What the filter is for
Every group has words it does not want. For one it is swearing, for another it is the name of a scam token, for a shop community it might be a competitor's name. Telegram has no word filter of its own, so this job falls to a bot that reads group messages and removes the ones that break the rule.
Build the list with categories
In the forbidden words module every word or phrase you add gets a category: profanity, hate speech, insults, adult content, spam, scams, crypto scams, drugs, violence, competitors or custom. Categories keep a long list readable, and each word can carry a severity from low to critical so you can see at a glance which matches matter most.
Choose how each word is matched
The match mode decides what counts as a hit, and it is where most false alarms are prevented:
- Whole word: matches the word on its own, so a short word does not fire inside a longer, innocent one. A good default.
- Exact and contains for stricter or looser matching.
- Pattern: wildcards for families of spellings.
- Regular expression, for admins who know them.
- Leetspeak: catches the word written with numbers or symbols in place of letters.
Pick the action
For each word you choose what happens when it appears. Delete removes the message. Warn hands the case to the warnings module, so repeat offenders reach your warning limit and its automatic action. Mute, kick and ban remove the message and deal with the sender. A new word starts with delete as its action, which is a safe default for most lists.
Test before it acts
Before switching the module on, paste a few real messages from your group into the test box on the module page. It shows which words would match. If an innocent message is caught, change that word's match mode to whole word or remove it. Settings for text normalisation and evasion detection catch extra spacing and symbol tricks.
Turn on the admin exemption so your team can quote a banned word when explaining a rule, and add a notification rule if you want to hear when a forbidden word is detected.