🔑 Claude sucks.

Reading time: 5 minutes / Become my affiliate / Sponsor this newsletter

Greetings from above,

Why did the AI user cross the road? To start a new chat because message 30 just ate their entire token budget. (Turns out the real limit was inside you all along.)

ALEX'S STORY: I used to blame Claude for cutting me off mid-workflow. "This AI is broken," I'd tell Robert. I was 28 messages deep, re-pasting the same project context, asking follow-up questions that forced Claude to re-read everything from scratch.

Then I ran the numbers.

Message 30 in a single Claude chat costs 31x more tokens than message

1. Not a typo. Thirty-one times. Every message re-processes the entire conversation history. By message 30, 98.5% of your tokens are spent re-reading old context.

Only 1.5% goes toward your actual question.

I was burning through my daily limit in two sessions. Not because Claude was stingy. Because I was wasteful.

Last week I gave you 7 rules to stop wasting Claude tokens. Today: 10 more.

These aren't beginner tips. These are the advanced habits that took me from "I run out by lunch" to "I haven't hit a limit in months."

Today's newsletter will show you:

Let's build your competitive advantage!

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more

Claude doesn't count messages. It counts tokens. And here's where it gets expensive.

Every time you send a message, Claude re-reads your entire conversation. Message 1 costs 1 unit. Message 2 costs 3. Message 10 costs 55. Message 30? That's 961 units for a single reply.

The total cost grows as a triangle, not a line. Two messages feel cheap. Twenty messages are a budget fire.

This is why you hit your limit at 2 PM every day. Not because Anthropic is cheap. Because long conversations are exponentially expensive.

💡 You already know the basics: edit instead of following up, front-load context, start fresh chats, use Projects. These go deeper.

Habit 1: Set custom instructions once, skip the preamble forever.

Claude's system prompt and custom instructions persist across every conversation. Put your role, industry, preferred output style, and constraints there. That's 200+ words you'll never paste again. Every chat starts pre-loaded.

Habit 2: Write prompts as numbered checklists, not paragraphs.

Prose prompts invite misinterpretation. Misinterpretation means follow-ups. Follow-ups mean wasted tokens. Instead: "Do these 4 things: 1. [X] 2. [Y] 3. [Z] 4. [W]." Claude hits all four on the first try. One message instead of three.

Habit 3: Specify exact output length and format.

"Write a blog post" is an open invitation for Claude to write 1,500 words when you needed 400. Every generated word costs tokens. Be surgical: "Write a 300-word blog post. Use 3 subheadings. No introduction paragraph." You'll cut output tokens by 50% or more.

Habit 4: Turn off features you're not using.

Extended thinking, web search, and artifacts all consume tokens on top of your message. If you're doing a simple text edit, toggle all three off. Thinking mode alone can double your token cost per message. Only activate what the task actually requires.

Habit 5: Ask for structured output instead of prose.

Tables, JSON, bullet lists, and markdown headers are more token-efficient than paragraphs. A comparison table uses roughly 40% fewer tokens than the same information written as flowing text. Plus, structured output rarely needs a "rewrite this" follow-up.

Habit 6: Route simple tasks to Haiku.

Quick summaries, reformatting, basic Q&A, simple code fixes. None of these need Sonnet or Opus. Switch to Haiku for anything that doesn't require deep reasoning. Your premium model quota stays untouched for the work that actually needs it.

Habit 7: Use artifacts as living documents.

When Claude generates an artifact (code, a document, a table), you can edit it in place. Most people ignore this and type "now change the third paragraph to..." as a new message. That re-reads everything. Instead, click into the artifact and ask for the specific edit there. Targeted changes, minimal tokens.

Habit 8: Copy outputs, not conversations.

When you need to continue work in a new chat, don't re-explain the whole project. Copy Claude's last output and paste it as your opening context. "Here's where we left off: [output]. Now do X." You carry forward the result without carrying forward the bloated history.

Habit 9: Split big tasks into phases with clear handoffs.

A 500-word mega-prompt that tries to do research, analysis, AND writing in one shot will fail. Then you retry. Then you clarify. Three attempts at a complex task costs more than three clean single-purpose prompts. Phase 1: research. Phase 2: outline. Phase 3: draft. Each in its own short chat.

Habit 10: Sequence your day by token weight.

Do your heaviest work first. Complex analysis, long-form writing, multi-step strategy sessions. These eat the most tokens. Push lightweight tasks (reformatting, quick questions, brainstorming lists) to the afternoon when your budget is thinner. Treat your token budget like a battery: heavy loads in the morning, light loads at dusk.

What you learned today:

Your Claude limit isn't broken. Your workflow is.
No more rage-quitting at 2 PM because you "ran out of messages."
You now have 10 advanced habits to make every token count.

And as always, thanks for being part of my lovely community,
Keep building systems,

🔑 Alex from God of Prompt

P.S. What's your biggest Claude frustration right now? Hit reply and tell me.

The top answers become next week's newsletter.