In the Support category, user jimmy0017 asks if Discourse automatically strips unsupported HTML tags used to bypass watched words. They note that users are successfully hiding terms like “discord” and “tampermonkey” by inserting tags (e.g., dis</n>cord), allowing these posts to bypass the review system without triggering moderation alerts. pfaffman responds that watched words are a “crude” feature easily circumvented by determined users, suggesting that such behavior warrants punishment. He further recommends exploring Discourse AI - Spam detection as a more robust solution.
tobiaseigen agrees with pfaffman, emphasizing that watched words cannot stop determined misbehavior and that active moderation is necessary to enforce community rules. jimmy0017 clarifies that their primary goal is preventing users from seeing this content, as they deal with bot registrations and have banned several users for bypassing filters. They are considering the AI spam detection feature but prefer the simplicity of watched words. Finally, pfaffman challenges the idea that watched words are more straightforward than AI, noting that bots are often more sophisticated. He suggests an alternative “straightforward” approach: requiring approval for all posts from new users (TL0/1) to prevent spam from appearing immediately.
We’ve recently added a couple watchwords into our instance.
However, our user found out the ways to around it like below. Does discourse automatically remove unsupported html tag or not? These are actually build into the post and didn’t trigger the review system.
dis</n>cord discord
tampe</big>rmonkey
Our watch word list:
*discord*
*tampermonkey*
Watched words are pretty crude. There are many ways like that to get around them.
If a user does that then they did it it on purpose and you can punish the user accordingly.
I’m afraid you’ll need to look into Discourse AI - Spam detection
@jimmy0017 did you get an answer that satisfies your needs? Jay’s reply is spot on. Watched words can be overcome by people who are determined to misbehave. You need to use moderation to make sure they know the rules and that they need to abide by the rules in order to be part of your community.
Yeah… But then the problem is we tends to get more “bot” registration as well. We do have a couple moderator on watch for these things. But mainly, we just don’t want our user to see those content in the first place. We’ve banned quite a couple users “forever” for purposely work around the watched words..
We will think about the AI spam detection and see if that’s something we are going to do. Just because this is a more straight forward way to “watch words” than using an AI LLM.
I’m afraid if that were true you’d not have written this post. You’ve got bots who are more clever than the “watched words” feature. Also, it’s a pretty good bet that they are actual humans (though bots may have gotten so much better in the past year that the humans that were creating new accounts have lost their jobs).
The other thing you could do would be to approve all posts from TL0/1 users. If you don’t get lots of new legit users who’ll be sad if their post doesn’t show up immediately, that could be a “straight forward” solution.