Picture a parent scrolling through Facebook late in the evening, already exhausted. They see a comment from someone in a parenting group: “HPV vaccines increase the risk of neurological problems.” Their stomach tightens. They have heard about the vaccine, maybe asked their pediatrician, but the blunt confidence of the post feels alarming. Misinformation like this spreads because it taps into real anxieties, not because people are foolish. And increasingly, the response to such falsehoods is automated: an artificial intelligence fact-checker appears beneath the comment, offering a correction. But according to new research from Washington State University, the mere existence of that correction is not enough. The AI needs to say the right thing in the right way. For some people, a calm, factual statement is most convincing. For others, a warmer, understanding tone works better. In fact, the study found that the tone of the correction mattered more than whether the correction came from a human being or an AI agent. This is a crucial discovery for a world drowning in medical misinformation, because it suggests that the fight against falsehoods is not just about facts—it is about how those facts are delivered and how people perceive the deliverer.
The study, published in the International Journal of Human-Computer Interaction, was led by Porismita Borah, a professor at Washington State University’s Edward R. Murrow College of Communications. Borah and her team conducted a randomized online experiment with 857 parents whose children were in the age range recommended for the human papillomavirus, or HPV, vaccine. HPV is a common sexually transmitted infection that can cause serious health problems, including several types of cancer. The vaccine is safe and effective, yet it has become a magnet for misinformation. Participants were first assessed for their level of anthropomorphism—the tendency to assign human qualities to non-human things like machines and computers. Some people think of AI as a tool, like a calculator or a search engine. Others think of it as more person-like, capable of emotion and understanding. Then participants were shown a simulated Facebook comment thread that began with the false claim about HPV vaccines and neurological problems. An AI corrections account replied to the claim, sometimes with neutral, direct language: “That’s not true. Scientific studies have shown no link between HPV vaccines and any of those scary neurological conditions.” Other times, the AI used an empathetic tone: “I hear you, but scientific studies have shown…” The results were remarkably consistent: the correction was most effective at reducing misperceptions when the tone matched the respondent’s beliefs about AI. People who saw AI as purely technical were persuaded more by the neutral response. People who believed AI could be humanlike were persuaded more by the empathetic response.
Why does tone carry so much weight? Because being told you are wrong is never just an exchange of information. It is a social and emotional experience. When someone corrects a deeply held belief—especially a belief tied to protecting a child—it can feel like an attack on their identity and intelligence. The natural reaction is to dig in, defend the belief, and find reasons to distrust the source of the correction. A flat, clinical response can make the recipient feel lectured or dismissed. An empathetic response, on the other hand, acknowledges the person’s fear and validates their concern before offering accurate information. This is not a new idea in human communication, but it is surprisingly complicated when applied to AI. Previous research on empathetic corrections has produced conflicting results, with some studies suggesting that warmth is helpful and others finding it ineffective. Borah’s team added a crucial layer by considering what the person expects from the AI itself. If someone believes AI is essentially a machine, an empathetic tone might feel creepy or manipulative. It does not match their mental model. But if someone believes AI can be like a human, a neutral tone might feel cold and uncaring. The same message can therefore succeed or fail depending on who is receiving it—even when the source, an AI agent, is identical.
This insight has enormous practical implications. Social media platforms, government agencies, news organizations, and public health institutions are all looking for better ways to combat misinformation. They have tried flagging false content, removing posts, and partnering with human fact-checkers. AI correction systems offer a scalable solution, but their effectiveness depends on personalization. The study suggests that platforms could design a simple onboarding step for users—perhaps a short questionnaire when someone signs up or opens a settings menu—to assess their anthropomorphic tendencies. Based on that information, the AI could adjust its conversational tone. For some users, it might respond with straightforward statistics and citations. For others, it might begin with “I understand why you’re concerned” before presenting evidence. This kind of tone matching is already common in marketing and customer service, where companies tailor their language to different personality types. Applying it to misinformation correction could make AI fact-checkers significantly more persuasive. It would be a shift from treating all users the same to recognizing that people are complex and respond to communication differently depending on their worldview.
At the heart of this research is a simple but profound truth: humans are not logical machines. We are emotional, social creatures who make decisions based on trust, identity, and belonging as much as evidence. Borah has been studying misinformation for a decade, and her work continues to show that there is no one-size-fits-all solution. The tone of a correction is important, but so is race, gender, and a host of other factors. A message that works for one person might backfire with someone else. The study’s authors—including Ziyao Zhang, a doctoral student at WSU; Xiaohui Cao, a doctoral student at the University of Wisconsin-Madison; and Danielle Ka Lai Lee, an assistant professor at Hong Kong Shue Yan University—hope their findings will encourage more nuanced thinking about how to communicate with the public. They are not suggesting that facts are irrelevant. Facts are the foundation. But facts alone rarely change minds. As Borah put it, “The problem of misinformation is critical, and it’s not going away. 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. We’re ultimately trying to study humans—and humans are remarkably complex.”
There is something hopeful in this research. It suggests that we are not doomed to a future of endless, hostile argument online. If artificial intelligence can learn to correct misinformation in a way that respects a person’s fears and meets them where they are, perhaps it can help break through the noise. Imagine a parent who is anxious about vaccinating their child. They see a false claim, and then an AI response appears. For that parent, who has always thought of AI as just a machine, the response might be calm and clinical, offering studies and clear statistics. For another parent, who wonders if there is something more to artificial intelligence, the response might be gentler, acknowledging their concern before sharing the science. In both cases, the goal is the same: to give people accurate information in a way they can actually hear. The research reminds us that effective communication is an act of empathy, whether it comes from a human or a machine. And in a time when misinformation can shape everything from personal health choices to public policy, that reminder could not be more urgent. The future of fact-checking is not just about being right. It is about being understood.