What is token optimization?

What is a token?

Language models read text in small pieces called tokens. A token can be a whole word, part of a word, a space or a punctuation mark. Every provider splits text with its own tokenizer, so the same sentence can be a different number of tokens in ChatGPT, Claude or Gemini. As a rough rule, one token is about four characters of English text.

What actually uses tokens?

Your visible prompt is only one part. A chat also sends earlier messages, attached files, system instructions and tool results, and the answer the AI writes uses tokens too. Some products count messages or credits instead of tokens. That is why a shorter prompt helps most when you write many prompts, and why it is never a promise of a lower bill.

Four ways to spend fewer tokens

• Trim the wording. Remove greetings, polite wrappers and filler words. This is what Typo's token language does.
• Send less context. Paste the section that matters, not the whole document. Point to the exact file instead of the whole folder.
• Reuse context. Keep background you repeat in a Project or in custom instructions instead of pasting it every time.
• Ask once, clearly. A prompt that says what you want, in which format and with which limits avoids a second round.

What Typo's token language does

It is a second language inside Typo, the autocorrect app for Windows and Mac. When you write in an AI tool, Typo first fixes your spelling. Then the token language removes filler from the corrected text at the level you chose, from 1 (tidy spaces only) to 10 (telegraphic). Level 3 is the default. Press Backspace right after a change to undo it.

Why shorter is not always better

A shorter prompt that loses a number, a limit or a “not” costs more in the end, because you have to ask again. Typo is built around that: protected text is kept byte for byte, words like not, never, only and must are never added or removed, and every rule is checked before it is kept. If a check fails, the rule is dropped for that text.

Try it

Paste your own prompt into the Platform Lab, pick your AI tool and a level, and see exactly what changes. It runs in your browser. Nothing you paste is sent anywhere.