[extrapolate]

Quote from a previous article (so I wouldn't attribute all 8% to bots):

  We’re experimentally removing CAPTCHA and using other methods for spambot detection.
  We know it’s annoying for humans, and are starting to worry the bots are actually enjoying it.
  Plus, our visual CAPTCHA was failing on accessibility for non-visual redditors.

[AdamGibbins]

Sorry, to be clear, I wasn't attributing the lack of CAPTCHA to bots. Rather the improved usability and increased human sign-up rate to the removal of a painful CAPTCHA.

[jbeja]

What other method for spambots could be used?

[bobbles]

In addition to plenty of stuff I cant think of, you could see:

1) High percentage posts flagged as 'reported'

2) High percentage posts flagged as 'spam'

3) High volume of posts containing URLs

4) Posts containing identified spam messages (like how an email spam filter may work)

5) Multiple accounts from similar/same IPs

Almost all spam in a subreddit I mod for is from accounts < week old. (I believe there may already be a limit on how much you can post after registration though)

[ubernostrum]

I mod a subreddit with >100k subscribers. We use AutoModerator and have it set to remove anything posted by an account less than a day old.

It catches some false positives from people who sign up just to post something, but it's also drastically decreased the spam. Most spambots are still at the "sign up and then blast shotgun-loads of posts before the algorithms catch you" level of behavior.

[tokenizerrr]

Do you review the removals and undo those for legitimate users? If not, wow. Talk about being hostile towards new users.

[ubernostrum]

Most people never notice. The ones who do get an explanation of what happened when they ask about it in modmail.

It can be rough on a brand-new account that wanted to jump into a discussion right away, but at the same time the quantity of spam hitting your subreddit once you pass the 100k mark is just ludicrous, and this is by far the most effective way to stop it.

[jobposter1234]

In terms of community building, that might not be the worst thing in the world, especially if the community objective is quality over quantity.

[extrapolate]

IIRC they probably mean to detect them through how they operate on the website, rather than how they register (tracking navigation, how they upvote/downvote, and what content they post).

[eli]

Simply testing for javascript still works pretty well for drive-by bots. For ones targeting Reddit specifically the captcha probably wasn't doing much anyway.

[UrMomReadsHN]

Adding a simple honeypot would probably cut spam in half.