Is Twitter Useful?
Part 1 of 12
A month ago I started a public experiment to decide whether Twitter is useful.
Despite all the hype, Twitter has always been the worst performing channel across most of my startups (in close competition with Facebook). When other entrepreneurs ask for my advice, I tell them that they can safely ignore Twitter except as a channel for customer service.
Social Media advocates have countered that I’m “just not doing it right.” They are correct about this, so it’s only fair for me to put it to the test before I render summary judgement. I’m making an honest effort to figure out Twitter. After 12 months if I cannot iterate to success (including crowdsourced advice from the community), then I will have pretty good evidence that Twitter is useless. That is to say, if somebody were to start today, I hypothesize there is insufficient value to extract from the platform to be of any merit.
Iterating to Success
What do I mean by success? I picked two metrics out of the blue. If you suggest alternate metrics I would consider them.
- Double followers (from 1084)
- Double engagement (from 1.7%)
I’m also pledging to do so organically, to simulate the effect of somebody new wandering into this platform in 2020.
Of course, I could easily shell out $5 and get a thousand bots to follow me, or build a couple of fanbots to artificially double my engagement rate. But that seems very much outside the spirit of what I’m building.
I’m also trying not to leverage the other Twitter accounts I manage (which have upwards of 100K followers) to juice my stats. You see, back in the day it was relatively easy to start random Twitter accounts and get followers. Once upon a time there was a lot more oxygen in the room, not to mention Twitter was tacitly encouraging bot followers to juice their growth numbers en route to IPO. Now that they’ve been absurdly and baselessly accused of having anything more than marginal influence in real-world politics, they are clamping down on bot, making it difficult for new accounts to gain “traction” in the traditional manner.
In this way social networks have a similar economic structure to a pyramid scheme, where the people at the bottom get a ton of bricks to the face. Every social network has a natural half-life, which is why Facebook’s business model is the “hot potato” strategy of continually buying up promising new social networks faster than they can depreciate.
Similarly but perhaps more controversially, I’m also trying not to leverage my other non-Twitter assets. RezScore alone draws enough traffic that I could fire off an email to our millions of users and beg for a follow to my personal account. It would be poor business judgement and incredibly self-serving — but then again, going to extraordinary lengths to get attention is the American way:
Nonetheless, I’m a purist at heart, and want to see if it’s possible to hone my “raw Twitter skills”, whatever that means, and build a following purely by leveraging this platform internally. So is it possible in 2020, or is this bus is already full?
UPDATE: I already messed up the final point with a viral work article in my name, though nothing seems to be contaminated, check the addendum for details.
Month One Results
I had some hope that I could end this experiment after a month. After years of neglect, I thought that simply making a daily post would boost my engagement enough that I could claim victory (defeat?) and get off the platform. No luck. A month of posts and I got the double deuce — boosting my engagement 0.2% percentage points from 1.7% to 1.9% and gaining a whopping 2 followers.
The good news is that it does give some raw data to crunch. Despite my skepticism about this platform, I’ve pledged to make a good faith effort to solve the riddle with analytic rigor. With 57 tweets made in this time and some engagement data, this means I can run some basic analysis to see what works and what doesn’t work, to help me iterate into the next month.
Given the results thus far, it seems unlikely that we’ll be able to optimize towards the second goal (double followers) — I would need to double my following every month to be on track. So we’ll need to ditch this and focus on engagement rate.
For the 57 tweets I sent over the month, I tagged them into various categories:
- Tone: Was the Tweet attempting to be “funny”, “smart”, “bragging”, “mean”, or “asking for help”
- Subject: Was the Tweet about “Bitcoin”, “science”, “health”, “technology”, “work”, “politics”, or “Twitter”
- Format: Was the Tweet a “reply”, “subtweet”, or “tweetstorm”
- Content: Did the Tweet contain a “mention”, “external link”, “image”, “hashtag”, “emoji” or “carriage return”
- Numerical: Time and length of Tweet.
Top 5
Before things get more technical, let’s see what the top 5 look like (excluding replies) in descending order from best performing to worst performing:
They have at least one thing in common… they are all pretty much among what I personally considered my worst tweets. At best, only one of these would crack my personal favorite top five. Then again, you’re probably noticing by the tenor of this article that I’m fairly uncommon.
One thing that pops out… 80% of them actually have a link to interact with. It’s sort of a “well, duh” moment that posts with links will have better engagement, since they have “something to engage with” built into them, so expect to see more links.
The first one may have tricked people into thinking there was a hot girl if you clicked, so I’m going to experiment over the upcoming month with a lot more hot girls. There was also a sense of drama and story that people crave, in that her life was in danger. Maybe I should overhaul Twitter account to do nothing but threaten the lives of supermodels.
