Long AI conversations have a hard limit: once you hit the context window ceiling, the tool stops accepting input and any unpreserved work is effectively lost. The "summarize this chat" technique is the primary defense against this.
Every AI chat session has a finite context window — the total amount of text (input + output) it can hold at once. Claude's context window is large (~128,000 tokens, roughly 90,000–100,000 words), but extended working sessions — especially those involving large data exports, long generated pages, or iterative back-and-forth — can exhaust it faster than expected.
When you hit the limit:
- The tool refuses new input
- You must start a fresh chat
- The new chat has no memory of prior work
- If you haven't summarized recently, reconstructing context is slow and lossy
At regular intervals during any substantive chat, send:
"Summarize this chat."
Claude will produce a concise summary of decisions made, content generated, and open questions. This summary becomes your portable context record.
To check remaining capacity, ask:
"How much context window do I have left?"
Claude will estimate the remaining space. If it reports ~10% remaining, summarize immediately and prepare to migrate.
When a chat is nearly full:
"Here's a summary of what I was working on. Please continue from here: [paste summary]"
The new chat now has full context and a fresh token budget. This is especially useful when switching between tools (e.g., moving work from Claude to ChatGPT when Claude's daily limit is exhausted).
Summaries also serve as in-chat landmarks. Because summaries use consistent language, you can use Ctrl+F within a chat to jump directly to them — useful when a single chat has grown long and you need to locate a specific decision or output without scrolling through everything.
This is particularly valuable in stream-of-consciousness working sessions where good ideas get buried in exploratory back-and-forth.