Saturday, October 18, 2014

Early Analysis of Insult-O-Bot 2000



Insult-O-Bot is a twitter bot I designed primarily to help teach myself Python and the twitter API, but also because I find something inherently funny about using advanced technology for potty humor.

You can check up on him via his twitter page or you can tweet him directly: @InsultOBot.

He has been live for 10 days now, and here is an analysis of his performance so far:


  • 158 followers
    • 81% male
    • Interests: 86% Comedy (movies and television)
    • Interests: 61% Comedy (hobbies and interests)
    • 61% from the USA
    • 6% from Canada
    • 19% from the UK
    • 28% also follow Conan O'Brien
  • 343 tweets
    • 63 retweets (18%)
    • 178 favorites (52%)
    • 2048 tweet engagements (6 clicks per tweet)
    • 29,249 total impressions
    • 7.0% click/view ratio

Most of these stats are fairly obvious.  A potty-humor-robot appeals primarily to men who enjoy comedy.  One interesting point, however, is that his engagement rate is incredibly high for a robot, which I believe comes from the underlying rule-set he is governed by.  Since people "nominate" their friends to be insulted, on top of the entertainment and the content Insult-O-Bot provides, there is also a perception that Insult-O-Bot's tweets actually "came from" a good friend: the person who nominated them.  This has a lot of interesting implications for twitter bots designed for marketing purposes, where engagement with a brand is the primary goal.

Here is a breakdown of the engagement rate by subject, and by the type of insult:

The primary lesson to be learned here is that if you want to insult someone, make it about how much of a loser they are, and not about how stupid their mother is.

I will go through and tweak the variables to improve the engagement rates, but there is another metric I would prefer to optimize on:  (retweets+favorites)/Impressions.  That metric is a more genuine representation of enjoyment, but unfortunately, the counts are too small to run a proper analysis on yet. I will have to wait until Insult-O-Bot has a few thousand tweets first.

As for bugs and issues, I had a few bugs I needed to iron out in the first couple days since he went live (including trouble recognizing nominations), but all of that has been taken care of now, and I can be relatively hands-off, with the exception of some moderation for the insults submitted by followers.

Here are a few other interesting observations:
  • People seem to genuinely enjoy insulting insult-o-bot, even via direct messages, where nobody else can see the conversation.
  • Nominations get a lot of favorites and replies
  • Even though insults about "your mom" had lower engagement rates than insults about you, a preliminary calculation of retweets+favorites per impression for "your mom" insults was 25% higher.  More data is needed to properly evaluate the impact of "your mom" on insult enjoyment, but it should also be noted that she is so fat, even her shadow has a significant impact.