There are a lot of bots on Twitter. Some are trying to sell things, some are stage one in an elaborate scam, and some are run by international intelligence agencies for any number of reasons.
Spotting these bots isn’t necessarily hard: just scroll through the timeline and see whether their activity resembles that of a human. Do they converse with friends, as humans do, or do they just say things to users who never talk back? Do they have a diverse range of interests, as humans do, or do they stick to one topic? Keep these things in mind and you can get an idea of whether something is a bot.
For those times, however, that you just can’t tell whether you’re looking at a bot or a person, Botometer can help. This tool, from Indiana University and Northeastern University, considers over 1000 factors, and then gives you a probability that a given Twitter user is or isn’t a bot. It isn’t perfect, because this is a hard problem to solve, but Botometer is a great tool to have around.
To get started, sign in to Botometer with your Twitter account, and then start adding any username you’re curious about. You’ll see the result quickly:
What does this mean? The higher the percentage on the “Bot Score,” the more likely a given user is a bot. According to the Botometer FAQ page:
Roughly speaking, one can interpret a bot score as a likelihood that the user is a bot. As such, bot scores closer to the extreme values of 0% and 100% are more confident assertions of the account’s bot-ness.
In this case, Botometer thinks there’s only a 16 percent chance my co-worker Harry is a bot. It’s a reasonable conclusion. I’ve worked with Harry for years, and still occasionally suspect he isn’t real—but only like 16 percent of the time.
There are a few things we can dig into using the “Details” link at right of the results. For example, we can see a timeline of when the user was last mentioned and retweeted.
You can also see a breakdown of the kinds of sentiments the user defaults to, and a breakdown of word usage (noun/verb/adjective/etc.) These are just a few factors used by the service, but diving into them can prove fascinating.
I ran this by a few known bots, and a few people I’m reasonably certain are humans. Precise percentages varied, but for the most part I found the results reliable. The main exceptions tend to be Twitter accounts run by multiple people, including those of politicians and brands. This makes sense to me, because such accounts frequently behave in bot-like ways—they tend to be focused on single topics and often don’t engage in conversations the way normal users do.