AI Spams and Comment Moderation

by Rian

Many websites have some common unwritten rules on comment moderation — most common of which is locking or closing comments after certain duration. I believe Reddit has it at 6 months, under some of the news articles they have it at 14 days, and some chatting-like discussions close it up faster after 48 hours. The Mad Tea Party is now closing comments (Trackbacks and others still work) after 30 days.

For some context, Akismet has been filtering most of the comments. By the plugin’s own statistics, it is stopping roughly 2-3k spams per month. Starting from July 2025, I believe, there was the unusually higher traffic of spam. And these spams were more carefully designed than before.

Some of the spams read benign at first, as if the author genuinely had questions or wanted to add to the topic. Akismet flagged them for manual moderation. After I did some digging, I determined Akismet was right to flag these comments as potential threats. Author’s identity was inconsistent, as if AI filled them out with random information: suspicious service recommendation (likely the goal of a spam), email address not matching up, and the blank website submitted as their “blog”.

The others were more obvious; the kinds I would have deleted them as part of moderation effort. It was a typical foot-in-the-door tactic, trying to get a response (which would automatically approve the IP address). These kinds of comments often do not have a focus, likely because it is not written by human, and it shows. For instance, a human reader could rage over how a simple how-to piece did not work, and the comment would make sure to mention it. However, when likely an AI or a spam bot writes a similar snarky text, it’s completely oblivious of “why” it came to write it. If I were to hazard a guess, the behavior is likely a result of the fact that these are prompt-driven AI; it assumes “why” is already within the user.

The problem for me, however, is the number of these comments is simply increasing. Akismet, before July, rarely pushed a spam for manual moderation, perhaps one or two a month at most. Nowadays, I’m moderating (i.e. identifying if the author is a real human being) the “pending” box on daily basis. These spams almost never target newer publications. Using the rule of the thumbs as a guideline, I’m hoping 30 days would cut down the illegitimate traffic.

The other issue, aside from prevalence of more advanced spams, is that many of the pieces I write do have a limited lifespan: for how-to, there is a good chance the new update might have broken it, for app-to, there is a good chance the app might be broken, for op-ed, it may not be up to date with the latest developments. As much as I hate to say it, search function on the website exists for that very reason. Updating and maintaining an old article can be done only so much.