How to shorten a prompt
- Paste a long prompt, system message or set of instructions.
- Choose which rules to apply: collapse whitespace, remove politeness phrases and wordy expressions, drop filler words, strip Markdown, remove repeated sentences, turn sequences into a list, or use short forms.
- Compare the estimated token counts, read the compressed version carefully, edit it if needed and copy it.
When prompt compression helps
Every token in a prompt costs money and time, and system prompts, few-shot examples and instructions are sent again with every request. Trimming a 1,500-token system prompt by a third saves that many tokens on every call, which adds up quickly for a busy chatbot or batch job. Shorter prompts can also be clearer: a model follows "Create a 3-day Lisbon itinerary under €500 with restaurants and evening activities" just as well as a paragraph of pleasantries and repetition.
Politeness is not harmful, and modern models do not need it, but some phrasing does matter. Keep words that change meaning, such as "only", "never", "exactly" and "do not", which this tool never removes. Compression is rule-based and transparent, so you can see every change, unlike methods that use a second model to rewrite prompts. It works best on instructions in English; do not run it over code, data or text that the model must quote exactly. Token counts are estimates; use our Token Counter for exact OpenAI numbers. Everything happens in your browser.
Does removing "please" change the answer?
Rarely in a meaningful way. Clear, specific instructions matter far more than politeness for answer quality.
Can it break my prompt?
It can, for example by shortening text inside quotes. Always read the result before using it.
How much can I save?
Typically 10–30% on conversational prompts, less on prompts that are already concise.