1Setup Is Where Most Moderation Goes Wrong
Adding a moderation bot to a Telegram group takes about a minute. Setting it up so it actually helps — without banning real members on day one — takes a little more thought, and that gap is where most first-time admins stumble. They flip on every filter at once, watch the bot mute three regulars in an hour, and switch the whole thing off in frustration.
It doesn't have to go that way. A good setup follows an order: get the permissions right, turn on the cheap and certain protections first, and — most important — let the bot watch before it acts. This guide walks through that order step by step, so by the end you have a bot you trust rather than one you're fighting.
2Step 1: Give the Bot the Right Admin Rights
A moderation bot can only do what its admin permissions allow. Add it to your group, promote it to admin, and grant exactly the rights its job requires — no more. For a moderator, that usually means the ability to delete messages, ban and restrict users, and (if you want entry gating) invite or approve members. Leave everything else off.
Resist the temptation to make the bot a full admin "to be safe." Broad rights don't make moderation better; they just widen the blast radius if the bot is ever misconfigured or compromised. Least privilege is the same principle you'd apply to any account with power over your group.
- Delete messages — the core of removing spam and rule-breaking content.
- Ban and restrict users — mute, kick, or ban when a rule is broken.
- Manage new members — needed only if you're gating entry with approval or verification.
- Skip the rest — don't grant pin, promote, or change-info rights a moderator never uses.
3Step 2: Turn On the Cheap, Certain Protections First
Not all defenses are equal, and you don't enable them all at once. Start with the layers that are fast, high-confidence, and almost never wrong about a real member. A shared spammer blocklist is the obvious first choice: connecting to a database like CAS auto-catches accounts already flagged across thousands of other communities, before they post a single word — with very few false positives.
Next come basic content filters and entry verification. Simple link and keyword rules absorb the high-volume, obvious spam, and a lightweight CAPTCHA at the door stops automated accounts from joining in the first place. These are the workhorses; save the nuanced, AI-driven checks for the following step, once the fundamentals are running cleanly.
- Shared blocklists first — near-zero false positives, catches known spammers on sight.
- Link and keyword filters — cheap rules for the obvious, high-volume cases.
- Entry CAPTCHA — stop automated joins before they reach the chat.
4Step 3: Watch Before You Enforce
This is the step that separates a bot you trust from one you resent, and it's the one most people skip. Before letting the bot ban or delete anything, run it in a monitoring mode that logs what it would have done without actually doing it. Let it observe your real community for a few days and produce a record of every action it would have taken.
Then review that log against your own members. Did it flag a regular's normal message as spam? That's a false positive you want to catch now, while it costs nothing, rather than after a real person has been banned. Once the log looks right, switch on enforcement one layer at a time — high-confidence blocklists and rules first, AI-based actions last. This is exactly how Telm is built to work: it starts in Monitoring Mode so you see its judgment before it ever acts.
5Step 4: Layer In AI and Fine-Tune
With the basics enforced and trusted, add the layers that catch what rules miss. AI classification judges a message by context and intent, catching reworded scams no keyword list anticipated, and OCR reads text hidden inside images and screenshots so picture-based scams aren't invisible to your filters. Because these run only on messages the cheaper layers didn't already resolve, they stay fast and accurate.
Setup isn't a one-time event. Keep an eye on the moderation log for the first couple of weeks, mark any mistakes, and adjust thresholds where the bot is too aggressive or too lax. A moderation bot is a tool you tune to your community, not a switch you flip once and forget.
- Enable AI classification for context-aware detection of novel scams.
- Turn on OCR so image-based spam isn't a blind spot.
- Review the log weekly at first; correct false positives and adjust thresholds.
6A Bot You Trust, Not One You Fight
The setup that works isn't complicated, but the order matters: grant the minimum admin rights the bot needs, turn on the cheap and certain protections first, let it observe before it enforces, then layer in AI and tune as you go. Skip the observation step and you get a bot that punishes real members; follow it and you get one that quietly does its job.
Done this way, the whole thing takes an afternoon and pays off every day after. Your members stop seeing scams, your evenings stop being spent deleting them, and the bot earns your trust the honest way — by proving itself before you hand it the keys.