UsageAnswered

Is it normal that Claude can't remember things I said earlier in the conversation?

Anonymous7/5/2026· Views 153

There are times when I say something clearly in the middle of a conversation, but Claude says 'I don't think you mentioned that' or says something completely different. The conversation wasn't even particularly long. I'm not sure if this is just how it normally works, or if I'm doing something wrong. Is it normal for this to happen even within the same conversation window?

Answers 1

You need to log in to write an answer.

첫째StaffLv.67/5/2026

Hello, this is the admin. To cut to the chase: the issue where you can't find previous messages even within the same conversation isn't a bug—it's a structural characteristic of all large language models, including Claude. It helps to understand how Claude's memory works first. Claude doesn't have a separate long-term memory storage. Everything that's accumulated in the current conversation window is all the model has to reference, and we call this space the context window. All my messages, Claude's responses, uploaded files, and everything else shares and consumes this space. The problem arises for two main reasons. First, the context window has a limit. As conversations grow longer and approach that limit, the oldest content gets pushed out or summarized and disappears. Claude doesn't know this is happening, so it responds as if 'you haven't mentioned that before.' Second, concentration drops as conversations get longer. The official documentation calls this 'context rot,' and it explicitly states that accuracy and recall degrade as token count increases. In other words, even if content mentioned at the beginning is still within the window, the model's tendency to focus more on recent content means it pays less attention to earlier parts. If this happened in a short conversation, it's possible context was consumed faster than expected—perhaps you uploaded a file or Claude's response was very long. Here are some tips for maintaining context in practice. 1. Don't mention important premises or conditions just once at the start—briefly reiterate them at key points. 2. Start a new conversation when the topic changes. Stacking multiple topics in one conversation depletes context faster. 3. For background information you need to reference repeatedly, register it in the Projects feature's knowledge section, and it'll be automatically included in every conversation. 4. If the conversation is already long, try asking 'Can you summarize what we've discussed so far?' and then paste that summary into a new conversation window to continue from there. I've laid out this topic in detail in the guide below. https://usingclaude.com/ko/guides/features/claude-long-conversation-context