Claude pillar guide
Everything you need to know about long Claude conversations: how the 200,000-token context window works, what to do when Claude says the conversation is too long, how to save and export your history, and how to stop attention drift from erasing earlier reasoning.
Claude's context window is 200,000 tokens, roughly 150,000 words, or about 500 pages of text. For the vast majority of Claude users you will never hit the hard limit in normal use.
The caveat: even well before 200K, performance can degrade in very long conversations. Claude, like all large language models, pays differential attention across a long context. Content near the beginning receives less attention than content near the end, so instructions established early can quietly lose influence.
If Claude can still respond to short messages, ask for a comprehensive summary: goal, every key decision, all outputs produced, important context you've given, and what still needs doing. Copy that summary and start a new conversation, pasting it as your first message with an explicit "I'm continuing work from a previous session, please pick up from here."
If Claude can no longer respond, scroll through the old conversation and manually copy the most important outputs and decisions. Claude Projects are the best long-term solution for extended work: instructions and uploaded documents live in a separate persistent layer, so you can work across many conversations without filling the active window.
Claude does offer self-service export. In the Claude web app go to Settings, then Privacy, then Export data (see Anthropic's help article on exporting Claude data). Anthropic emails a link to a .zip containing your conversations as JSON. It is a full-account export rather than per-conversation, and delivery can take up to 24 hours.
For a single thread, copy-paste or the share link are still the fastest paths. The export is your raw backup; a thredly summary is the working handoff you paste into a fresh chat, or into a different AI tool entirely, without re-reading the whole transcript. See the cross-platform walkthrough on how to export AI chat history across platforms.
Two distinct phenomena feel like forgetting. The first is a hard context limit at 200,000 tokens, rare in normal use. The second, more common and more subtle, is attention drift: content near the start of a long conversation receives less attention than content near the end, so early instructions have less influence on later responses.
Each new conversation starts with a fresh 200K window. Claude does now carry some context between chats via chat search, chat-history memory summaries, and Projects (see Anthropic's help article on chat search and memory), but those are Claude's own background summaries, not a visible, editable handover document for one specific long thread that you can paste anywhere.
No. The conversation still exists and is fully readable. You just can't add more messages to it. All outputs, documents, and discussion remain visible.
Claude's 200K window is larger than the default window on most ChatGPT plans. The exact ChatGPT window varies by the current model and plan, and both platforms still hit a finite limit on very long conversations.
Partially. Project instructions load fresh for each new conversation, so persistent instructions aren't buried under history. For truly long single conversations, re-anchoring is still needed.