Well, she may have never existed in the traditional sense, but she nonetheless she lived a rich life in the imagination of both man and machine:
It’s not just me that noticed the “links” is a common feature. R also had something to say about it when I throw these factors into a regression to see what pops out as significant:
If you can interpret this you already know what’s coming next, but this is like a TL/DR up front — links and images work, tech topics do too, and multiline tweets apparently do worse.
Format
Let’s start by looking at if the tweet was a “reply”, “subtweet” or “tweetstorm”.
This is a good intro to these boxplots — I’ll be doing a lot of these because it’s quick and easy to generate in R and I don’t want to waste too much time on this article. It’s not the best way to visualize the data —but it has the nice property of being a fairly neutral way to present the data for anybody to draw their own interpretations. Of course, please contact me if you are interested in the raw dataset.
We see pretty quickly that replies look pretty good — it’s our best weapon to boost reply rate over 4%. HT to Charles Jo for recommending this strategy, the only social advice expert advice I followed that showed positive results.
Subtweets look to be major losers, which is sad because it’s my favorite format. Tweetstorms look to have a higher floor but lower ceiling.
Tone
I’ve tried different tones, based on recommendations from social media experts, but I don’t see a lot of difference.
The best performing tone looks like “asking for help.” Indeed, something I’m observing during this experiment is whether Twitter has value — so if the community is good at “crowdsourcing” help I should think this would be value. This is baked into the premise of my Twitter experiment — if Twitter is indeed good at crowdsourcing help, then I should be able to get the community to solve my experiment.
There’s also the “modern-day presidential” qualities of braggadocious behavior and attack tweets. Sometimes among fellow lefties, they will lament that although the person holding the office is contemptible, they identify these behaviors as effective and wish a leftist firebrand could utilize the platform so effectively.
You all seem to respond well to the self-aggrandizement, which is good news for you because I have not yet scratched the surface of how great I am. I’ve not yet found the right art to the “meanness.” The data suggest I should cut back, but that will be tough for me because I’m so often been described as the “world’s biggest A-hole” (← “see, there he goes bragging again!”)
Whether the style was intellectual or jocular didn’t seem to matter… or maybe I just missed the mark on this content by a country mile.
Content
OK, here’s some of the factors that might make a difference:
Links really do matter — if I get this right, I can get almost all the way to the goal just by putting interesting link in my content. Same with images to a lesser degree. Mentions might be a bit of a factor, but tough to tell. Hashtags actually look like they hurt things somewhat. #counterintuitive
It’s a bit surprising that mutliline tweets do poorly. I had assumed that they would increase engagement, since the full content would sometimes fall “below the fold” and require a click to see the full content. Perhaps my teasers are insufficient.
Subject
Wow does it look like tech tweets are a winner. But, it’s an awfully small sample size. It was my least common subject, so I’m taking it with a grain of salt.
Sad that you all don’t respond to the Bitcoin tweets — if you ignore everything else I say, you should at the very least buy bitcoin.
I had to test politics tweets, because people have said politics Twitter is so dominant nowadays. Politics twitter is insufferable though, so I’m glad that’s not hitting so far.
Science and health stuff seem to do ok. Also relieved that the work stuff isn’t turning anybody off, because there is so much exciting stuff going on at RezScore.
After I generated these graphics I also took a look at “Twitter bashing”, which anecdotally seemed to be popular, but the evidence didn’t look as strong as I had suspected. I do still suspect that the people who choose to engage with the platform are defensive of it, so I may tweak this and try more.
Time
Best time of day? I’m not allowed to use Twitter while I’m at work, but it looks like you kids are up late — 7 PM PT looks like the optimum best time. The party may go later than that for all I know, but I have a tendency to kiss you the Irish goodbye early and leave the party.
Length
How about tweet length? Looks like the less I say the better. Better to be thought a fool than open your mouth and confirm it…
If you like my shorter tweets better, you’re in luck… my hypothesis gives me a nice offramp from Twitter, so we’ll see if there’s an asymptotic discontinuity as my length approaches zero.
Coming Up Next Month:
Now that we have some of the basics covered, here’s some of the theories I may be delving into next month… you can get an early read here:
I’ll also present the notes of a chat I had with a Twitter expert with over 20x my followers, who confirmed my belief that the platform is “essentially worthless”
UPDATE: The “Hacker News” effect
This article was slightly delayed in its release due to my startup unexpectedly squatting on the frontpage of “Hacker News,” getting an unusually warm reception from a typically testy crowd.
While this is great news for my startup, this means I violated my above pledge not to leverage my outside assets before I could even publish this article! On advice of our team I published the article under my personal name instaead of the company name, which in theory means the attention (7000 reads and spreading as of publication) should leak to my personal account.
So what is the effect of getting major exposure on the front page of a popular tech site? Exactly one new follower. The “Twitter-verse” really is a bubble.
If I were to hack this strategy to resolve this experiment in the affirmative, I would need to generate innovations at a pace that would make Thomas Edison blush, occupying three slots at a time on the frontpage of the site for the rest of the year.
