You open Claude for one quick task and get the wall: usage limit reached, come back later.
So you blame the plan and start pricing the next tier up. More often, the leak is how each chat runs.
Claude counts far more than your words. Long chats, big attachments, web search, extended thinking and a high effort setting all draw on the same pool. Claude Code and the desktop app share that pool too.
6 habits that stretch it
1. One seat, one job. Give each task its own fresh chat. Usage grows with the length of the conversation. When the topic changes, open a fresh chat and paste a short summary of where you are.
2. Send the whole brief up front. Three follow-ups cost three full rounds. Batch them into one message, and have Claude flag missing details before it starts.
Paste this: "Read the brief below. If a detail is missing, stop and list what you need. Once I answer, do all three: [task 1], [task 2], [task 3]."
3. Repair the brief and resend. When an answer misses, edit your original prompt and resend it. You skip a correction round, and the chat stays short.
4. Size the settings to the task. Web search, Research, extended thinking and idle connectors all add usage. For a quick question, turn them off and drop the effort level.
5. Park your reference files once. Upload your style guide once instead of attaching it to every chat. Project files are cached, so each reuse costs less.
Receipt: Anthropic's
usage help page says Project content "counts less against your limits when reused." The cache expires after a break, so batch that work into one sitting.
6. Read your two clocks first. Settings > Usage shows your five-hour session bar and your weekly bar. The session clock starts with your first message, so start the heavy build on a fresh session, never at 90%.
Do this today
- Look: open Settings > Usage and note where you sit.
- Reset: take your longest open chat, summarize it in three lines, and start fresh.
- Move: put one file you attach every week into a Project.
↳ Still hitting the wall? The model you pick matters too. That was the August issue: send each job to the smallest model that gets it right.
Want the screenshots, the prompts and the worked examples? Read the full version on Substack
Reply and tell me the job that burns your limit fastest. I read every reply, and I'll tell you which habit fixes it.