T | TERSE | ISSUE 003 · JUNE 2026 |
|
THE TERSE LETTER · 3 MIN READ The politeness tax
on your AI bill"Could you please possibly help me with…" — every soft word is a token you bought. |
We're trained to be polite. "Please," "thank you so much," "if it's not too much trouble." To a human, that's warmth. To a tokenizer, it's dead weight you pay for. It adds up faster than you'd think — and the model doesn't answer one bit better for it. |
Filler can be 15–20% of a typical prompt.Hedges ("I think maybe"), politeness ("would you mind"), and throat-clearing ("I was just wondering if") inflate every message. Across a year of daily use, that's millions of tokens spent on words the model strips out anyway. |
✗ WORDY PROMPT "Could you please…" "I was wondering if…" "if it's not too much trouble" "thank you so much!" |
| ✓ TERSE PROMPT "Refactor this fn" "List the edge cases" "Fix the type error" "Explain line 12" |
|
|
| 02 | The counterintuitive part |
Terser prompts often get better answers.Removing hedging forces clarity. "Maybe we could possibly look at improving this somehow" becomes "Cut this function's runtime." The model has less to guess at — and you get a sharper response for fewer tokens. |
Terse rewrites filler and politeness on the fly — your intent stays, the token cost drops. |
Three things to do right now. |
1 | Cut the opener Delete "I was wondering if you could" and start with the verb. "Write," "fix," "explain." |
2 | Let Terse do it automatically Normal mode strips filler, politeness, and hedging without changing your meaning — set it and forget it. |
3 | Watch what gets removed Terse shows the before/after diff. You'll quickly learn which habits cost the most. |
|
macOS · Free forever tier · 1,500 optimizations/week |
|
NEXT WEEK Your system prompt is 3× too long. Most CLAUDE.md files repeat themselves, over-explain, and bury the rules that matter. I'll show the trim that keeps every instruction and halves the size. |
|
Until then — stay terse. — James
Builder of Terse · terseai.org |
