Prompt optimization for Claude

Why the whole conversation matters

Claude reads the conversation so far every time you send a message. Long chats, large attachments and stronger models all use more of your allowance. Anthropic's own tips point the same way: start a new chat for a new task, keep reused material in a Project, and be specific about what you want.

Trim the wording, keep the evidence

Removing “Hi Claude”, “could you please” and “thank you so much” changes nothing Claude needs. Removing a quote, a figure or a file name does. Typo's token language keeps quotes, code, links, numbers and file paths byte for byte and only trims the words around them.

Projects for recurring context

If you work on the same material often, add it to a Claude Project once and keep the project instructions short. Each new message can then focus on the current question instead of repeating the background.

Claude Code

In Claude Code, point to the exact file, function and error instead of pasting a whole folder or log. Typo stays out of terminals and code editors on purpose, so it does not shorten what you type in a terminal. For longer instructions, write them in the Platform Lab or in a chat field, check the result and paste it in.

Which level for Claude

Level 3 suits most chats, and 4 or 5 works for long prompts without source material. When a prompt contains quotes or documents, stay at level 3 or below so the words around your evidence read exactly as you meant. Use levels 8 to 10 only for short, simple instructions.

How Typo switches on

Auto mode recognises the Claude desktop app and claude.ai. Backspace right after a change undoes it. On Windows, Ctrl+Alt+T shortens the whole prompt and Ctrl+Alt+Z restores the original.

Token counts on this site are estimates. Claude counts tokens with its own tokenizer, and chat length, files and the model also affect your usage.