Twitter reveals that 8.5% of accounts may be bots (update)

Updated 1:37 p.m. Pacific on Aug. 12 with Twitter clarifications

In a report just filed with the Securities and Exchange Commission today, Twitter has finally put some numbers on the problem of bots posing as human account holders. Advertisers bank on the accuracy of real Twitter accounts to estimate the reach and value of their sponsored tweet spends.

Twitter reports that as much as 8.5 percent of its monthly active users are bots — small pieces of software that automatically pull in data and involve no human interaction with the service. For example, a bot might grab a certain type of update and post it via a third-party mobile app.

Later today, Twitter clarified that what it really meant is that up to 8.5 percent of monthly active users are accessing its service solely through applications that have the capability to auto-pull content from Twitter without the user telling them to do so (like HTC Sense or Flipboard). However, Twitter says there is a possibility that some of these bots can publish content to Twitter autonomously as well.

Twitter also said later today that it estimates that about 5 percent of its user base are spam bots, or bots that publish content automatically to Twitter. That number is not counted in the 271 million monthly active users, Twitter says.

In real numbers, this means that as many as 23 million of the 271 million monthly active Twitter users read and engage with other users or with ads.

What the filing doesn’t call out is how many of these accounts might be spam bots, which don’t pull from but rather post to the service. The idea that bots are counted among Twitter’s account totals is nothing new to advertisers and investors, but the number of bots may be a surprise.

From the filing

“Our metrics are also affected by third-party applications that automatically contact our servers for regular updates with no user action involved, and this activity can cause our system to count the users associated with such applications as active users on the day or days such contact occurs,” the filing reads.

“Up to approximately 8.5 percent of all active users used third-party applications that may have automatically contacted our servers for regular updates without any discernible additional user-initiated action,” the filing reads. “The calculations of MAUs presented in this quarterly report on Form 10-Q may be affected as a result of automated activity.”

It should be noted that Twitter introduced the “bot” measurement itself. It debuted this quarter. Previously Twitter had lumped all accounts with any automation (with human involvement or not) in together, and the company could have presumably continued doing so.

Counting errors

Twitter has had other troubles estimating real user engagement levels in the past, the filing shows.

In the third quarter of 2013, Twitter estimates that it incorrectly logged “a small percentage of timeline views” as a result of a product update. “We believe this estimate to be reasonable, but the actual numbers could differ from our estimate.”

Twitter was unable to track timeline views on mobile devices until June 2012, the filing states.