Proofreading is both a tedious and difficult task. It is expensive to hire a professional to do the job, and it may be also difficult to find someone competent. Newspapers skip proof readers to save money, and most writers try to do the job themselves. The consequences are clearly seen from errors in text written by professionals to incomprehensive e-mails and blogs. This creates misunderstandings, offers false data, and, in the worst cases, may give a very bad impression. This paper presented an automatic proof reading tool that works in any language. The principle is to compare the sentences offered by the user to similar sentences in a large text repository and then offer suggestions for improvement in the cases where the user's choice gets a zero or low frequency and where other alternatives get higher scores. The prototype that is presented here works on rather small text repositories but can still offer valuable suggestions in many cases. In a product version, large repositories are required. In order to be able to process such amounts of data by the brute force technique that is presented here, a large cluster of computers will be needed.