT, Copilot, Cursor, Codex, Grok or your own agents. Corrections become rules the AI has ready for the next question, and you can see which knowledge an answer used. It is a plugin that runs in our cloud or on your own server, and one person can start with it for their own agents.

Related questions

Is ChatGPT's memory full when it forgets things?

Not necessarily. OpenAI says there is no separate storage limit for what ChatGPT can reference through chat history, and that memory does not keep every detail of every conversation. Forgetting can also be selection, or a conversation that got too long.

Why does the AI forget things within the same chat?

In a long conversation the window fills up. Anthropic's documentation describes context rot, lower accuracy and recall as the token count grows, and notes that chat interfaces can manage the window first in, first out, so the oldest messages fall out first.

How do I make the AI remember something important?

Say it explicitly. Both ChatGPT and Claude let you ask the assistant to remember something, and both have settings where you can review and edit what is stored. For rules that must always apply, use custom instructions or project instructions.

Does a temporary or incognito chat use my memory?

Claude's incognito chats are not saved to memory or chat history. ChatGPT's temporary chats do not create or update memories, and you can choose before starting whether they use existing memories.

Sources

  1. Context windows (Claude API docs)
  2. Effective context engineering for AI agents (Anthropic, September 2025)
  3. Memory in ChatGPT (OpenAI Help Center)
  4. Use Claude's chat search and memory (Claude Help Center)
The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.