🔱  Sign up for a free demo today to see how the latest AI and digital advocacy tools can transform your work.

2.5 Minute Read

From November of 2024 through March of this year, researchers from the University of Zurich conducted a covert four-month experiment on a Reddit forum where people go to have others attempt to change their minds on contentious topics. The results of this experiment have profound implications for the future of public affairs advocacy, voter advocacy, and persuasion in general. So we're devoting two newsletters to this study to give it the airing it deserves. In today's newsletter, we'll discuss what the study showed, and how you can try a light touch version of the experiment for yourself. In next week's newsletter, we'll discuss what it might mean for the future of public affairs.

The experiment deployed AI-powered bots that posted over 1,000-1,700 comments to test whether large language models (LLMs) could persuade users to change their minds on controversial topics. The study tested three approaches: generic AI responses, community-aligned responses using a fine-tuned model, and personalized responses that analyzed users' post histories to infer demographic details like age, gender, political orientation, and ethnicity. This personalized approach didn't just edge out humans—it crushed them. AI proved three to six times more persuasive than human commenters and achieved persuasion rates that placed it in the 99th percentile of all users.

A few provisos: nothing in this study should be taken as gospel. Given that the Reddit users agreed to have others attempt to change their minds, there's certainly the possibility for opt-in bias at play here. And the results would need to be replicated before drawing definitive conclusions. Finally, let's be clear: much of the study was downright unethical. The bots assumed fabricated identities including a rape victim, a Black man opposed to Black Lives Matter, and a domestic violence counselor. But the results are striking nonetheless and have enormous implications for any form of persuasion.

Throughout the entire four-month experiment, not a single user suspected they were interacting with AI rather than humans. The bots blended seamlessly into authentic online discourse, racking up over 10,000 "karma points" and 137 "delta awards"—the subreddit's gold standard for successfully changing someone's mind.

After moderators discovered the unauthorized experiment and filed ethics complaints, Reddit's response was swift and unequivocal. The Chief Legal Officer called the research "deeply wrong on both a moral and legal level," banned every account associated with the study, and initiated formal legal proceedings against the University of Zurich. The researchers, facing the backlash, ultimately decided not to publish their findings. But the AI genie was already out of the bottle.

Try This Self Experiment To See AI's Power To Persuade

Skeptical? Here's a 60-second exercise to see the power for yourself.

Open ChatGPT, Claude, or your AI tool of choice. Feed it some basic demographic details—real or imagined. Age, gender, political leaning, educational background, where you live, etc. Then ask it to craft an argument tailored specifically for you on any topic. Pick something you're genuinely conflicted about. Or pick whether pineapple belongs on pizza (although it absolutely does not).

Here's what you'll see: The AI won't just give you a generic argument. It will reverse-engineer your psychology and hand you a message that feels like it was written by someone who understands you. Someone who speaks your language. Someone who knows exactly which pressure points to press.
That's not the persuasion we use today. That's precision-guided influence. And if it works on you—someone who knows it's AI—imagine how well it works on someone who doesn't.

Be sure to read next week for a peek into the future about what this means for public affairs advocacy.

One Way To Operationalize This Email

🔱 Try the experiment above to see for yourself AI's power to persuade.

🔱 If you're enjoying this content, please consider forwarding this email to a colleague or friend. 

🔱 If you're not already a subscriber, please sign up here to stay up to date on the latest developments in political technology.

🔱 If you have suggestions for future newsletters or any other feedback, we'd love to hear from you.