22 Claude Rules to Stop Hitting Your Weekly Limit

Few things are more annoying than hitting your Claude weekly limit on a Wednesday afternoon. You’re halfway through a project, things are going well, and then everything stops. I’ve been there more times than I’d like to admit, so this post caught my eye right away.

An AI professional on LinkedIn shared a no-nonsense list of 22 rules for using Claude without burning through tokens. The hook was bold: “I don’t think you should use Claude anymore.” The second line explains it: not until you’ve read these rules. The author is clearly frustrated with how much people waste, and the fixes are simple.

I went through every rule and added some context so you can see why each one works. Some will feel obvious. A few surprised me.

⚙️ Pick the Right Model and Effort

The first group of rules covers model choice and effort level. These settings drive token use more than almost anything else.

1. Use Medium effort. Never Max. The creator’s breakdown: Low for grammar, reformatting and translations. Medium for everything else. High for super-ambitious work. Never Extra and never Max. More effort means more thinking tokens, and most tasks don’t need them.
2. Save your free reset for a bad day. If you get a limit reset, don’t spend it on a normal Tuesday. Keep it for a deadline or a day when everything goes wrong.
3. Use Opus 5.5. Every chat. Never Fable. The author’s take is that Opus 5.5 handles daily work well. Reaching for a heavier model by default eats your allowance much faster.
4. Never change model between turns. Switching models mid-chat means the new model has to process the whole conversation again. You pay for the same context twice.

💬 Run Your Chats Smarter

Once your settings are right, the next big leak is how you handle conversations. Every message in a chat sends the full history again, so long and messy threads cost a lot.

5. Never come back to an old chat. Old chats carry a lot of history. Adding one more question means Claude rereads all of it. Start fresh and paste in only what matters.
6. Add “No web search. No files.” to your questions. If you just need a quick answer, say so. That stops Claude from reaching for tools that pull a pile of extra content into the conversation.
7. Plan in text. Ask for the file last. Agree on the outline and the details in plain chat first. Generating a full document over and over while you’re still deciding is expensive.
8. Highlight. Never ask for a full redo. If one paragraph is off, point to that paragraph. Asking for a full rewrite regenerates everything, including the parts you already liked.
9. Load your apps on demand. Connected apps and tools add to the context. Turn them on only when a task needs them.
10. Use a short custom instruction. The original poster recommends pasting a specific two-sentence prompt into your Claude instructions. The wording is only shared through the post itself, so check it there. The idea is that a tight, well-written instruction does a lot of work for very few tokens.

📎 Trim What You Feed It

Inputs are the next area to fix. Images and documents can be surprisingly heavy, and background tasks quietly drain your allowance.

11. Crop every screenshot. Convert every PDF. A full-screen screenshot is full of pixels Claude doesn’t need. Crop it to the part that matters. Turn PDFs into plain text when you can.
12. Pause every scheduled task you don’t read. Automated reports you never open still use tokens. Be honest about which ones you actually look at.
13. Start every task with “ask me questions.” This one surprised me. Letting Claude clarify first prevents wrong first drafts, and wrong drafts are where a lot of waste happens.
14. Put all your asks in one message. Five separate messages mean five rounds of rereading the conversation. Batch your requests.
15. Speak your prompt. Don’t type. Talking lets you share more context faster. Richer first prompts mean fewer back-and-forth fixes later.

🧹 Keep Things Clean

The last batch covers habits. These are the ones that separate people who run out on Wednesday from people who never notice the limit.

16. Never correct Claude. Start over. When a chat goes sideways, every correction adds more confusing history. A clean restart with a better prompt usually costs less.
17. Stick to one topic or goal per chat. Mixing topics bloats the context and muddies the answers.
18. Keep your instructions under 2,000 words. Custom instructions load with every message. A huge instruction block is a tax you pay on every single turn.
19. Put every file you reuse in a Project. Stop uploading the same brand guide or reference doc again and again. Keep it in one place.
20. In Claude Code: one task, then /clear. The /clear command wipes the session context. Finish a task, clear it, and start the next one fresh.
21. Split your day into 3 sessions. Spreading usage across the day works better with how usage windows reset, so you’re less likely to hit a wall all at once.
22. Ask Claude for images. But rarely. Visual output is handy, but it’s costly. Save it for when you really need it.

Why this matters: most token waste doesn’t come from hard tasks. It comes from small habits like long old chats, uncropped screenshots, model switching and endless corrections. Fix the habits and your limit lasts much longer.

🚀 Where to Start

You don’t need all 22 rules on day one. If I had to pick three, I’d start with Medium effort as your default, fresh chats instead of old ones, and “ask me questions” at the start of every task. Those three alone should change how far your weekly allowance goes.

Then layer in the rest. Crop your screenshots. Batch your asks. Use /clear in Claude Code. It all adds up.

What I like about this list is how practical it is. The author doesn’t tell you to use Claude less. The point is to use it with less waste, and that’s a much better deal.

Check out the full LinkedIn post for the original rules and the exact two-sentence instruction the creator mentions in Rule 10.