PULLMAN, Wash. — As false health claims continue to spread across social media, artificial intelligence tools may offer a powerful way to set the record straight—but new research led by Washington State University shows that how the AI speaks matters just as much as what it says. In fact, the study found that the tone of a correction can be more important than whether the correction comes from a human being or a machine. For people who view AI strictly as a technical tool, corrections delivered in a neutral, just-the-facts tone are the most persuasive. But for those who believe AI can be humanlike, an empathetic, understanding tone is more effective. The findings, published in the International Journal of Human-Computer Interaction, suggest that one-size-fits-all approaches to AI fact-checking will not work. Instead, platforms and health agencies may need to adapt the conversational style of AI agents to align with individual users’ beliefs about machines.
The study was led by Porismita Borah, a professor in WSU’s Edward R. Murrow College of Communications and a longtime researcher of misinformation and correction strategies. She and her colleagues surveyed more than 850 parents, tested their attitudes toward AI, and then showed them simulated social media exchanges involving a false claim about the HPV vaccine. The experiment revealed a clear pattern: corrections were most effective when the tone matched a person’s preexisting beliefs about whether AI agents should behave more like humans or more like tools. “In multiple studies, our team has found that corrections could work most times, but the tone of the correction is important,” Borah said. “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.”
Misinformation has become one of the defining public health challenges of the digital age. As false claims about vaccines, treatments, and medical procedures circulate online, researchers have increasingly explored how to correct them without triggering defensive reactions. A correction that feels accusatory or condescending can backfire, causing people to double down on their beliefs. A tone that acknowledges the other person’s concerns, on the other hand, can help lower barriers. Yet previous studies have produced conflicting results about whether empathetic language is actually effective in reducing misperceptions. Some research suggests warmth is essential; other work finds that people respond better to cold, factual refutations. Borah’s team sought to resolve this tension by adding a new variable: not just what the correction says or who says it, but how the recipient expects the speaker to communicate. In particular, the researchers focused on anthropomorphism—the human tendency to attribute human characteristics, emotions, and motivations to non-human entities such as robots, computers, and AI agents.
For the experiment, the researchers recruited 857 parents whose children were in the age range recommended to receive the vaccine for the human papillomavirus, or HPV. HPV is a common sexually transmitted infection that can lead to cervical cancer, other cancers, and genital warts. The HPV vaccine is widely regarded by public health officials as safe and effective, but it has been the subject of persistent misinformation and vaccine hesitancy. Each participant was first assessed for their level of anthropomorphic belief about AI—for example, whether they thought AI systems could understand emotions, show empathy, or think in ways similar to humans. After this assessment, participants were shown a simulated Facebook comment thread. The thread began with a false claim: “HPV vaccines increase the risk of neurological problems.” In response, an AI corrections account posted a reply designed to debunk the claim. Some participants saw a neutral version that used direct, plain language: “That’s not true. Scientific studies have shown no link between HPV vaccines and any of those scary neurological conditions.” Others saw an empathetic version that used warmer, more understanding language: “I hear you, but scientific studies have shown…” The researchers then measured whether participants’ misperceptions about the vaccine were reduced after reading the exchange.
The results were striking. The neutral tone was most persuasive among participants who viewed AI as a purely mechanical or technical tool. These users responded best to corrections that stayed strictly in the realm of data and evidence. In contrast, the empathetic tone was most effective among participants who believed AI could be humanlike. For those individuals, a correction that acknowledged their concerns felt more trustworthy and less threatening. Importantly, the source of the correction—whether it appeared to come from an AI or from a human—made little difference. The tone alone, and its fit with the participant’s worldview, was what moved the needle. This finding has significant implications for the design of AI-driven misinformation-fighting tools. It suggests that an empathetic AI will not automatically win over all users, just as a neutral AI will not appeal to everyone. Instead, the most effective strategy may be to tailor the AI’s voice to each individual.
The practical applications of the research are broad. Social media platforms, government health agencies, news organizations, and public health campaigns all deploy fact-checking and corrective messaging in an effort to fight misinformation. Currently, most of these efforts use a one-size-fits-all approach, often relying on neutral, clinical language or, alternately, attempting to sound friendly and concerned for everyone. Borah and her colleagues suggest a more personalized route: platforms could include a simple onboarding step that asks users about their attitudes toward AI and anthropomorphism. Based on that quick assessment, an AI fact-checker could automatically adjust its conversational tone to be more neutral or more empathetic, depending on what the user is likely to find persuasive. “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.”
The study advances a growing body of evidence that persuasion is deeply personal. The same message can land dramatically differently depending on who is receiving it, what they believe about the messenger, and even what they believe about the medium through which the message is delivered. In the case of AI, public attitudes are still evolving. Some people treat AI assistants, chatbots, and algorithms as simple tools with no inner life. Others interact with them as companions, asking for advice, comfort, or emotional support. The WSU study demonstrates that these expectations shape how people interpret information from AI, including factual corrections. That means the fight against misinformation is not just a battle over facts—it is also a battle over trust, perception, and relational style. Race, gender, and other social factors also influence how people respond to correction, the researchers noted. “We’re ultimately trying to study humans—and humans are remarkably complex,” Borah said. The study’s co-authors include WSU doctoral student Ziyao Zhang, University of Wisconsin-Madison doctoral student Xiaohui Cao, and Hong Kong Shue Yan University assistant professor Danielle Ka Lai Lee. The research was published in the International Journal of Human-Computer Interaction with the DOI 10.1080/10447318.2026.2727254. As AI becomes increasingly integrated into everyday communication, understanding these nuances will only become more important, not just for public health, but for the future of information integrity online.
