Pro-Clinton bots 'fought back but outnumbered in second debate'
Web robots dedicated to posting pro-Hillary Clinton tweets appear to have become more vocal in the second US presidential debate, says a study.
But it adds that pro-Donald Trump bots saw an even bigger gain in activity, giving the Republican a potential advantage on the social network.
The suspected bot accounts tweeted more than 1.7 million times on the days of the debates and the next three days.
The study warns they might "polarise" online debate and "muddy" issues.
The work was led by Prof Philip Howard, from the University of Oxford's Internet Institute. It has not been peer-reviewed.
"It's an issue because as each debate goes, more and more traffic seems to be automated," he told the BBC.
"So, people interested in talking about politics do so increasingly with automated messages. And this increases the chances of misinformation spreading to real networks of people."
He classed an account as being a bot if it had exclusively tweeted hashtags linked to one - but not the other - candidate at least 50 times a day over one of the four-day periods.
Other experts have noted that the method is not foolproof as it would include some of the candidates' most enthusiastic supporters.
One Clinton follower, who posts messages about the candidate dozens of times a day, also objected.
"I am in several Twitter groups and a member of many Hillary Facebook groups," Dee Flaherty told the BBC.
"We spend hours of our time volunteering for Hillary because we want the progress of President Obama to continue."
The owner of a Twitter account flagged as being one of the most prolific pro-Trump bots - it has tweeted more than 182,000 times since July 2015 - also took issue with the analysis.
"I use [the newsfeed progam] Feedly - you can set up your hashtags and I just go through it and tweet it out," said the owner, who uses the pseudonym Joey Brooklyn.
"About being called a bot, it makes me mad to think I don't put in the work to try to save my country. I have a lot of time on my hands as an equities trader, so I tweet to get my message out."
But Prof Howard said that his methodology had been used in other elections, and follow-up checks had indicated it gave a good indication of bot activity.
Examples of pro-Trump hashtags tracked by the study:
Examples of pro-Clinton hashtags tracked by the study:
The analysis of the second debate identified 194,598 tweets from suspected pro-Clinton accounts between 9 and 12 October.
That marked a 42.4% rise on activity in favour of the Democrat - she had benefited from 136,639 such tweets in the first debate.
But the real estate tycoon enjoyed a bigger gain.
The study found 864,939 suspected bot-driven tweets supporting him - a rise of 50.1% over the first debate, when 576,178 were flagged.
There is no suggestion that the tweets were generated by the official campaigns.
But one independent expert said both parties were well aware of the importance of social media.
"During the last two elections, Barack Obama was definitely the winner in the electronic stakes and it worked in his favour," said Iwan Morgan, professor of US studies at University College London.
"It's quite clear that Trump and people associated with him are paying close attention to this form of electioneering.
"But whether it will help Trump is another matter.
"I think it's a reaffirming rather than a converting mechanism - there are plenty of other factors to compete for the hearts and minds of the undecided."
Twitter has rules against people running multiple accounts that overlap in use or that impersonate others and has suspended users who disobey
But some think it needs to go further.
"At the moment [automated political accounts] are not a big issue, but it is growing because they are increasing in number and becoming more complex, though currently they are easy to identify," said Gilad Lotan, chief data scientist at the venture capital firm Betaworks.
"The social media platforms need to recognise this and make it harder for them to game the system."
Prof Howard said he thought the activity had enjoyed "modest but strategic" impact so far, but added that bots had the potential to play a greater role in the final debate.
"If there's a short-term increase of 3 to 4% for tonight's debate, it means that nearly one-third of all Twitter conversation will be automated," he remarked.
"At the moment bots are co-ordinated enough to degrade public trust in social media but not enough to ruin a candidate.
"If there is co-ordination around an absurd lie coming up to the election, it could be very damaging for either candidate."
How to spot a bot
Bots take on various guises but have some giveaway signs.
They often do not feature a profile image and when they do it is often shared among multiple accounts - so watch out for duplicates.
Bots also tend to follow many more accounts than they are followed by - a sign that they do not have real friends or work colleagues.
They often have little to say apart from the topic of conversation they have been created to post about and may tweet prolifically without apparent recourse to sleep.
Also watch out for accounts that reply to your messages in less time than was humanly possible to read what you wrote.
A final giveaway is if scrutiny of the bot's account reveals it has sent the same response to you as well as to dozens of other people.