IGP Panel Examines the Business of AI Slop Ahead of 2026 Midterms
Ahead of US midterm elections, Columbia SIPA’s Institute of Global Politics (IGP) organized a roundtable on September 25 to discuss AI slop – the cheap, easy-to-produce, algorithmically optimized content flooding voters' social media feeds.
Much of the discussion focused on the challenges of defining AI slop, the scale of the problem, and the use of behavior-based rules in places like Taiwan as an alternative to content moderation to address the problem.
Nobel laureate and SIPA Professor Maria Ressa opened the event.
Panelists included Zach Praiss, head of state strategy at Reset Tech; Isabel Linzer, elections and democracy policy analyst at the Center for Democracy & Technology; Aidan Walker, creator-in-residence at the Carnegie Endowment for International Peace; Jen Weedon, lecturer and researcher at Columbia SIPA; and Audrey Tang, Taiwan's founding minister of digital affairs and an IGP Carnegie Distinguished Fellow.
Camille François, assistant professor of practice at Columbia SIPA and codirector of IGP's Technology & Democracy Initiative, moderated the conversation.
Linzer said that 2026 is a different electoral environment from 2024. Back then, campaigns largely avoided generative AI for fear of media backlash and voter distrust. “That norm is gone,” she said. Chatbots are now part of how voters find information, and they draw information from online sources, including slop, whether right or wrong. Linzer pointed to data voids – “corners of the internet where there just isn't a lot of information” – including local races and candidates across more than 10,000 election jurisdictions in the US.
Before the panel could address the problem, it had to define AI slop, the subject of an IGP report published back in June. The panelists described the ambiguity and challenges involved. Linzer asked, “Do I want X deciding that?” Jen Weedon, a researcher who previously worked with Meta and coauthor of the report, expressed caution, arguing that the panel needed to distinguish between real politicians using slop-like aesthetics in their political advertising and marketing, content creators producing AI slop to make money, and influence operations through which governments deliberately try to steer what people see online. The very definition of AI slop could be used by politicians to remove content that they dislike.
Walker, a content creator who previously worked at Know Your Meme, said the bigger concern with AI-generated content, whether text, images or video, is not that it is always malicious but rather that it displaces original content. He compared AI slop to fast food at McDonald’s: feeding people cheap, good-enough food at scale is more efficient than investing in a food system that would earn big companies less. Slop is similarly easy to produce and ready for mass consumption.
Praiss said the money at stake for tech firms, as well as content creators, explains the current flood of AI slop. The platforms reward creators who post engaging content at volume, and they earn more as that content keeps people online longer. Coordinated campaigns run from countries such as Sri Lanka, Pakistan, and Indonesia target Western audiences, he said, because “some people's attention has more value than others.” Praiss described an academy in Sri Lanka where one entrepreneur, by his own account, raised hundreds of thousands of dollars from slop and now teaches others to do the same. The IGP report puts the gap in numbers: YouTube ad rates reportedly run about $1 per 1,000 views in Pakistan and $5 to $7 in the US.
Another example showed how a 17-year-old student in Burkina Faso posted an AI-generated video that drew more than 12 million views on Facebook. The video falsely claimed there had been a coup in France. The French president's request to Meta to take down the video was declined. The teenager earned about 7 euros from the video, then used the attention he received to advertise his course on creating AI slop videos. Ressa noted that even a head of state lacks the power to get such content taken down.
Tang described a response that Taiwan built without touching content. Deepfake ads of Nvidia chief executive Jensen Huang crowded out small-business ads, and people lost millions to the scams that followed. Under Tang’s direction, Taiwan convened a 447-person citizen assembly. The assembly backed three measures: verifying advertisers, making platforms jointly liable for scam losses, and slowing the videos of platforms that ignore the rules. The last applies to all of a platform's videos. “It's not content specific. It's entirely behavior,” Tang said.
Once implemented, impersonation scam ads fell by more than 94 percent by 2025 and 99 percent this year, Tang said. Japan is now working on a similar approach, with a citizen assembly and a crowd-trained classifier.
Panelists discussed whether interventions should focus on labeling, data provenance, platform design, monetization, privacy, or digital literacy. One speaker suggested that platforms could potentially continue hosting certain kinds of political expression while limiting the financial incentives attached to accounts producing large volumes of problematic content. Others emphasized the limitations of regulation and the different approaches available in the United States and elsewhere.