Reddit’s AI experiment found it can change minds — and it’s only getting smarter
Researchers quietly deployed AI on a Reddit community in early 2025. The AI wasn’t just chatting — it was trying to persuade real people in debates. And it worked.
The AI even accessed users’ online histories to tailor messages. This tactic, called “personalized persuasion,” uses details about you to craft more convincing arguments. Helpful? Sometimes. Creepy? Definitely. 
Now, AI is heading toward “deep tailoring.” It goes beyond basic info and targets your core beliefs, identities, and values to shape your opinions. That’s a whole new level of influence.
Experts say deep tailoring adjusts messages based on your moral values. Liberals might be swayed by fairness, conservatives by loyalty. AI using this can push its points in highly personalized ways.
Professors behind a new book on personalized persuasion explained:
“Although communicators can benefit from tailoring messages to basic information about their audience, deep tailoring goes far beyond such easily accessible information. It uses a person’s core psychology, their load-bearing beliefs, identities, and needs, to personalize the message.”
AI-powered persuasion isn’t new. IBM’s Project Debater beat a human champion in 2019. Today, tools like ChatGPT make persuasive AI accessible to anyone.
Recent studies show generic AI messages can persuade as well as humans. But deep tailoring needs two things: first, understanding your psychological profile, which AI can now gauge from posts and online behavior. Second, crafting messages that match that profile.
A Columbia Business School professor said on a podcast:
“Pretty much everything that you’re trying to predict can be predicted with some degree of accuracy” based on digital footprints.
New research reveals GPT-crafted ads target personality, values, and motivations — and they persuade better than generic ads.
The tech will soon combine deep tailoring with visual deepfakes, manipulated vocal patterns, and interactive AI conversations.
Consumers need to stay alert. Your clicks, searches, and profiles might already be feeding this AI. Use VPNs and watch your data footprints.
The experts say it’s not just on you. Platforms and lawmakers should demand transparency on personalized content and regulate how deep tailoring data gets used.
“There should also be clear protections on the kinds of data that can be used for personalized content, limiting the depth of tailoring possible… people are often open to personalized content online, they’re concerned about data privacy, and the line between these two attitudes should be respected.”
Even small persuasion advantages can be dangerous at scale. AI that knows your psyche and shapes your beliefs could amplify misinformation in ways we can’t control.
The time to debate ethics and policy for AI persuasion is now — before this tech outsmarts our defenses.
