How many of Trump's followers are fake?
The President of the United States of America likes to boast about how many Twitter followers he has. But who are they?
I was curious, so on Thursday 31 August 2017 I used the Twitter API to download the unique account IDs of one million of Donald Trump’s most recent followers.
After waiting a day or so to give any new users time to set up their profiles and send their first tweet, I downloaded the metadata (username, number of tweets, profile picture, etc.) associated with each of these IDs.
Here’s what I learned.
Of the 1,000,000 accounts in my sample:
- Around 700,000 followed Trump within a few hours of being created. The chart below shows a plot of 1/1000th of the dataset. You can clearly see that most of the accounts fall on the lower frontier, i.e. are less than 12 days old (note the log scale).
Around 490,000 have no basic profile information. These accounts have no profile picture added, no description, location or URL entered, and have not changed their default theme / background.
Around 390,000 are silent. These accounts have never tweeted, nor have they ever retweeted or liked anyone else’s tweets either.
The fact that most accounts are new isn’t massively surprising in and of itself; Twitter suggests people to follow as part of the signup process, and the President of the United States is a pretty obvious candidate for inclusion. Interestingly, however, even for the oldest cohort of this type of account in my sample, most still didn’t add any profile information or send any tweets even after 11-12 days of being on Twitter.
Based on these figures, my guess is that only 51% of recent Trump followers are real.
Although this number is pretty rough and ready, it’s consistent with other estimates from services like TwitterAudit. In reality there are real accounts with no profile information, and fake accounts that are superficially convincing - and armed with more time and firepower there are increasingly sophisticated ways to discriminate between them. But for a quick analysis, using the absence of basic profile information as a predictor is good enough.
Other fun facts:
My sample contained around 140,000 accounts with usernames ending in 8 digits. This is the default naming convention that Twitter uses during signup if you skip the step to choose your own username.
My sample also contained around 7,000 accounts that follow Trump and no one else (which is a pretty offensive waste of a Twitter account IMHO).
And finally, around 500 accounts were deleted during the time I waited between downloading the original list of IDs and going back to retrieve the associated metadata.
In any event, the bottom line is that it looks like Trump is currently gaining around 90,000 followers per day. At this rate, he’s set to gain more than 2.5 million followers per month. And half of them are probably fake.
If you want to replicate this analysis, either for Trump or for anyone else on Twitter, then I recommend hitting the following API endpoints:
The first of these will return the unique account IDs belonging to your target account’s followers, starting with the most recent (these are numeric identifiers, not the @usernames you see on the front end). The second will convert blocks of IDs into fully-hydrated user objects, including @usernames and other profile info.
The API is rate limited, so you’ll also need to be patient if you want to pull a large sample to work with.
GET followers/ids returns up to 5,000 IDs per call, and allows 15 calls per 15 minute window, so retrieving 1,000,000 IDs takes around three-and-a-half hours.
GET users/lookup operates on blocks of up to 100 IDs, and allows 300 calls per 15 minute window, so converting 1,000,000 IDs takes around eight-and-a-half hours.