How to protect a Telegram group from spam

Spam reaches a Telegram group by several routes at once: blanket mailings across every chat, advertising disguised as a real message, fresh accounts that write once and vanish. No single setting closes all of that — but there is no need to build everything at once either.

Below is the order worth switching things on in: from the cheapest and most harmless to the most delicate.

Step 1. Remove the noise that is not spam#

Before fighting spam it is worth clearing out what makes it hard to notice. The Service messages block removes Telegram's notices about joins, leaves and other housekeeping events. In an active chat that is half the feed.

It breaks nothing, works on any plan and takes a minute to set up. There is a separate walkthrough: how to remove "joined the group" messages.

Step 2. The simple blocks#

Two blocks that take out a noticeable share of the junk without knowing a single word of your chat:

• Text length — cuts off empty messages and one-word spam, and walls of text while it is at it;
• Duplicates — catches both one member repeating themselves and the same text sent at once by different accounts. The second is a direct hit on coordinated mailings.

Neither needs maintenance: switch on and forget.

Step 3. Links#

A link is a spammer's main instrument. Decide which mode suits you: an allowlist if links are not really needed in your chat, or a blocklist if they are but junk domains have to go.

Modes, subdomains and different punishments are covered in how to block links in a Telegram group.

Step 4. Words#

Phrase moderation covers what repeats verbatim: "DM me", scheme names, profanity. A rule is built from parts with connectors and exceptions, so it touches real conversation far less often than a plain stop-word list.

The builder is explained in detail in how to set up a word filter. If you need patterns rather than words, there are ready-made regular expressions.

Step 5. AI moderation#

Everything above catches what you described in advance. Spammers rewrite their text faster than you can edit your sets — that is where a model that looks at meaning comes in.

Ticking the built-in topics is enough: illegal substances, gambling, crypto scams, phishing, job scams, adult spam. A punishment is applied only on high confidence.

If you need to catch what is out of place specifically in your chat, add your own rules in plain words. And to keep that from costing extra — how not to overpay.

Step 6. Punishments#

Without a punishment chain everything above is decorative statistics: the rule finds a violation and does nothing.

For spam sets the chain is short: the very first step is a ban, amnesty off, and deleting all of the member's messages makes sense with the ban. For human lapses like profanity, a ladder with a warning and a mute. Covered in detail in how to build punishments.

What works by itself#

Two things you do not need to configure.

Trust levels. The bot is stricter with new and "empty" accounts: what it is not entirely sure about counts as a violation for them, while the same message from a long-timer goes through.

The allowlist. Chat administrators and members on the allowlist bypass every rule, so your own moderators and helper bots are in no danger.

What not to do#

Switch everything on at maximum strictness. Keep punishments mild for the first few days — delete and warn — and watch the moderation log. False positives show up immediately, and rules get fixed before anyone is banned for nothing.

Configure blind. The moderation log shows which rule caught a message and what the bot did. A couple of visits after enabling a new set saves a week of arguing in the chat.

Rely on a single block. Spam varies: some of it is caught by a domain, some by a word, some only by meaning. Three simple blocks together work better than one complicated one.