Lately, Joe Pulizzi has been posting videos on Twitter:
In a recent episode of the This Old Marketing podcast, Joe mentioned that he’s trying to crack the code on what makes a video take off vs. get ignored. I’m not sure if this is the sort of answer Joe (or anyone else) is looking for, but for me it comes down to one word:
Of course, videos and related content need to be humorous, captivating or otherwise interesting in order to become popular. But I believe that random factors are a larger force.
Here’s my speculation on how the Twitter algorithm works.
(Side note: I believe Twitter uses AI and machine learning, so it doesn’t behave like a conventional algorithm. The system learns and adapts based on a changing data set. So it’s operation is opaque and difficult to document, but I attempt to explain the current manifestation of its AI.)
Testing the inner circle
For each user, Twitter takes the entire list of followers and creates a small subset that I’ll call the inner circle.
These are folks who consistently engage with your tweets. I see that a set of 10 people consistently engage with many of my tweets. And for me, it’s the people I often engage with who get pushed to the top of my feed, a number that’s also about 10.
Twitter tests your tweet by presenting it first to your inner circle.
Serendipity, part 1: Is your inner circle on Twitter right now?
Serendipity, part 2: If they are on Twitter, are they actively checking their feed or are they responding to Notifications or participating in a Twitter chat? Maybe they’ve been watching a live video broadcast for the past 30 minutes.
The popularity of your tweet depends in part on how much your inner circle engages with it, during the first 30–90 minutes of its existence.
If your inner circle is not on Twitter, but logs in later, they’ll be presented with your (aged) tweet and there’s a chance momentum will build if they (and others) interact with it.
The ideal scenario, though, is for your inner circle to have a burst of engagement shortly after you tweet it (e.g., within the first 15 minutes).
Testing a random sample
Twitter expands its test and exposes your tweet to a wider set of your followers. If the algorithm could speak, it’s saying, “OK, let’s put a whole bunch more out there. What’s happening, what’s happening? Are people buying in?”
Let’s say there’s limited engagement from your inner circle and zero engagement from the random sample. I think Twitter stops showing the tweet to the rest of your followers.
Of course, there’s things like hashtags and tagging other users. Those are wildcards that may get your tweet noticed and amplified. There’s also Twitter search, where people may find your tweet based on the words you used.
Serendipity, part 3: Are your other followers online right now and are they liking, retweeting or commenting on your tweet?
Serendipity, part 4: Do people come across your tweet by way of search, hashtag, etc.?
The snowball effect
So here’s how things can take off. One or more of the serendipities hits and your tweet gets noticed. It’s being liked, retweeted and commented on.
I don’t recall when, but Twitter made a newsfeed change that now shows you tweets from people you don’t follow. So long as a friend (e.g., someone you follow) liked it, the tweet may appear in your feed. Here’s an example — I don’t follow Melanie, but Michelle does:
At birth, your tweet is a snowflake.
Based on some serendipity, the snowflake turns into a snowball. Then, it starts rolling down the hill, getting bigger and picking up steam as it rolls along. This downhill motion is the likes, retweets and comments, which gets your tweet into more people’s feeds — even people who don’t follow you.
ADDENDUM (December 11, 2019)
Recently, I tweeted that I established an LLC for my consulting business:
A bunch of replies came in from folks congratulating me (thanks for that!). The replies continue to stream in — every few hours, I get another one. I predict that this will continue for a few more days.
What’s happening here?
Playing amateur psychiatrist, I’ll call this the “join in with the merry crowd” phenomenon. People who come across my tweet see the high engagement counts. They see that lots of other people replied to my tweet. As a result, it’s an easy decision to join in and write a reply.
If there were ZERO replies, however, a user may not deem the tweet to be worthy. The thinking might be, “Eh, that’s interesting, but I wonder why no one has replied yet?” You don’t want to be the first to reply, so you don’t.
This tweet has reached medium snowball size, though. The Twitter algorithm is already pushing it further down the hill. This means that more people will come across it for the first time. And more replies will come in.
Here’s another tweet that became popular:
I believe that a few acts of serendipity allowed it to catch on. From there, it got the benefit of the snowball effect.
If the forces of serendipity hadn’t intervened, this same tweet could have received minimal attention and engagement. If I tweet the same exact thing 20 times over the next 6 months, it may never see anywhere close to the engagement the original got.