Hook, bait and camouflage: making a honeybot

For the past few months I’ve been working on a bunch of what I call “honeybots” — bots that act as honeypots for jerks who term-search on Twitter. It’s worked really well, and it’s kind of cathartic to watch the neo-Nazis and movement atheists yell at them. But with Twitter still refusing to ban most harassers, it’s gonna take a lot more honeybots than the ones I make to really make a dent in the problem. My formula’s already been improved on by arguetron, which any aspiring honeybot maker should look at for inspiration, but I want to do some overview stuff on how these work so we can get a bunch more people making them.

First of all, my bots use Tracery and Cheap Bots Done Quick; I’ve also been setting some of them up to auto-retweet using labnol’s scripts and go through lists of generic, non-bait tweets that I put in Zach Whalen’s Google Spreadsheets-to-Twitterbot tool.

You can use any combination of these and/or other tools — a lot of Twitter bots use Markov generators, and if you’re generating text based on a corpus that uses “bait” words a lot, you can make them with those too. I’m a really bad programmer and tend to use pretty simple tools for my bots, but if you have computer science and/or linguistics training, you can make honeybots out of anything that posts to Twitter.

For my purposes, the honeybots have three parts: bait, hook and camouflage.

Bait is stuff that people term-search Twitter for looking to pick fights; people who engage with strangers as performance will often share bait to get other people to join them. When they do this with regular Twitter users, it’s kind of a jerk move, and often it’s harassment; the sharing with their followers often incites a mob. When they do it to a bot, though, it’s just them shouting into a void — they don’t harm real people, and sometimes they even are fun to laugh at.

Honeybots pretty much only work on people who are term-searching — they aren’t usually interesting enough for people to want to follow them on their own. I say “pretty much” because you can occasionally hand-bait trolls tagging a honeybot into a conversation with them, but this is a pretty hit-or-miss tactic.

You can use a ton of things for bait. A lot of alt-right/neo-Nazis working for places like Infowars and Breitbart search for their own names and retweet critics to get their followers to harass them. A lot of right-wing politicians have fans who name search them looking for people to harass. Some conspiracy theories work well too — I get a couple “flat earth” people arguing with my bots every month. Evangelical atheists also term search “atheism” looking for people to convert to atheism and/or to performatively argue with.

This is just the stuff that occurred to me as stuff that trolls search for; you may know of more topics.

Second, we have the hook. This is what the bot replies with. You get as advanced as you want with this; my bots have 20 or less replies that they choose from at random. You can put in hundreds of replies if you want. You can have it generate Markov text. You can plug in ELIZA. The #1 thing I’ve found to keep people engaged is to have it occasionally make spelling/grammar mistakes; the kind of troll that uses Twitter like this generally is incapable of not correcting them. I also have been experimenting with my bots occasionally grabbing phrases from the list of “bait” phrases as replies.

Third, we have the camouflage. Since most people will be going directly from search or a mob harasser’s retweets to your bot, this is technically optional. But this is the part that makes people think your bot is human. I’ve been experimenting with a bunch of boring life-related tweets that they cycle through and having them retweet news and product accounts.

That’s all you need! Make one. Make a dozen. You can straight-up clone them; I sometimes give them “themes” in their graphics and retweets — one of mine retweets a bunch of game companies and accounts for camouflage but says the exact same things for bait as several others.

The more of these there are, the more these people will have their search results cluttered with bots and the more they’ll be wasting their time abusing bots instead of people.

A side note — these are an experiment. They don’t substitute for the things that need to happen to make society better — if you only have the time/energy to make some honeybots or call your congresscritters, make the calls instead. They also are a band-aide on the problem of web harassment; they only occupy one kind of troll on one website.

Here’s a short FAQ for common questions, though.

How do you get around Twitter’s duplicate filter?

You can randomly choose between different unicode spaces, or even have it stick a random number of spaces at the end of posts. Or you can do what I do, which is have them stick randomly-chosen emoji on the end.

How do you monitor replies?

I do a Twitter search for “to:[HONEYBOTNAME]” in incognito mode (so I can see users I have blocked).

What about shadowbans?

Twitter occasionally shadowbans accounts — I think it’s a spam-prevention thing— which keeps them from showing up in searches. If this has happened to your bot, it’s pretty much shot, but these are so trivial to set up that you can just make a new one. Or a dozen.

Why do you give them names?

Most of these trolls are misogynists, so giving the bots “girl” names makes them a bit more successful. It’s also funny watching them try their usual patronizing bullshit on robots.

How do you choose icons/graphics?

I use public domain photos. I’ve mostly been trying to, when I pick a “human” avatar for them, make sure that I make one that the person isn’t identifiable in, because I don’t want to accidentally make some stock photo model a target.

Aren’t you worried about the “other side” doing this too?

Neo-nazis/the alt-right already are doing similar things, except they’re funded by billionaires. The main reason I’m not worried about this, though, is it only traps people who are using abusive/harassing tactics; that’s not something the “good guys” really do. Sure, there are some left-wing folks who term search — if you say “irony bros” a bunch of them will show up to yell at you — but they aren’t actually doing anything useful anyway, so if people waste their time, it’s no skin off my back.

We’ve been telling Twitter for ages that their platform is easy for trolls to exploit for abuse and they’ve chosen to do nothing; this is a way to automate the counter-trolling so that human beings can spend their time getting the real work done.