As Ghana approaches its presidential election on December 7, researchers have uncovered a network of 171 bot accounts on X that use ChatGPT to write posts favorable to the incumbent political party, the New Patriotic Party (NPP).
According to new research conducted by NewsGuard, a website that offers tools to assess the accuracy and reliability of news outlets, the bot accounts have uniformly been promoting NPP candidate Mahamudu Bawumia and his right-wing talking points, often using the hashtags #Bawumia2024, #NPP, and the NPP slogan #ItIsPossible. The accounts appear to have been active since February.
“Admiring Dr. Bawumia’s commitment to transforming Ghana,” reads an AI-authored bot post from September. “His profound impact on our national policies and progress cannot be overstated. #Bawumia2024 #NPP.”
The accounts — which appear to have AI-generated profile photos, and have names such as “Glenn Washington,” “Netflix Series&Movies,” and “Patriot” — also disparage John Mahama, the presidential candidate from the rival left-wing party National Democratic Congress. These posts often use hashtags such as #mahamaisaliar and #DrunkmaniMahama, accusing Mahama of being a drunkard. (Mahama has denied this.)
“The primary goal of the network appears to be to amplify pro-NPP messaging, promote the Bawumia administration and take aim at the opposition National Democratic Congress,” McKenzie Sadeghi, an editor of AI and foreign influence at NewsGuard who contributed to the research, told Rest of World.
Dimitris Dimitriadis, NewsGuard’s director of research and development, said in an interview with Rest of World that the bot accounts tend to post at “regular” and “predictable” intervals, often ten or more times per day. He said the accounts — which tend to be active between 8 a.m. and 6 p.m. in Ghana — mostly receive likes and reposts from others in the bot network. These regimented patterns in timing and style tipped the researchers off that the accounts may be fake.
NewsGuard’s research team fed all 171 of the posts from the accounts into a tool from Pangram Labs, which assesses the likelihood of text being generated by AI, Dimitriadis said. The tool concluded that it was “highly likely” that all of the accounts were posting AI-generated content created by ChatGPT.
In the past two years, NewsGuard has “followed and tracked down a number of instances of manipulated behavior on X,” Dimitriadis said.
X’s content moderation has significantly diminished since 2022, when Elon Musk acquired the company, then called Twitter. Musk quickly fired most of the website’s content moderators and an estimated 80% of its trust and safety engineers, who identified and removed foreign influence campaigns. X later cut half of its election integrity unit, just weeks after CEO Linda Yaccarino promised to expand it. A 2023 report from the European Unit alleged that under Musk’s leadership, the volume of mis- and disinformation posts has exploded compared to other social media platforms.
The lack of oversight on the platform has created openings for political influence networks. With tools such as ChatGPT, these networks can craft new social media content at scale with little to stop them. “AI seems to be increasingly employed by bad actors who are seeking to influence elections, or influence democratic outcomes,” Dimitriadis told Rest of World.
Sadeghi said that NewsGuard has observed many changes on X since Musk took the reins.
“We found that verified users on X accountfor some of the most viral false or unsubstantiated claims,” Sadeghi said. “The blue check allows misinformation to appear higher and with greater prominence on the platform, which turned out to be a boon for bad actors.”
She added that a recent NewsGuard surveyfound that one quarter of X users who responded believed that users with blue checks were more reliable than those without.
This year is a vulnerable political moment in Ghana. As many as eight people in Ghana were killed during its last election cycle in 2020, according to the US Department of State. Two of the killings were perpetrated by civilians, and two others were perpetrated by military and police officers in the country’s Elections Security Task Force.
NewsGuard said in its report that the network of 171 bot accounts “appears to be the first secretly partisan network using AI to influence elections in Ghana.”
NewsGuard shared the findings of their research with X and OpenAI, Dimitriadis said. At the time of writing, two of the 171 accounts have been suspended, and two have been restricted. Neither company responded to requests for comment from Rest of World.
OpenAI has expressed concerns about people using ChatGPT to generate content that influences political discourse. According to a report OpenAI released in October, influence networks have been using ChatGPT to make political social media posts and blog posts “in the United States, Rwanda, and (to a lesser extent) India and the European Union.” The company claimed that it “disrupted” the activities of more than 20 of these networks.
The network in Ghana did not introduce anything new into the country’s political discourse, Dimitriadis said, since all of the hashtags and talking points in the network’s posts were already popular. He added that it is “inherently difficult” to assess how much impact these accounts have had on Ghana’s political discourse.
One of the objectives of the network was to very clearly drive traffic into a specific set of pro-NPP hashtags,” Dimitriadis said. “But obviously, because of the nature of hashtags, and because there’s thousands of accounts every day that promote these hashtags, I think it’s very difficult to disentangle that impact.”
