Robots Are Doing Better Than Humans on Twitter

This is all about an experiment I started back in July 2009, called TwiBots. Initially it was supposed to be a simple 24/7 online tweep (Michael Davis) saying a bunch of stuff randomly. But then we (Michael and I hehe) started picking out certain topics, feeding content from certain RSS feeds, filtering all content by keywords and regular expressions.

When the new Retweet API came along, I wrote an RT module which uses the Twitter Search API to find relevant tweets and users, then retweets those messages or just somehow interacts with a user. As soon as the Twitter Lists API was announced, I started working on the Lists module, which eventually became a simple “sorting-the-tweeps” based on their keywords – web design, design, wordpress, etc. Soon enough, I found out that Lists couldn’t contain more than 500 members (although some glitch made 501 possible). Web design 2, wordpress 2, etc wouldn’t be as fancy. I also tried building a conversation list of tweeps that by any means talked to the robot, but then again, the 500 limit broke all my hopes ;)

There were other modules which I worked on really hard, such as RThx module or Random Buzz, DM Control. Some of them worked, some of them were turned off after a few days (yes, you guessed it – Random Buzz, that really made some noize ;)

So, what did I achieve? Me – nothing. Michael did though, in 6 months he went up to 4500+ followers, while following a little more than 200 people himself, has been featured in ~ 250 lists, sent ~ 55,000 tweets and retweets all based on four keywords (or hashtags) – design, web design, wordpress and jquery. Built 4 lists based on these keywords, 500 members in each. Total list followers is a little less than 150 (which is quite good actually).

For comparison take a look at my account – (@kovshenin), in a little bit more than a year I got ~ 1700 followers. It took Michael a couple of months to reach that. The chart below illustrates the followers growth during the last three months. Human (me – blue) vs Robot (Michael – red). Yeah, I added a new module in mid December ;)

Twibots Chart: Robot vs. Human

I manually logged into Michael’s account recently to check out how he’s doing, and I was kind of surprised to see that people really are talking to the guy, thanking him for retweets, asking him for further reads, wishing him a great day and handing over some coffee. Michael doesn’t usually reply to these and he’s a little bit shy sometimes, besides, he never drinks coffee ;)

A few days ago I decided to give Michael a rest, so tuned his backend to a new Twitter account with a few different settings, especially in timing. I’ll be switching to other keywords and feeds in the next few weeks. So let’s see if he’s as good as Michael, or perhaps better? ;)

I’m not sure what I’ll do with Michael’s account. TweetValue said it’s worth over $5k … anyone? ;) Or should I just throw it away.. Or run a contest for his password? …

About the author

Konstantin Kovshenin

WordPress Core Contributor, ex-Automattician, public speaker and consultant, enjoying life in Moscow. I blog about tech, WordPress and DevOps.

17 comments

  • I always suspected that I'm talking to robots on Twitter rather than to humans :))

    Talking seriously, there are many users on Twitter who are infected with follower-mania. The larger counter value for “# of followers” they have the more they happy. At some moment of such growth many start using different 3rd-party tools to facilitate automatic account growth. This leads to a large community of those who has accounts charged with tons of friends that aren't interesting to each other at all.

    And I think thanks for retweets and wishing good luck is a reflex. Nothing more.

    Do you have a statistics of how many of 4K followers click a link that you tweet out? That would be interesting. I thing the percent for such account will be low. Hope I am mistaken :)

    Great post!

    • Victor,

      Awesome idea, I'll shoot out a link with some tracking code to find out the click-through rate but hey, I understand your larger # of followers theory, but it's not like we went followers crazy with Michael.

      I personally didn't look into the stats that much and I never really checked who is following the robot, but I've seen other humans (I hope) retweet the guy, so I need at least one click through ;) As long as I make him tweet something interesting.

      You see, I'm pretty sure that at least 50% of the followers are using/used/interested in #wordpress #webdesign #design and #jquery, otherwise they wouldn't have followed. Right? ^_^

      Cheers, and thanks for the comment! ;)

    • Okay so according to bit.ly I got 70 clicks in 4 hours on four different tweets scattered through a 15 minute interval. So what is this supposed to mean? I have no idea. It could have been 10 people who clicked 7 times each, it could have been 1 person who clicked 70 times (I doubt this one) or it could have been 70 people who clicked once.

      Maybe I should use something else rather than bit.ly ;)

      Then again, this doesn't show much. It still depends on the time zone, tweet intervals, tweet topics, etc. So I guess this should be measured a little bit differently. Perhaps I should resume the robot's activity, then track the bit.ly links generated via his API key pair, and sum up the total on a 24 or 48 hour basis.

      Thoughts?

    • This means that you're quite influential. I'm surprised a bit with results, they're pretty good. I compared your account to Dan's in comment bellow.

      [Thinking about creating robo-account….]

