Prompt optimization for ChatGPT

What makes a good ChatGPT prompt

OpenAI's own advice is to be specific: say the task, give the context that matters, describe the output you want and name the limits. A short prompt with a vague goal often leads to a second message, and that costs more than a few extra words.

• The task: summarize, draft, compare, fix.
• The context: who it is for and what it is about.
• The output: a table, five bullets, an email under 120 words.
• The limits: what to leave out and what must not change.

Where Typo helps

Typo's token language works on the wording around those parts. It removes greetings and sign-offs, “please” and “could you”, filler like “basically” and “really”, and from level 5 long phrases like “in order to”. Your task, numbers, quotes and links stay as you wrote them.

Use Projects for text you repeat

If you paste the same background every time, put it in a ChatGPT Project with its own instructions and files. Then each new prompt only needs the current task. This is about focus and convenience. It is not a guarantee of lower usage.

Which level for ChatGPT

Level 3 is a good start for everyday questions. Levels 4 to 6 work well for longer prompts with background. Levels 8 to 10 also remove articles, linking words and “is” or “are”. ChatGPT usually copes with that, but read the result before you send anything important, or turn on safe mode.

How Typo switches on

In Auto mode Typo recognises the ChatGPT desktop app and chatgpt.com in your browser, and the token language works as you type, one finished word at a time. Press Backspace right after a change to undo it. On Windows, Ctrl+Alt+T shortens the whole prompt at once and Ctrl+Alt+Z brings the original back.

Token counts on this site are estimates. ChatGPT counts tokens with its own tokenizer, and your plan decides how usage is counted.