Profanity Filter and Bad Word Checker

Strong profanity, crude insults, mild vulgarity and — flagged separately — slurs that target protected groups. You choose how strict the check is.

Yes. It normalises leetspeak, look-alike letters from other alphabets, repeated and spaced-out characters and common prefixes before matching, so simple bypasses are caught.

By normalizing the text before matching it, because a filter that only compares literal words is defeated by the first person who types an asterisk. Several tricks are undone in turn: leetspeak, where digits stand in for letters; look-alike characters borrowed from another alphabet, where a Latin word is spelled with a Cyrillic letter that renders identically; letters spaced out or separated by punctuation; and repeated characters stretched for emphasis. All of those collapse back to the same word before the comparison happens. What it deliberately does not do is match on fragments, because that is how a filter ends up censoring ordinary words that happen to contain a shorter one.

Because they are different problems and merging them makes the result useless for either. Ordinary profanity is a register question: too strong for a customer email, unremarkable in a message to a friend, and entirely a matter of who is reading. A slur aimed at a group is not a register question at all — it does not become acceptable in a casual context, and a tool that reports "3 flagged words" without saying which kind lets the serious one hide behind the trivial ones. Three tiers are scored apart for that reason: slurs, strong profanity, mild vulgarity. A draft with twelve mild hits and no slurs needs a different decision from one with the opposite.

Yes, and it is worth doing for two opposite reasons. A shared list cannot know your context: a community, a game, an industry or a family has words that are unacceptable there and meaningless anywhere else, and the only way those get caught is if you add them. The other direction matters just as much — a word that is flagged here but is ordinary in your field can be excused, so the report stops crying wolf and you keep reading it. A filter you have stopped trusting is a filter that is no longer working, and the usual cause is not a missed word but too many wrong ones.