New Free Tool Trims Your AI Prompts by 40% to Cut Costs

Every time you type a long instruction into an AI chatbot, the company behind it charges by the word — and long messages also take longer to answer. A new research paper from Shamin Chokshi describes a free, open-source tool that automatically trims those instructions before they're sent, keeping the meaning but dropping the fluff. In tests, it removed about 40% of the words on average while the AI's answers stayed almost the same.

The trick is old-fashioned language clean-up rather than new AI. The tool strips out filler words like "basically" and "just," swaps phrases for shorter versions ("in order to" becomes "to"), trims unnecessary adjectives, and protects names and key facts so they don't get mangled. Crucially, it runs on a regular computer processor — no special chips, no second AI model, and no randomness, so you get the same result every single time you run it. A gentler setting that only removes filler words cuts about 30% of the text with almost no change to the answers.

The study tested this across 1,242 real prompts and 11 kinds of tasks, from math problems to casual chat, generating more than 18,000 side-by-side comparisons. The results show a clear trade-off: the more you trim, the more the answers drift. Most tasks hold up well, but questions requiring common sense or everyday reasoning broke down badly when the compression got aggressive — a rare, puzzling answer where a normal one was expected.

For anyone paying for AI by usage — freelancers, small businesses, customer-support teams — this is a straightforward way to cut costs without switching providers. The author published both a free Python package and all the test data so others can check the work. Just don't run it on your trickiest reasoning prompts without checking the output first.