🧮 Cut your Claude usage in half: the cheat sheet

Whether you're hitting free-tier limits or watching a paid bill climb, the fix is the same: most people burn 2-3x the usage they need. Here's the cheat sheet — what actually consumes your quota and the habits that cut it roughly in half.

What actually costs (the mental model)

Every message re-reads the whole conversation — the chat so far, pasted files, project context. Long chats get expensive per message and dumber, because the important stuff drowns. Usage isn't about how much Claude writes; it's mostly about how much it re-reads.

The seven habits

1. New task, new chat. The single biggest saver. Finished the email? Start fresh for the plan. Continuing a 50-message chat to ask an unrelated question re-reads all 50 messages for nothing.
2. Use the handoff brief. When a long chat is still needed, compress it: "Summarize this conversation as a brief for a fresh chat" — then restart with the brief (prompt #5 here).
3. Paste excerpts, not documents. The relevant two paragraphs, not the 40-page PDF. If Claude needs the whole thing once, fine — but don't leave it in a chat you keep using.
4. Put standing context in a Project. Project instructions are handled far more efficiently than re-pasting your business brief into every chat — and you stop forgetting it.
5. Batch related questions. Five small questions as five messages = five re-reads. One message with five numbered questions = one.
6. Ask for the format you want up front. "Answer in a table" once beats generating prose, then asking for a table — every re-do is a full re-read plus a full re-write.
7. Match the model to the job. If your plan offers model choice: routine drafting doesn't need the heavyweight model. Save the big model (and extended thinking) for work that deserves it.

The false economies (don't do these)

• Starving context to save tokens. A vague prompt that needs three retries costs more than one rich prompt. Cut re-reads, never clarity.
• Cramming unrelated jobs into one chat "to save chats." Chats are free; context is what costs.