A disclosure first: we build Group Ward, one of the bots in this list. So I've tried to be fair: the other bots do things ours doesn't, and I say so, and there's a section on a mistake we made. Every fact about the other bots comes from their own docs, sites and repositories, checked on September 23, 2026. When the docs don't mention something, I write "not documented" rather than "can't".
The short answer. Almost every bot catches spam with links, so pick on convenience. The real difference is scams without links: "Looking for people, remote work, 2 hours a day, DM me." Link filters and word lists don't see those. Of the bots here, only two read the message itself: tg-spam, which you host on your own server, and Group Ward.
1. The three kinds of spam in Telegram groups
Links. "Earn $300 a day 👉 t.me/…", crypto giveaways, "check yourself in @some_bot". The oldest and easiest case: the message carries a link, a bot mention or a forward from a channel. A link filter removes it without any intelligence.
Scams without links. "Hiring for remote work, paid daily, DM me." "I'm an admin, message me about your withdrawal." No link, no banned word. The scammer starts the conversation in the group and does the actual scam in private messages. Catching this needs a bot that understands what the message is trying to do.
Bot raids. Dozens of accounts join within a minute and post the same thing. A join captcha, limits for new members and an "under attack" mode handle this.
A good setup covers all three. Most bots' docs don't mention the second one at all.
2. How to stop spam in a Telegram group without a bot
Before adding anything, turn on what Telegram already has.
- Aggressive Anti-Spam. Telegram's built-in filter for groups with more than 200 members. It has no settings and doesn't publish its rules. You can report its mistakes from Recent Actions. Free, one switch in the administrator settings.
- Slow mode and restrictions for new members, such as no links or media.
- Join requests, if the group is private.
For a small, quiet group this can be enough. You need a bot when spam arrives faster than you can delete it, or when it comes without links.
3. Telegram anti-spam bots compared
| Bot | Link spam | Scams without links | Captcha | Undo a mistake | Price |
|---|---|---|---|---|---|
| Combot | yes, plus the CAS spammer list | not documented | for new members | commands and action log | free and paid plans, prices not published |
| Rose | yes, very flexible locks | only via word lists you write yourself | every new member | commands | free |
| Shieldy | yes, plus CAS | not documented | every new member, can skip older accounts | not documented | free |
| GroupHelp | yes, plus word filters and its own ban list | not documented | for new members, several types | commands | free; Pro 150 Stars once, UltraPro 150 Stars a month |
| tg-spam | yes, plus CAS | yes, own classifier and optional AI | none | buttons in an admin chat | free, but self-hosted |
| Group Ward | yes, plus CAS | yes, AI | one tap on an emoji; suspicious newcomers only by default, can be every newcomer | "Undo" button in your DM | free |
4. Each bot in short
Combot is the best-known group manager. Its strongest feature is the CAS list: if someone has already spammed other groups, they're banned the moment they join. It also has a web dashboard, a filter that screens join requests by profile bio, and an anti-raid mode. The gap for this comparison: reading what a message means isn't documented, and we couldn't find prices for the paid plans.
Rose is free and very flexible. You can lock almost anything: links, forwards, buttons, specific scripts, "zalgo" text. But everything is a rule you write yourself. Its docs are honest about the limit: a blocklist entry for "the admins suck" won't match "these admins suck". You set it up with chat commands; there's no web dashboard.
Shieldy is a simple captcha plus CAS and a block on links to other channels. It works, but it has mostly stopped evolving: the last commit, in February 2026, added legal documents, and an issue titled "bots pass captcha" has had no reply since 2024.
GroupHelp is the biggest of them all: its Telegram page shows about 3 million monthly users, and the bot speaks 10 languages. It covers a lot for free: link and word filters, flood control, a captcha with several types, and its own global list of harmful accounts. Pro costs 150 Stars once for five groups and adds things like an anti-porn filter; UltraPro is 150 Stars a month per group. Everything is configured in menus inside the chat, and reviewers say those menus get confusing on big setups. Like Combot and Rose, it's rule-based: catching scams without links isn't documented.
tg-spam is an open-source project and the strongest in this list at reading message content. It compares text against known spam and can also ask an AI model. By default it checks only a new member's first message. One real downside: you run it yourself with Docker, so it's a tool for admins comfortable with servers.
5. Group Ward: how it works and what we got wrong
We built Group Ward on our own platform, FlowCastle, and the whole bot is an ordinary flow on it. This is how it works:
- Regular members are never checked. Only newcomers are.
- By default nobody is checked on joining. If a newcomer's message looks risky, like a link, the bot asks for one tap on an emoji. You can turn on a check for everyone who joins in the settings.
- Links, hidden-character tricks and spammers on the CAS list are removed automatically.
- AI reads newcomers' messages that have no links and tells a scam from ordinary talk. Free.
- Mute always comes before ban, and every action has an Undo button in the admin's DM.
One mistake of ours. In September we looked into why some groups removed the bot, and found a real problem. Right after install, the bot treated everyone as a newcomer, even people who had been in the group for years. In a Russian gaming chat, the AI flagged a two-word message about a game as spam. In another, it flagged advice to a member: "promote it without a link, message people directly". The bot deleted the messages and muted the senders, and the admins removed the bot within 10–15 minutes. We fixed the first cause: members who were in the group before install are no longer checked. The second remains. The AI can still misread slang, especially "DM me"-style phrases. That's why the first action is always a mute, not a ban, and undoing it takes one tap.
When Group Ward is the wrong choice:
- You need a bot outside Telegram.
- You want chat statistics and a web dashboard: that's Combot.
- You want full control and your own server: use tg-spam.
- The AI checks up to 300 messages a day per group. Above that, the ordinary filters still work.
6. Which one to pick
- Small, quiet group: Telegram's own settings may be enough.
- Spam is mostly links: Rose or Shieldy, free.
- Large community, you want stats and a dashboard: Combot. If you prefer lots of fine-grained rules in chat menus: GroupHelp.
- Scams arrive without links and you don't want to run a server: Group Ward. If you do have a server, look at tg-spam too.
We're preparing a live test: the same set of spam and normal messages for each bot in its own group, with a table of what each one caught and who it wrongly blocked. When it's done, the results will be added here, Group Ward's misses included.
You can try Group Ward here: tap "Add to a group", pick the group, and Telegram asks for the rights it needs.
Run a group and know a bot that should be in this list? Tell us, and we'll add it to the test.