ClaudeModels & limitsLevel 1: Basic prompts

When Should You Start a New Claude Chat?

Start a new chat when the topic changes, when Claude keeps returning to ideas you dropped, or when you're close to a usage limit. Running out of room is a less common reason than it used to be, because Claude can now summarize older messages and keep going.

Prompt to copy 
I'm moving this work to a new chat. Write a handoff note I can paste there, with:
1. The goal, in one or two sentences.
2. Decisions made so far, and why.
3. The current version of [the document or code], in full.
4. What we tried and rejected, so it isn't suggested again.
5. The exact next step.

Leave out the back-and-forth. Keep everything except the current version under [300] words.
how off track the chat is →
New chat, handoff notename what you rejected
New chatcarry nothing over
Keep goingClaude can summarize older messages
New chata short summary at most
how far the topic has moved →
Rule of thumb, not an Anthropic chart. Near a usage limit, Anthropic's help page also suggests a new conversation.

Try it now

  1. 1In the long chat, paste the prompt above, then read the note and correct anything that's wrong.
  2. 2Open a new chat, paste the note, and ask for the next step.
  3. 3On a paid plan, if you later need something from the old chat, ask Claude to find it. This uses Search and reference chats under Settings > Your Claude.
  4. 4You'll know it worked when the next answer follows your decisions and doesn't bring back anything from the rejected list.

Use it when

When your next question would make sense without the earlier chat, when you've corrected the same thing twice, or when you've hit a length limit. Anthropic says to start a new conversation or use a project in that last case.

Skip it when

When the history is the point: a long debugging session where Claude holds the findings so far, or a document built up with many agreed decisions. Keep going if answers are still good and you aren't near a limit. Compaction can handle the length.

Why it works

Each turn sends the whole conversation back to the model, as Anthropic's context window docs describe. So a long chat costs more and gives Claude more old material to weigh, including drafts you rejected. Anthropic's help page says conversation length affects usage, and suggests a new conversation if you're approaching your limit in a longer chat. Compaction solves the space problem but not the others: it lets a chat continue, but those chats use more of your limit. Anthropic's Claude Code guide explains the quality side. After two failed corrections, a third mostly adds to a conversation that already holds two wrong attempts, and one new prompt that uses what you learned almost always works better.

Example

Say a chat started with a blog outline, went through three rewrites, and Claude keeps returning to a structure you rejected an hour ago. Ask for the handoff note, check that it names the rejected structure, and paste it into a new chat with the final outline. The old structure is now one line marked as rejected, rather than three full drafts for Claude to reread.

Going further

Claude Code has specific commands for this. /clear starts over, and /clear oauth-migration names the session you're leaving so you can return later. /compact keep the API changes and the test commands summarizes with a focus, though the docs warn that it's a big request on a long session. /rewind rolls back code, conversation or both. If you build on the API, note that Anthropic's Fable 5 guide says the model may suggest a new session in very long runs, most often because your harness shows it a token countdown. That suggestion may not mean space has run out.

Common mistakes

  • Asking for a fourth fix in a chat where three have failed, rather than starting over with what you learned.
  • Copying the whole old conversation into the new one, which brings the same clutter back.
  • Starting fresh without saying what was rejected, then discussing the same options again.
  • Assuming a new chat in the same project inherits the last one. Anthropic says context isn't shared across project chats unless it's in the knowledge base, so add the handoff note there.

Sources

Checked 6 November 2026. Products change often, so confirm details with the vendor.

Related

Concept How Much Can You Fit in a Claude Chat? Technique How Do You Keep a Long Claude Draft Consistent Across Chats? Workflow How Do You Give Claude Memory Across a Long Project? Workflow How Do You Design a SaaS Database With Claude?

Beyond the course

Glossary Large language model Pricing guide Claude pricing, plans and models