Donald Trump support during presidential debate was inflated by bots, professor says

Many of the Twitter users supporting Donald Trump after the presidential debates were bots, according to a new analysis.

More than four times as many tweets came from automated accounts that supported Mr Trump than they did backing Hillary Clinton, according to Philip Howard from the University of Oxford.

The robot tweets helped give the appearance that Mr Trump had more support than he did, according to Professor Howard. That apparent surge in support was referenced repeatedly by Mr Trump, who claimed that despite what the official polling showed he had actually won both of the debates.

The software used to create the bots also help manipulate public opinion and “muddy political issues”, according to the authors of the paper.

The research looked at tweets posted on 26 September, the day of the debate, and the three days that followed it. It looked at which accounts were posting pro-Trump and pro-Clinton hashtags, and found accounts that were only posting one and not the other.

About 1.8 million tweets were sent supporting Trump, and only 613,000 tweets were sent for Clinton.

The researchers found that 32.7 per cent of the Trump ones appeared to be sent by accounts that were bots. Only 22.3 of the pro-Clinton ones were sent from such accounts.

But the definition of bot used in the research has prompted some complaints.

An account was classified as a bot if it had tweeted at least 50 times a day in the time being studied, meaning it sent at least 200 tweets over four days. But it’s possible that those accounts may simply be actively used – there certainly are a number of accounts that tweet more than that and are legitimate – and so the numbers could also reflect that people who tweet more often are more likely to support Donald Trump.

Professor Howard said however that the findings showed that “on the balance of probabilities, if you pulled out a heavily automated account the odds are four to one that you'll find it's a bot tweeting in favour of Trump”.

He said that most of the apparent bot tweets were sent overnight and shared similar information. They showed other behaviour that is “not human”, Professor Howard said, and tended only to tweet about the candidate that they had chosen.

The team chose the number of tweets per day after analysing data generated during the Venezuelan election and the Brexit vote. They found that when they looked at accounts that broke through the criterion, most of them tended be inhuman, and that most legitimate accounts don’t tend to tweet enough to reach the criterion.

It isn’t possible to know who those accounts were created by and what they were being used for, the researchers note. There is no suggestion that any of the tweets were sent or paid for by the official campaigns, or even necessarily that those accounts were being used primarily to support the candidates they were posting about.