"Once a tweet goes through one filter that says 'yes this probably toxic', it then goes through another, there's seven of them. Then once it scores 90 percent 'most likely to be toxic', the bot replies with a positive tweet to balance the negativity on Twitter.
"We also did an experiment in the New Zealand election with replying to some of the candidates with a positive tweet, so if they received a toxic tweet, we wouldn't reply to that toxic tweet, we would just reply to the candidate with a positive message so they actually started to see some of that within their own timeline."
However, the bot solely focused on women candidates to "raise awareness that women candidates tend to face greater toxicity on social media", she said. So there was no data to compare 'toxic' messages to male candidates in New Zealand.
The women often received "awful" messages, with the filters able to pick up on profanity, talk of sexual violence, rape and death threats.
"They are quite vile and quite disappointing," Comer said.
While it's unclear how far the reach of the positive tweets went, she said there was great community engagement in New Zealand with the bot, which has also run in Canadian elections and the current US presidential election.
"I think we had more community engagement [in New Zealand]. For some reason, people here really got what Parity BOT was all about, were really supportive, we got lots of questions within the community on Twitter and then on our website."
- RNZ
