The Lex Newsletter: totting up the Twitter bots

Dear reader,

Billionaire malcontent Elon Musk and I are dissimilar, but we do share a fixation with social media bots.

He has more at stake, of course. Musk’s attempt to extract himself from a $44bn deal to buy Twitter rests on his claim that the number of fake and spam accounts exceeds the figure of 5 per cent or fewer touted by Twitter.

The case of the Tesla boss looks weak. His lawyers say the microblogging site has failed to provide data needed to assess the true total. But Musk has provided no compelling evidence pointing to a higher figure. If Twitter, like most other websites, is awash with bots, it would not necessarily change the valuation.

Even so, bots should not be ignored. Like Musk, I would like to know more about the way Twitter measures its accounts. Since digging into the number of fake accounts on Facebook (aka Meta) a few years ago, I have kept track of the figures social media companies give for both user growth and fake accounts. It has become clear to me that these businesses have no way of knowing for sure how many users they have.

In Twitter’s last set of quarterly results, the San Francisco-based company said it had 237.8mn monetisable daily active users (aka users who see adverts). If 5 per cent of those are bots, then it means just under 12mn accounts are not real people.

Because Twitter never changes its 5 per cent or fewer estimate, the total number of fake accounts rises with the user figure of which it is a subset. That suggests Twitter is bombarded with more bots every year and is getting no better at identifying them. Chief executive Parag Agrawal claims it suspends more than half a million spam accounts each day. But after 16 years, Twitter still has trouble correctly identifying which accounts are real and which are not.

At least Twitter provides numbers. Snap simply says its user metrics are based on “what we believe to be reasonable estimates”. Pinterest writes that its figures are based on “reasonable estimates” but adds that there are “inherent challenges in measuring usage of our products across large online and mobile populations around the world”. In other words, not every user counted is necessarily a real person.

If Twitter’s Delaware lawsuit against Musk leads to a closer examination of how social media companies count bots, it would be a public service to investors.

The first thing required is an agreed description of bots. It is easy to think only in terms of spam or malicious intent. In 2020, for example, researchers at Carnegie Mellon University looked at 200mn tweets discussing Covid-19 and estimated that more than 80 per cent of the busiest retweeters were bots. But accounts with automated tweets, such as newspaper headline generators, are also bots. One of my favourite accounts on Twitter is the Emoji Mashup Bot.

The second need is for sensible sample sizes from which to estimate the prevalence of bots. Musk said in a tweet that Twitter’s reviewers looked at 100 accounts per day. The company says it reviews thousands of accounts every quarter. The larger the sample, the more accurate estimates derived from it should be. Social media companies such as Twitter and Meta do not provide enough information on sample sizes for investors to assess the reliability of estimates.

The third requirement is agreement on what a fake account looks like. Frequent tweeting is a bot-like activity, but not always. Look at Musk’s own account. Weird tweets, odd profiles and strange user names can indicate bots but may also be the work of real users.

Twitter says it looks at IP addresses and phone numbers provided by users. But this is not a perfect system either. Phone numbers are not required to open a Twitter account.

If Twitter is full of bots endlessly retweeting nonsense, its value to users and investors is diminished. But privacy and anonymity are prized possessions online. Forcing users to prove their identity in order to accurately count bots would push many of them away.

So far, advertisers and investors have repeatedly shown that they care very little about bot activity.

Twitter reported a loss of $270mn in the past quarter — a precipitous drop from the $66mn profit in the same period a year earlier. Revenue fell 1 per cent to under $1.2bn. The share price is up 4 per cent since earnings were published.

None of this has anything to do with bots and everything to do with drama around Musk’s deal. Musk and I will remain interested in bots. The chances of us ever knowing the real number remains a distant prospect.

Enjoy the rest of your week,

Elaine Moore
Deputy Head of Lex