AI-generated information is becoming more prevalent, raising concerns about how such models can be influenced, particularly with negative narratives. A recent experiment by a GEO agency tested whether AI models could be swayed to return damaging information about a fictional persona, Fred Brazeal. By strategically publishing false information across various established websites, the researchers aimed to see if they could manipulate how AI models like ChatGPT and Perplexity perceived and reported on this made-up individual. To set the stage, the researchers ensured Fred had no prior online presence. They selected ten credible third-party websites to host fabricated claims about unethical behavior and legal troubles. The content mimicked a realistic biographical profile, making it more likely to be accepted by AI models. After publishing, the team closely monitored responses from 11 different AI systems, using a tracking tool to analyze how and when these models referenced Fred and the negative claims. Over several weeks, they found that some models began citing the test websites and incorporating the negative information into their responses. While only Perplexity and OpenAI’s ChatGPT referenced the test sites, Perplexity consistently included the damaging claims, albeit with cautious language. This experiment highlights how easily AI models can be influenced by seemingly credible, yet false, information published online, raising questions about the integrity of AI-generated content and its potential impact on public perception.