19 Laws to Stay Under Your Claude Limit
Smart token strategies — stay efficient, get better results, never hit the wall
⚡ Token Limit Mastery
Token limits are real. Plan around them with these 19 laws — smart usage beats brute force every time.
19
Laws
Prepare Before You Upload
Do This
Convert and clean files first. Paste text instead of raw attachments.
Example
PDF report (12 pages) → Converted text summary
Token Impact
~12,000 → ~2,000 tokens (83% reduction)
Think First, Build Second
Do This
Map the plan in Chat before jumping into doing.
Example
Building plan first → 500 tokens vs. jumping in → 5,000+ tokens
Token Impact
Saves ~4,500 tokens per major task
Ask Better Questions
Do This
Be clear and specific about what you need. Add context or examples.
Example
"Summarize Q1 results" (vague) vs. "Summarize Q1 results with 3 key insights and risks."
Token Impact
~200 vs. ~800 tokens saved per request
Start With the Real Problem
Do This
Address the root cause, not just the symptom.
Example
"Fix the error" (surface) vs. "Why is this error happening and how to prevent it?"
Token Impact
~300 vs. ~1,200 tokens (75% reduction)
One Request, Full Picture
Do This
Put everything in one message. Don't split into multiple follow-ups.
Example
3 follow-ups: ~600 + 600 + 600 = ~1,800 / One full request: ~900
Token Impact
Saves ~900 tokens
Reuse Before You Rewrite
Do This
Keep what works. Swap only the parts that need changing.
Example
Rewrite whole prompt: ~800 tokens / Reuse structure, change part: ~200 tokens
Token Impact
Saves ~600 tokens each time
Polish Before You Send
Do This
Edit your input for clarity and focus. Remove fluff.
Example
Original draft: ~1,200 tokens / Polished version: ~600 tokens
Token Impact
Saves ~600 tokens (50% reduction)
Choose the Right Tool
Do This
Pick the best model or mode for the task.
Example
Haiku for quick fact: ~200 tokens / Opus for deep analysis: ~1,500+ (tokens vary by depth)
Token Impact
Using the right tool can save ~1,000+ tokens
Keep It Short, Sharp, and Clear
Do This
Cowork tends to read this every time.
Example
Long description: ~400 tokens / Concise version: ~100 tokens
Token Impact
Saves ~300 tokens per request
Start Fresh When Needed
Do This
If things go off track, start a new chat instead of dragging the old one.
Example
Bloated chat (50k tokens history) vs. New chat with summary (1k)
Token Impact
Saves ~49,000 tokens
Summarize Often
Do This
Every 15–20 messages, summarize key points and decisions.
Example
20 messages raw: ~10,000 tokens / Summary: ~1,500 tokens
Token Impact
Saves ~8,500 tokens
Don't Upload Unnecessary Files
Do This
Only share files essential to the task at hand.
Example
Whole folder (15 docs): ~20,000 tokens / Only needed doc: ~2,000 tokens
Token Impact
Saves ~18,000 tokens
New Topic, New Conversation
Do This
Switch topics? Start a fresh chat.
Example
Continuing old chat: +10,000 tokens of history / New chat: ~0 tokens of history
Token Impact
Saves ~10,000 tokens
Turn Off What You Don't Need
Do This
Disable web search, plugins, or extras unless necessary.
Example
Search on: +500 tokens per query / Search off: ~0 tokens
Token Impact
Saves ~500 tokens per search
Use Projects for Recurring Work
Do This
For repeat tasks or topics, use Projects instead of starting over.
Example
New chat each time: ~3,000 tokens of re-explaining / Project memory: ~500 tokens of context
Token Impact
Saves ~2,500 tokens per session
Set Preferences & Trim Memory
Do This
Memory can waste tokens. Set preferences once and keep it lean.
Example
Long memory: ~5,000 tokens / Trimmed memory: ~500 tokens
Token Impact
Saves ~4,500 tokens
Don't Ask for The Impossible
Do This
Stay realistic and specific. Claude can't break the rules of physics (or reality).
Example
"Build me a time machine" vs. "Analyze time travel concepts in sci-fi"
Token Impact
Avoids wasted tokens (~500+) on impossible requests
Use Real Examples & Data
Do This
Real-world examples, data, and references help Claude think like you.
Example
Vague request: ~300 tokens / With real data: ~800 tokens (but better output, fewer retries)
Token Impact
Better inputs = fewer retries (avoid 2–3k tokens of back/forth)
Know the Limits. Work With Them.
Do This
Token limits are real. Plan around them and use strategies that stretch every token. Smart usage beats brute force.
Example
Blowing limit (restarts): ~10,000+ wasted / Smart planning: stays under limit
Token Impact
Saves ~10,000+ tokens and time
Key Principles at a Glance
Before You Send
- → Prepare & clean your files first
- → Plan before you build
- → Ask specific, rooted questions
- → Polish your prompt, remove fluff
During a Session
- → One complete request beats follow-ups
- → Summarize every 15–20 messages
- → New topic = new conversation
- → Disable unused tools & search
Long-Term Habits
- → Use Projects for recurring work
- → Trim memory, keep only what matters
- → Start fresh when a chat is bloated
- → Real examples = fewer retries