Twitter's user growth problem is worse than you think

Twitter said it only added 4 million users between Q3 2014 and Q4 2014, meaning the growth of its user base appears to be grinding to a halt. CEO Dick Costolo said Apple's iPhone operating system update, from iOS 7 to iOS 8, cost the company 4 million users it would otherwise have added, because the new iPhone system didn't automatically display tweets to users in the same way the old one did.

But details inside Twitter's earnings disclosure show the underlying problem may be worse than that. The percentage of users on the system who aren't actually doing anything stayed the same despite the loss of the Apple users, Twitter said. The company did not say why that portion of users who are on the platform "without any discernible additional user-initiated action" was unchanged.

Those users "without any discernible additional user-initiated action" could include apps like Blinkfeed, which comes installed in HTC phones, and pulls social content from a variety of sources including Twitter to create a feed of updates on a user's smartphone homepage. The user has to first authenticate the app, but from there it just automatically pings Twitter to obtain tweets, without the user needing to do anything.

A Twitter spokesperson tells us these are real users who created Twitter accounts with real usernames and passwords, receiving Twitter content through third-party applications. They're not bots or machines, they're real people who want Twitter content, the spokesperson says.

Users who aren't actually doing anything

According to Twitter's press release, the number of people it counted as "monthly active users" (MAUs) rose by 4 million people to 288 million. But in that number there are a whole bunch of "users" who aren't actually doing anything. Twitter's statement describes them like this (emphasis added):

In the three months ended December 31, 2014, approximately 8.5% of users used third party applications that may have automatically contacted our servers for regular updates without any discernible additional user - initiated action. As such, the calculations of MAUs presented in our earnings materials may be affected as a result of automated activity.

If you minus out that 8.5%, then the monthly active users on Twitter who do actually perform actions totals only 263.52 million users. That's an increase of only 3.66 million users from the same number in the prior period - slightly less than the 4 million increase that is Twitter's headline number.

Here is what that looks like in a chart:

Here is the population of Twitter users who "used third party applications that may have automatically contacted our servers for regular updates without any discernible additional user-initiated action":

Nearly 25 million Twitter users are not doing anything on Twitter, except maybe passively reading tweets, according to the company's own numbers. (That big decline in non-active users in Q2 is explained here.)

"We don't expect to get those users back ..."

That percentage of non-active "active" users was the same as the prior quarter, 8.5%. That is a surprise because the loss of the iOS 8 users would seem to imply that the percentage of automated user activity should have gone down, because a portion of people passively receiving automatic updates from Twitter were lost in the Apple update. Here is how Costolo explained the iOS 8 issue to Business Insider yesterday:

There were two issues. One was Safari auto-polling, and that was 3 million users and we don't expect to get those users back. The other issue that was more complex was an encryption issue related to the Twitter integration into iOS, such that when users integrated, a lot of them weren't able to launch Twitter successfully. That was a much more complex issue, it did not have a one-size-fits-all fix, so the team here worked as quickly as possible to address it but it caused a large number of users to not be able to use the product, even those who were trying repeatedly to figure out ways to get in.

In iOS 7, if you were logged into Twitter, the Shared Links section on your phone automatically showed you recent tweets from people on your connected Twitter accounts. In iOS 8, that changed. Note also that Costolo describes the issue as "auto-polling" from Safari, Apple's web browser. That seems to be something to do with users who had Twitter bookmarked on their Safari browsers. The bookmark would ping Twitter to keep itself updated even if users weren't looking at it.

The fact that the percentage of automated users stayed the same despite Twitter losing users from the iOS ecosystem implies that elsewhere on the Twitter platform automated activity increased.

We cannot say that for sure, however. We're only looking at the numbers Twitter publishes. Business Insider asked Twitter for comment but we have yet to hear back.

In one way, an increase in automated users is actually good news for Twitter - it means people want to use apps or devices that receive tweets. The service is useful to them. But their level of engagement seems passive. It's not clear whether those users are actually reading the tweets or whether their apps and devices are just pinging the system for updates even though the users aren't seeing them.

Twitter stopped counting "timeline views"

The company also stopped counting another user metric, "timeline views." The metric was thought to be useful because the more users look at Twitter, the more engaged they are, in theory. Costolo had previously touted timeline views as a measure of the engagement users have with the product. The company later downplayed the measure, by not tweeting it on a subsequent earnings call the way it did with other charts.

But the metric was problematic for two reasons. First, it went into decline right after Costolo suggested analysts pay attention to it, making Twitter look bad, like its users were losing interest in the product. Second, timeline views would go down if the product became quicker to use - making the engagement appear to be in decline when in fact the product was becoming more useful and efficient.

Either way, Twitter says it will not talk about timeline views in the future, because it was counting them wrong anyway:

We present and discuss timeline views, but have estimated a small percentage of the timeline views in the three months ended September 30, 2013 to account for certain timeline views that were logged incorrectly during the quarter as a result of a product update.