T | TERSE | ISSUE 005 · JULY 2026 |
|
THE TERSE LETTER · 3 MIN READ Markdown is quietly
inflating your billEvery #, *, and ``` is a token. In long outputs, they pile up fast. |
Markdown makes answers readable — but every formatting character is a token you pay for. Headers, bold markers, bullet symbols, code fences: they're invisible to you and very visible to the meter. In a long, heavily-formatted response, syntax alone can be 8–12% of the output tokens. |
Tables and nested lists are the worst offenders.A markdown table spends tokens on every pipe and dash. Deeply nested bullets repeat indentation and markers on every line. Triple-backtick code blocks add fences around content you could read raw. |
Long formatted answer · output tokens | Full markdown | 4,200 tokens |
| → Across heavy output days, that's about $90/year. |
|
|
Keep markdown for humans. Strip it for machines.Reading the answer yourself? Keep the formatting. Piping output into another tool, a script, or a follow-up prompt? Ask for plain text — or let Terse's aggressive mode strip markdown automatically. |
✗ TOKEN-HEAVY ### Headers everywhere **bold** **everything** Nested bullet trees ```fenced blocks``` |
| ✓ LEAN OUTPUT Plain paragraphs Emphasis by words Flat short lists Inline snippets |
|
|
Terse aggressive mode strips markdown from anything you feed back into the model. |
Three things to do right now. |
1 | Ask for plain text on pipes When output feeds a script, add "respond in plain text, no markdown" to the prompt. |
2 | Use aggressive mode for re-prompts Pasting a formatted answer back in? Terse strips the syntax first. |
3 | Skip tables for small data Two or three values don't need a table. A sentence is cheaper and just as clear. |
|
macOS · Free forever tier · 1,500 optimizations/week |
|
NEXT WEEK The free-tier math. 1,500 optimizations a week sounds abstract. I'll show exactly how many dollars that is for a typical Claude Code user — it's more than the paid plans cost. |
|
Until then — stay terse. — James
Builder of Terse · terseai.org |
