Sponsored by
Reading time: 6 minutes / Become my affiliate / Sponsor this newsletter
Greetings from above,
Why did the Claude user call tech support? Because they burned through their entire daily limit asking "Can you say that again but better?" fourteen times.
ALEX'S STORY: Last month I was deep into a client project. Building a full content calendar, SEO briefs, ad copy. The kind of session where Claude becomes your second brain.
By 2 PM, I hit the limit. Dead stop. Mid-workflow. The brief was half-done, the copy unfinished, and the client wanted everything by 5 PM.
I switched to ChatGPT. It didn't have the context. The outputs were off. I spent two hours rebuilding what Claude already understood. A total waste.
That night I sat down and mapped out exactly where my tokens were going. The answer? About 80% of them were burned on things that didn't need to be burned. Long chat threads. Repeating context. Using Opus when Haiku would have nailed it.
I rebuilt my entire Claude workflow around 7 rules. Haven't hit a limit since.
Today's newsletter will show you:
Why your Claude limit runs out by lunch (it's a math problem, not a pricing problem)
7 rules that cut my token usage by 80% without reducing output quality
2 copy-paste prompts that save thousands of tokens per session
Let's build your competitive advantage!
How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads
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.
Here's what most people don't realize about Claude.
Every time you send a message, Claude re-reads your entire conversation from the top. Your first message might cost 200 tokens. But by message 30? That same simple question costs 50,000+ tokens. Because Claude is processing every previous message again.
Long chats are expensive chats. That's not a bug. It's how the model works.
Your Pro plan gives you roughly 45 messages per 5 hours on Opus (Claude's strongest model). Burn through those on a single bloated thread and you're locked out. But spread your work across shorter sessions and cheaper models? You can get 200+ quality outputs per day.
The difference between hitting your limit at lunch and never hitting it at all comes down to 7 habits.
📚 RULE 1: EDIT YOUR LAST MESSAGE. DON'T SEND A NEW ONE. 📚
This is the single biggest token saver. Most people don't know it exists.
When you send a follow-up like "Actually, make it shorter" or "Try again with a different tone," Claude processes the entire conversation history plus your new instruction. That's expensive.
Instead: click the edit button on your last message. Rewrite it with the correction built in. Claude re-runs from that point only. You skip the compounding token cost entirely.
The math: In a 20-message thread, editing instead of following up can save 30,000 to 50,000 tokens per correction. Do that 5 times in a session and you've saved an entire conversation's worth of capacity.
📚 RULE 2: START FRESH EVERY 15 MESSAGES 📚
After about 15 to 20 exchanges, your conversation becomes a token furnace. Every new message carries the weight of everything before it.
The fix: when a thread gets long, copy the key output or summary, open a new chat, paste the context you actually need, and keep going.
Think of it like clearing your browser tabs. You don't lose the work. You lose the dead weight.
Pro tip: Use the prompt below (Prompt #1) to create a compressed summary before you close a long thread. That way you carry forward everything important in 200 tokens instead of 50,000.
📚 RULE 3: BATCH YOUR QUESTIONS 📚
Don't ask one question per message. Claude charges you the full context window every time you hit send.
Instead, combine related questions into a single, structured message. Five questions in one message costs a fraction of five separate messages.
Bad approach:
"What's a good subject line?"
"Now write the intro."
"Can you add a CTA?"
Better approach:
"Write an email with: (1) a subject line under 5 words, (2) a 3-sentence intro that opens with a pain point, (3) a CTA button that creates urgency. Output all three sections."
📚 RULE 4: USE PROJECTS FOR RECURRING WORK 📚
Claude Projects let you attach instructions, files, and context that persist across every chat inside that project. The context loads once. It doesn't re-process each time like a regular chat would.
If you do the same type of work repeatedly (writing newsletters, analyzing data, creating social posts), set up a dedicated project. Add your brand voice guidelines, templates, and reference docs once.
Bonus: this also includes custom instructions and memory. Turn both on inside your project. Claude remembers your preferences without you retyping them every session.
📚 RULE 5: PICK THE RIGHT MODEL FOR THE JOB 📚
This is where most people waste the most tokens without realizing it.
Claude has three model tiers. Each one costs different amounts of your usage limit:
Haiku (fast, cheap): Perfect for brainstorming, simple rewrites, grammar checks, quick questions, formatting. Uses roughly 1/10th the capacity of Opus.
Sonnet (balanced): Great for drafting content, analysis, coding tasks, moderate complexity work. The sweet spot for daily work.
Opus (the heavyweight): Reserve this for complex strategy, long-form writing, nuanced reasoning, multi-step workflows. This is your premium fuel.
If you're using Opus to check grammar, you're driving a Ferrari to the mailbox. Switch to Haiku for simple tasks and your daily capacity instantly triples.
📚 RULE 6: FRONT-LOAD YOUR CONTEXT 📚
Give Claude everything it needs in the first message. Role, task, constraints, format, examples. All of it. Upfront.
When you drip-feed context across multiple messages ("Oh, I forgot to mention it's for B2B" ... "Also, keep it under 200 words"), each follow-up reprocesses the entire growing thread.
