Claude Limits Running Out? 24 Rules to Make Them Last

Hitting the “usage limit reached” banner in the middle of the week is the worst feeling, especially when you’re right in the middle of something good. I’ve been there more times than I’d like to admit. So when I came across this post from a LinkedIn creator who packed every Claude limit-saving habit into one list, I saved it right away.

The author calls them 22 rules, but I counted 24. The backstory is simple: the original poster’s team kept hitting Claude’s weekly cap until they worked out these habits. The author also says plainly that they’re not paid by Anthropic. This is just what worked for them.

I’ve grouped every rule into themes so it’s easier to skim. Each one comes with a short note on why it saves tokens.

⚙️ Pick the Right Settings First

The fastest wins come from settings you choose before you type anything. These rules are about not overpaying for horsepower you don’t need.

1. Use Medium effort. Never Max. Higher effort means more thinking tokens on every reply. Medium handles most everyday work just fine.
2. Low for grammar, high for one big prompt. Match the effort to the job. Fixing a typo doesn’t need deep reasoning, but one big, complex request might.
3. Save your one free reset for a bad day. Treat it like an emergency fund, not something to burn on a Tuesday afternoon.
4. Use Opus 5.5 in every chat. Never Fable. The author picks one strong default model and sticks with it instead of reaching for the heaviest option.
5. Never switch models mid-chat. When you switch, the new model has to process the whole conversation from scratch, and that eats into your budget.

💬 Keep Your Chats Short and Focused

Once your settings are sorted, the next big leak is the conversation itself. Every message you send makes Claude reread the whole thread. Long chats get more expensive with every reply.

6. Never go back to an old chat. That long history gets processed again every time you add a message.
7. Ask for a 10-line summary and paste it in a new one. You keep the important context and drop the dead weight.
8. One topic per chat. Mixing topics bloats the context with stuff that has nothing to do with your current question.
9. Never correct Claude. Edit your prompt or start over. Each correction stacks another round on top of the bad answer. Editing the original prompt replaces it.
10. Put all your asks in one message. One full request costs less than five small follow-ups.
11. Start every task with “Ask me questions before you start.” A few clarifying questions upfront beat three rounds of redoing work Claude guessed wrong.

📎 Trim What You Feed It

This section was the eye-opener for me. A lot of token burn has nothing to do with your prompts. It comes from the files, tools and instructions sitting in the background.

12. End quick questions with “No web search. No files.” This stops Claude from going on an expensive side trip for a simple answer.
13. Crop every screenshot. Turn every PDF into text. Images and full PDFs cost far more than the few lines you actually care about.
14. Upload files you reuse once, in a Project. Stop re-uploading the same docs into every new chat.
15. Keep your instructions under 2,000 words. Custom instructions load into every conversation, so long ones add up fast.
16. Paste Anthropic’s 2 sentences into Settings → General → Instructions. The author points to two specific sentences from Anthropic that belong in your global instructions. You’ll find the exact wording in the original post.
17. Load your apps on demand (+ → Connectors → Tool access). Every connected tool adds definitions to your context, even when you don’t use it. Only switch them on when you need them.

🛠️ Work Smarter Day to Day

The last group covers daily habits. These are small workflow changes that quietly save a lot over a week.

18. Plan your deck in text first. Ask for the file last. Generating files is expensive. Lock down the outline and content in plain text, then ask for the file once.
19. Highlight the part to fix, then click “Edit with Claude”. A targeted edit costs far less than regenerating the whole thing.
20. Pause every scheduled task you don’t read. Automations you ignore are still burning tokens in the background.
21. Speak your prompts. The author uses Wispr Flow for this. Talking lets you give richer context in one go, so you need fewer follow-ups.
22. In Claude Code: one task, then /clear. Clearing the context between tasks stops old code and logs from tagging along.
23. Split your day into 3 sessions. Spreading your work out gives your short-term usage window time to recover in between.
24. Ask Claude for images. But rarely. Visual output is heavy, so save it for when it really counts.

🗓️ The Sunday Max Day Trick

Now for the author’s favorite rule, and honestly mine too. According to the post, limits reset every Sunday at 5 pm.

So the original poster made Sunday their “Max day.” They go all out with Claude on Extra or Max effort and use up every token they have left before the reset. Any tokens you don’t use by the reset are gone anyway, so you might as well spend them.

I love this because it flips the whole mindset. You stay careful from Monday to Saturday, then spend what’s left on Sunday on the big, ambitious tasks you’ve been putting off.

One tip from me: check when your own account resets before you plan around it. Then save your heaviest work, like big refactors, long research or full drafts, for that window.

🚀 Where I’d Start

Don’t try all 24 rules tomorrow. If I had to pick three quick wins, I’d go with these:

• Default to Medium effort and only move up when a task really needs it.
• Start fresh chats often and carry over a 10-line summary.
• Turn off connectors you aren’t using and trim your custom instructions.

Those three alone fix most of the hidden token drain. After that, add the other habits one at a time.

Check out the full LinkedIn post for Anthropic’s exact 2 sentences and screenshots showing where to click for every rule.