  • Great experiment.

    A 4th option on the bit.ly is that it could have been 2 people and 68 robots as it appears bit.ly doesn't scrub robot clicks when you line up Google Analytics and Bit.ly traffic.

    • Correct, though I don't really understand what you mean in terms of robots. Google doesn't crawl through Twitter links because they're marked as 'nofollow', and even if it would, probably not that fast.

      Maybe posting a bit.ly link with a really really old article or something obvious that humans wouldn't click and track the robots rate? Or are robots smart enough to distinguish interesting and uninteresting? ;)

      Thanks for the comment!

    • Yes, it auto-follows back anyone who follows it (and similarly unfollows anyone who unfollows it), but it never pro-actively follows anyone (i.e. it's always just waiting around to see who it picks up 'organically').

      I do auto-following for all my bot-based accounts (@twitexperiment, @freelondon, @freenewyork, @reboundfinder) and even some non-bot accounts (@boxuk, @thejanuarist, etc), just because… I don't know, I just think it's polite! But for my personal account (@zambonini), I do manually check everyone before following back (I have certain criteria, but follow most real 'people' who use twitter as a conversation tool, but not really broadcast-only accounts).

    • I should point out, for the auto-following (or, rather, the auto follow-back and auto unfollow-back), I built my own simple PHP script, rather than using one of those 3rd party services, as I don't really trust them not to spam DMs.

    • Sweet list you got there and I liked your recent blog post ;) I stopped relying on 3rd party services as soon as I noticed on of them shooting a status update without asking me about it first.

      I'm working on a little platform right now called Twibots. I'm still unsure what it will be when finished, but the drafts are working ;)

    • Aaaaaa! I said this already, but I repeat: I always suspected that I’m talking to robots on Twitter rather than to humans.

      Now I'm not suspect, I'm sure! @twitexperiment twitted few weeks ago about my project, and I was happy. I followed him, thank him and he followed me back :) Till today I was sure that @twitexperiment is a man. How stupid I am, doh! :)

      Btw looks like an account of the author works more effectively since he got 70 clicks for 4 links. And I got 6 clicks once you twitted about me. You have had 4500+ followers at that moment.

    • Victor, you cannot judge the click-through-rate upon one link tweeted once. It seems that this is pretty difficult to measure, because it really depends on how interesting the tweet is, how many followers are actually looking through the tweets at the time it has been tweeted, and how many people were searching for similar content at the time the tweet has been fired.

      I'm working on the Twibots Platform right now and I've got a the bit.ly shortener as an addon that could be applied as a filter to the links at the RSS module (pretty complicated, huh?). I'm working on saving the links tweeted together with their generation time, so that I could measure at least an average possible click-through-rate.

      This doesn't save me from duplicate clicks and I had the idea to rank bots in the Twibots Directory according to their click-through-rate, but I'm afraid that people would just start clicking multiple times on their bit.ly links to rank higher, meh!

      Perhaps I should write a new URL shortener for the project ;)

      Regarding to robots, yeah, with a few regular expressions I was able to read through my @mentions and retweets, then gather them guys in groups of 5-6 (depending on their screen name lengths) and shoot a random "thanks for the RT", "appreciate the RT", "thank you", etc mentioning the guys.

      I also stored them into the database for later use. Each other hour on every Friday I groupped them guys again based on the most used hashtag they used when retweeting my tweets and shooted out a #followfriday these guys for awesome #hashtag content.

      This really made Michael look a little bit human y'know ;) Confusing hehe..

    • > Victor, you cannot judge the click-through-rate upon one link tweeted once

      Yep, agree. Highly depends on interest, but IMO does not depend on time zone too much when you have thousands of followers uniformly distributed around the globe (is it your case?). If there is a shift, TZ is important also of course.

    • Maybe, maybe not.

      First of all we cannot assume that Twitter is used all around the globe. Twitter in Russia for instance is not too popular.

      Based on a few of the latest people who followed Michael are scattered around USA, Western Europe and a couple of them from China – ( according to Foller.me: http://twitpic.com/15mukw )

      This means that if I tweet at 10 AM in America, then I'll get clicks from the US but Europe is still asleep. When I tweet at 10 AM in Europe I loose all the US clicks, and if I tweet at 10 AM in China, I loose both US and EU clicks.

      I hope that makes sense ;)

      Cheers.

  • […] Who wants to follow a long-ass stream of nothing but links? Do you?? And all of a sudden I’m asking myself…to what end? Sure I like Twitter okay, but I don’t looooove it the way I love knitting, web design and painting. I love people, but most of the time I feel like Twitter is just filled with robots endlessly sending their links out hoping for clicks. Sometimes I worry that to the other twitter users, I’m just another four Twitter accounts doing the same thing. Indeed a quick googling found this post that proposes that robots do better on Twitter than humans! […]