An army of fake Twitter accounts supporting Donald Trump are ramping up their output and will flood the social media platform during the third and final Presidential debates.

A new analysis of Twitter traffic by Oxford University's Internet Institute during the second debate shows messages from fake pro-Trump accounts increased 36.7% over the Republican's first clash with Clinton.

Of 11.5 million tweets collected between the debate October 9 and the four days afterwards, roughly 2.4 million tweets were generated by automated Trump accounts. Similar accounts supporting Democrat Hillary Clinton tweeted just 30% that amount.

Accounts that support the Republican are "tweeting much more aggressively" said Philip Howard, a Professor of Internet Studies at Oxford who gathered and analysed the data with colleagues from the University of Washington and Corvinus University.

"We've noticed that the bot traffic does seem to be increasing," he said.

What are bots?

Bots are automated social media accounts powered by computer code to post content themselves with occasional human intervention to curate those messages.

"The bots are set up with dozens of messages in advance and they're programmed to release those at certain times," said Howard. "They can tweet in interaction with somebody. If they see you tweeting about Trump they will add you into their tweets."

Howard expects the trend to continue during the third and final debates in Las Vegas, Nevada, tonight (October 19). "Increasingly, this kind of automated political campaigning is part of the process," he said.

One computer programmer, MIT postdoc Bradley Hayes, has even created a life-like Twitter bot known as DeepDrumpf, which he taught to adopt Trump's language by feeding it transcripts of his speeches. The bot can even interact with other users, but tweets nothing but nonsense.

Howard and his colleagues have an ongoing project to understand this automated propaganda in Russia and China. But they also examined bot campaigning on social media during the Brexit referendum in Britain too.

Angry messages

The good news is the Oxford analysis found that by and large traffic around the debates is being generated by real people. The bad news is "that bots tend to work well when they are carrying angry messages, crazy claims, nasty language," said Howard, adding that negative memes tend to go further than positive ones.

"A pro-Trump account with angry hashtags will get better pickup than a neutral message," he said.

Governments and political movements around the world are now using bots to "manipulate public opinion, choke off debate, and muddy political issues," the Oxford analysis said.

Spotting bots in the wild

Howard gave a few tips for spotting whether an account is actually a bot. "Highly automated accounts don't have a good profile picture, have 10k followers and don't follow anyone. And there are usually anti-social things about the profile," he said.

Their volume of output is also high, and they latch onto hashtags to help them generate and broadcast their content. A good baseline for spotting them, Howard said, is to look out for accounts that tweet more than 50 times per day.

Still, "it's difficult to relate the bots to changes in public opinion," he said. Sharing what happens with these bots in the third debate will have to wait, he said, since his team intend to publish the data in a peer-reviewed academic paper.

The final presidential debate takes place tonight at the University of Nevada in Las Vegas. It will be moderated by Fox News anchor Chris Wallace. Americans go to the polls on November 8.