Regular expressions for Telegram moderation: examples

Phrase moderation covers the overwhelming majority of cases and needs nothing but words. But there is a class of violations that words cannot describe: a pattern rather than a text. A card number. An invite link with an arbitrary tail. A word where every other letter has been swapped for a lookalike.

That is what the Regex block is for.

Where to set it up#

Moderation section → the Regex card. Rules go into sets, like phrases, and a set has its own punishment chain.

The block is part of the Base plan; Limited does not have it.

A rule has a name, the expression itself and flags. Give it a human name — it ends up in the moderation log, and in a month "Rule 4" will tell you nothing.

Ready-made examples#

Below are the ones people ask for most. Test them against your own messages before switching them on: someone else's chat is not your chat.

A bank card number — sixteen digits in groups, with any separators:

\b\d{4}[ \-]?\d{4}[ \-]?\d{4}[ \-]?\d{4}\b

A private Telegram invite link — the kind a domain rule cannot catch, because the domain is legitimate:

t\.me/\+[A-Za-z0-9_-]{10,}

An international phone number:

\+\d{1,3}[ \-(]?\d{2,4}[ \-)]?\d{2,4}[ \-]?\d{2,4}

Shouting in caps — several long uppercase words in a row:

[А-ЯЁA-Z]{5,}(\s+[А-ЯЁA-Z]{3,}){3,}

A word with substituted letters — the classic way around a stop-word list, where Latin letters are swapped for Cyrillic lookalikes:

[сc][rг][уy][рp][тt][оo0]

A TRON wallet — the USDT addresses spammers drop into chats:

\bT[A-Za-z1-9]{33}\b

How not to shoot yourself in the foot#

A regular expression knows no context and catches exactly what is written. Three rules that save nerves.

Test on real messages. Take a dozen actual messages from your chat — both spam and ordinary ones — and run the rule against them before switching it on.

Use word boundaries. \bspam\b will not touch "antispam" and "spammer", while a bare spam will.

Start with a mild punishment. Keep the set on "delete and warn" at first, look at the moderation log after a couple of days, and only then add mutes.

And a general thought: if a task can be solved with a phrase rule, solve it with one. Phrases are readable, a colleague can edit them, and they do not break over a stray bracket. Regex is worth reaching for when phrases genuinely fall short.

How regex differs from phrases#

A phrase rule works on normalised text: before checking, spaces, emoji and punctuation are stripped from the message and substituted Latin characters are mapped back to Cyrillic. That is why "с п а м" and "спам" are the same thing to it.

Regex looks at text much closer to its original form — that is both its strength and its weakness. The strength: you can describe structure that normalisation would have erased. The weakness: evasion tricks that phrases handle by default have to be anticipated by hand here.

If regex feels like overkill, start simpler: how to protect a Telegram group from spam.