Imagine this: you’re scrolling through social media, and a relative shares a scary post. It claims the HPV vaccine causes neurological problems. You know it isn’t true, but thousands of people might see it and wonder. Then an AI-generated correction appears in the comments. Do you believe it? According to new research led by Washington State University, your answer may depend less on what the correction says and more on how it says it. In an era of rampant medical misinformation, artificial intelligence can be a powerful tool for setting the record straight. But the study, published in the International Journal of Human-Computer Interaction, found that the tone of a correction is just as important—if not more important—than the source. For people who view AI strictly as a technical tool, corrections delivered in a neutral, just-the-facts tone are most persuasive. For those who believe AI can be humanlike, an empathetic, understanding tone works best. The researchers found that tone mattered more than whether the correction came from a human being or an AI agent. “In multiple studies, our team has found that corrections could work most times, but the tone of the correction is important,” said Porismita Borah, professor in WSU’s Edward R. Murrow College of Communications and the paper’s corresponding author. “In this study, it did not necessarily matter whether the correction came from a human being or an AI agent. What mattered was the tone and how the tone aligned with people’s beliefs about whether AI agents should be more humanlike or more machinelike.” In other words, it’s not just about being right. It’s about speaking the same emotional language as the person you’re trying to reach.
This idea gets to something called anthropomorphism—the tendency to give human qualities to machines, animals, or other non-human things. Some people look at AI and see a clever calculator: no feelings, no personality, just cold logic. Others naturally imagine AI as something more, almost like a friendly assistant or a wise digital companion. The WSU team realized that this difference in perception could be the key to making AI fact-checkers actually work. If you believe AI is basically a machine, then a correction that sounds like a machine—plain, direct, and unemotional—feels authentic and trustworthy. If you believe AI can be humanlike, then a correction that acknowledges your feelings feels respectful and considerate. But when the tone clashes with your expectations, the correction can fall flat, or even backfire. It’s a bit like calling customer service. Some people want the representative to get straight to the point, skip the small talk, and solve the problem efficiently. Others want to feel heard, understood, and maybe even sympathized with. The same information can be delivered in both ways, but only one will feel right to you. The authors of the study suggest that social media platforms, government agencies, and news organizations could use this insight to tailor AI fact-checking agents to each user’s level of anthropomorphism. For example, a simple onboarding question—like “Do you think AI can understand how people feel?”—could be enough to determine the best conversational style. Then the AI could adjust its tone accordingly. The study was co-authored by Ziyao Zhang, a PhD student at WSU, Xiaohui Cao, a PhD student at the University of Wisconsin-Madison, and Danielle Ka Lai Lee, an assistant professor at Hong Kong Shue Yan University. Together, they have opened up a practical and surprisingly personal path toward fighting misinformation.
Why is this so important? Because correcting misinformation is hard. It takes more than facts to persuade someone that a claim is incorrect. When people are confronted with evidence that they are wrong, they may feel challenged, insulted, or embarrassed. Their natural reaction is often to become defensive, to double down on what they already believe, or to reject the source of the correction entirely. As social media has grown, the spread of dubious health claims has become faster and more dangerous, and researchers have spent years trying to figure out how to correct falsehoods without pushing people further away. Borah has been studying this subject for a decade. Her earlier work and the work of others have produced conflicting conclusions about whether empathy in correction messages helps or hurts. Some studies suggested that a warm, understanding tone reduces misperceptions; others found that it made no difference or even made things worse. The new research helps explain why. The effectiveness of an empathetic tone, it turns out, depends on the person receiving the correction. People who see AI as humanlike respond to empathy because it matches their mental model. People who see AI as a machine may find empathy confusing, or even manipulative. For them, a straightforward, evidence-based correction is more persuasive. The key isn’t to choose one tone for everyone; it’s to match the tone to the individual. This is a subtle but powerful shift in how we think about persuasion. Instead of asking “What is the best way to correct misinformation?” the researchers are asking, “What is the best way to correct misinformation for you?”
To test this, the research team designed a randomized online experiment with 857 parents of children in the age range recommended to receive the HPV vaccine. HPV, or human papillomavirus, is a common virus spread through sexual contact. It can cause a variety of health problems, including several types of cancer. The HPV vaccine is widely considered safe and effective, yet it has been the subject of persistent and sometimes bizarre misinformation. That made it a perfect case study for understanding how to combat false health claims. In the experiment, participants were first evaluated for their level of anthropomorphism—how much they tended to see machines as humanlike. Then they were shown a simulated Facebook comment thread that began with a false claim: “HPV vaccines increase the risk of neurological problems.” An AI corrections account responded to the claim in the thread. For some participants, the correction was neutral and direct, using plain language like: “That’s not true. Scientific studies have shown no link between HPV vaccines and any of those scary neurological conditions.” For others, the correction was empathetic and warm, like: “I hear you, but scientific studies have shown….” The researchers then measured whether participants still believed the misinformation after seeing the correction. The results were clear: the correction was most effective at reducing misperceptions when the tone matched the respondent’s anthropomorphism beliefs. In other words, the people who thought AI was humanlike were more convinced by the empathetic correction, while those who saw AI as a machine were more convinced by the neutral one. The mismatch produced weaker results. It wasn’t that one tone was universally better. It was that the right tone for the right person made all the difference.
These findings have significant implications for the way we build and deploy AI systems in public communication. Social media platforms could identify users who are likely to respond to empathy and adjust their fact-checking bots accordingly. Government agencies and public health organizations could design AI assistants that adapt their language style to each person’s expectations. News organizations could use similar tools to engage with comment sections and correct falsehoods before they spiral out of control. The goal isn’t to manipulate people, but to communicate in a way that actually reaches them. If someone feels attacked or condescended to, they will stop listening. If they feel understood, they are more likely to accept the information being offered. Borah’s team is quick to note that this is only one piece of a much larger puzzle. “The problem of misinformation is critical, and it’s not going away,” Borah said. “The effectiveness of corrections depends on a lot of factors—for example the way you talk to someone when providing accurate information—an empathetic tone may often work better than a condescending one. Race, gender, and other factors also matter. We’re ultimately trying to study humans—and humans are remarkably complex.” That complexity is exactly why this research is so valuable. It doesn’t offer a one-size-fits-all answer. Instead, it reminds us that even artificial intelligence must understand human nature to be effective. A correction that ignores how the reader feels is just another piece of noise in an already noisy world.
In the end, this study is a reminder that misinformation isn’t just a problem of facts and falsehoods. It’s a problem of trust, identity, and emotion. People believe what they want to believe, and they reject what feels wrong—even if it happens to be true. Artificial intelligence, for all its power, cannot solve this problem by simply being smarter. It has to be wiser. It has to recognize that a person’s beliefs about technology shape how they respond to it. It has to adjust not only what it says, but how it says it. The good news is that AI is uniquely capable of doing this. It can be trained to detect subtle cues in language and tailor its responses in real time. It can learn to be neutral for some and empathetic for others. It can even learn to be both, depending on the context. But this ability must be guided by research like Borah’s, which shows that the human element cannot be ignored. As misinformation continues to evolve, so must our strategies for fighting it. We cannot simply shout louder or post more charts. We have to listen, understand, and adapt. The future of fact-checking may not be a single authoritative voice, but a chorus of AI agents, each speaking in a tone that resonates with the person on the other side of the screen. And if we can manage that, we might just have a real chance at restoring trust in the truth. Because in the fight against misinformation, the most important thing isn’t just being right. It’s being heard.