A while ago, while playing with Klout and the Klout API, I noticed how many Twitter users with very high Klout scores Tweet absolute rubbish, but quite a lot of it. It got me started on the topic of how exactly Twitter users accumulate followers.
Can you become powerful on Twitter merely by Tweeting like a lunatic? Or worse, can you automate Twitter follower accumulation? How much does the content of your Tweets actually matter if your Tweet frequency is high enough?
Conventional wisdom has it that you accumulate followers by Tweeting and reTweeting useful information, thus becoming a valuable resource on the Twitter network. Does this always hold, though?
To try and answer these questions (or rather, a small part of it), I created the following experiment:
I wrote four simple Twitter bots that Tweet the output of the fortune Unix command line application. The first bot Tweets every minute, the second every five minutes, the third every fifteen minutes, and the fourth every thirty minutes.
I made sure that these bots never follow any other Twitter users, they have no profile information, no avatars, and their Twitter.com profile pages are completely unstyled.
These bots have been running for just over a month now, and here are our findings:
In the above graph, each bot’s followers are plotted against the total amount of Tweets they have sent. As expected, it looks like the fifteen minute bot has a radically higher follower uptake than the two fast ones. Because of the potential higher nuisance factor of the faster ones, they tend to lose more of their new (and probably valuable) followers. (The fastest one has a nuclear follow cost of 39936.77 milliscobles! Robert beat that!)
Something’s up with the 30 minute bot though. It’s difficult to properly see what, because its graph spans such a small width. Let’s view the data as a function of time, so that all the graphs are the same width:
Now, let’s normalize the data, to make it fair. The 5 minute bot‘s data is multiplied by 5, the 15 minute bot‘s data multiplied by 15, and the 30 minute bot‘s multiplied by 30.
This is based on the assumption that, if the bots will continue tweeting until they have tweeted as many times as the 1 minute bot, their follower number will increase linearly. Not entirely accurate, but this assumption will have to suffice until I have more data (I will continue this experiment for a few months more, unless they’re disabled by Twitter before then.)
With the data at our disposal right now, it looks as if the 15 minute bot is doing comparatively better than the 30 minute one! We’ll need more data to say for sure, but this might be quite meaningful. It might mean that humans (Twitter humans, that is), is most comfortable with input that arrives every 15 minutes (or somewhere else in the 5 – 30 minute band).
Watch this space in another month’s time for more data.
Posted by Adriaan Pelzer