One detailed first message beats five clarification messages every time. Not just for token savings. The output quality is better too.
📚 RULE 7: PACE YOUR SESSIONS 📚
Claude's rate limit resets every 5 hours. If you blast through 45 Opus messages in 90 minutes, you're locked out for the remaining 3.5 hours.
Spread your heavy work across the day. Do your Opus-level strategy work in the morning. Switch to Haiku and Sonnet for afternoon tasks. Save a few Opus messages for evening review.
Don't sprint. Pace yourself. Your daily output will actually increase because you won't waste time waiting for resets.
⚙ PROMPT #1: THE CONVERSATION COMPRESSOR ⚙
💡 What this does: When a thread gets long, use this prompt to compress everything important into a tight summary. Copy the output, open a new chat, paste it as your opening message. You keep the context. You ditch the bloat.
#CONTEXT:
You are a Conversation Compression Specialist. Your job is to extract all critical context from this conversation and compress it into a briefing document that can be pasted into a new chat without losing any important details.
#ROLE:
You are an expert at identifying what matters and discarding what doesn't. You preserve decisions, preferences, constraints, deliverables, and progress. You discard small talk, failed attempts, and redundant exchanges.
#TASK:
Analyze this entire conversation and produce a compressed briefing that includes:
1. Project Context: What we're working on and why
2. Decisions Made: Every choice or direction confirmed
3. Current Progress: What's been completed so far
4. Active Constraints: Rules, preferences, and specs
5. Next Steps: What needs to happen next
6. Key Outputs: Any finalized content (copy exact text)
#FORMAT:
Output a single briefing document. Use headers and bullets.
Keep it under 500 words. Prioritize specifics over summaries.
If exact wording was approved, include it verbatim.
#TASK CRITERIA:
- Do NOT summarize vaguely ("we discussed marketing")
- DO include specifics ("Subject line approved: 'Free guide'", target audience: B2B SaaS founders, tone: casual")
- If files or data were referenced, note what they contained
- Flag anything unresolved or needing a decisionInput needed:
Nothing extra. Just paste this prompt at the end of any long conversation.
Output you'll get:
A tight briefing document (under 500 words) that captures everything. Paste it into a fresh chat and pick up exactly where you left off. No lost context. No wasted tokens rebuilding.
⚙ PROMPT #2: THE BATCH REQUEST TEMPLATE ⚙
💡 What this does: Instead of asking Claude one question at a time (expensive), this template structures multiple requests into a single efficient message. Use it whenever you have 3+ related tasks.
#CONTEXT:
I need you to complete multiple related tasks in a single
response. Process all sections below and output each one
clearly labeled.
#ROLE:
[INSERT ROLE - e.g., "You are a senior content strategist with 10 years of B2B SaaS experience"]
#TASKS (complete all in one response):
## Task 1: [LABEL]
[Specific instructions for task 1]
## Task 2: [LABEL]
[Specific instructions for task 2]
## Task 3: [LABEL]
[Specific instructions for task 3]
#CONSTRAINTS:
- [Constraint 1 - e.g., "Keep each section under 200 words"]
- [Constraint 2 - e.g., "Use casual tone, no jargon"]
- [Constraint 3 - e.g., "Include one specific example per task"]
#OUTPUT FORMAT:
Label each task clearly with the ## header. Separate sections
with a line break. Complete ALL tasks before stopping.Customize these variables:
ROLE: Match it to your industry. The more specific, the better the output.
TASKS: Add 3 to 7 related tasks. Keep them in the same domain for best results.
CONSTRAINTS: Add word limits, tone requirements, and format specs here. Not in each task.
📋 SUMMARY 📋
Edit, don't follow up. One habit that saves 30,000 to 50,000 tokens per correction.
Start fresh every 15 messages. Compress and carry forward. Kill the bloat.
Batch your questions. Five questions in one message beats five separate messages.
Use Projects for recurring work. Context loads once. Doesn't reprocess.
Match the model to the task. Haiku for simple work. Opus for heavy lifting.
Front-load your context. One detailed message beats five clarifications.
Pace your sessions. Spread Opus across the day. Don't sprint.
📚 FREE RESOURCES 📚
📦 WRAP UP 📦
What you learned today:
The token math is working against you - Claude re-reads your entire thread every message. Longer chats cost exponentially more.
7 habits separate power users from everyone else - Edit messages, batch questions, pick the right model, and start fresh before threads get heavy.
Two prompts keep your workflow lean - The Conversation Compressor and Batch Template let you carry context without carrying cost.
You're paying for Claude Pro. You should be getting Pro-level output all day. Not hitting walls by lunch.
No more burning tokens on bloated threads.
You now have a system to get 200+ quality outputs per day from the same plan everyone else is complaining about.
And as always, thanks for being part of my lovely community,
Keep building systems,
What did you think about today's edition?
🔑 Alex from God of Prompt 🔑
P.S. What's the one Claude task you waste the most tokens on? Reply and tell me. I'll build a dedicated prompt system for the top 3 answers next week.
