Researchers find deceptive bots coordinating online chatter in several major races

SAN FRANCISCO — Deceptive bots have tried to manipulate the online chatter in some of the tightest races of the midterms.

Researchers at Cyabra, a social media monitoring company, said last week they found signs of coordination by inauthentic social media accounts from Aug. 31 to Oct. 19 in several major matchups in Georgia, Kansas, Maine, Michigan and Ohio. In the contests for governor in Georgia and Ohio and for Senate in Kansas and Maine, fake bots attacked both the Republican and Democratic candidates, the research showed.

Cyabra reviewed discussions about six midterm races on Facebook, Instagram, TikTok and X. Out of 274,000 posts, researchers said roughly 1 in 9 about each race (and around 19,300 accounts out of 168,000 total) were inauthentic. Such bots also posted 1 in 4 replies to X posts from the candidates running to be governor of Georgia — Rick Jackson, a Republican billionaire, and Keisha Lance Bottoms, the Democratic former mayor of Atlanta.

Real accounts can also be automated and post heavily about certain topics — especially with artificial intelligence helping produce content and distribute it — but those accounts do not mask their identity, said Keith E. Noble, a Cyabra vice president. The company, he said, uses AI tools to analyze hundreds of indicators of coordinated inauthentic activity, such as synchronized bursts of posts around debates and other events, multiple comments on news articles, account creation date and narrative echoes.

In Kansas, Adam Hamilton, a Democrat at the head of the largest church in the state, and Sen. Roger Marshall, the Republican incumbent, are facing off for a Senate seat. One online profile weighing in on that race did not exist before April and has since posted nearly 11,000 times, sometimes 50 times a day, often at the same times.

“The story is bigger than just AI or automated models — you have these sock puppets, these managed personas where they’re pretending to be your neighbor next door,” Noble said. “There are human beings behind this inauthentic activity that have an agenda.”

It is unclear what — if any — effect the bots will have on the election discourse, or where they originated from. Real accounts still attracted 99.8% of views on X, the researchers said.

Inauthentic narratives are insidious, Noble said, spreading manufactured noise like a cancer while sowing confusion and animating anger. Other researchers have also found signs of bot manipulation in California, Texas and elsewhere. The result is a distorted sense of truth in an information environment already short on trust.

Access to AI tools will only make such campaigns easier to execute, he added.

“The system hasn’t built in antibodies,” he said. “It requires a different defense muscle that we’re all going to have to operate in this era because these inauthentic narrative campaigns are going to be a permanent component of our election systems moving forward.”