Can AI be a trustworthy health resource?

When a woman experiencing severe menstrual pain goes online for answers, she might turn to chatbot. But will the information she finds be accurate? That is one of the questions Viviane Ito, a third-year doctoral student in the UNC School of Information and Library Science, has set out to answer.

Ito arrived at Carolina with a background spanning communications, linguistics and computational linguistics, plus years of professional experience in digital marketing and data analysis. She came here specifically to work with two SILS associate professors: Francesca Tripodi, a principal investigator at the Center for Information Technology and Public Life who studies the intersection of online search engines and society, and Yue “Ray” Wang, an expert in natural language processing and health informatics.

“I’m in a very privileged space where I can work with a sociologist and at the same time have a secondary adviser focused on information retrieval and text mining,” Ito said.

A high-stakes search

Ito’s research focuses on the quality and trustworthiness of health information provided by artificial intelligence tools like Google’s AI Overviews and ChatGPT. As more people rely on these systems for health guidance, the accuracy of what they say matters enormously, she said.

Women in particular face challenges when seeking health care. For example, women with endometriosis often go seven to 10 years without a correct diagnosis. Many turn to online communities and search tools for guidance, which can be unreliable.

“Those gray areas are dangerous, because you can get answers that sound credible but rely on inaccurate sources,” Ito said. “These patients are already in a very vulnerable state, and they end up frustrated with what they find.”

Perimenopause and heart failure are other examples that Ito cites, conditions that are either understudied or too generically named to make reliable information easy to find.

The fellowship that recognized the work

Last fall, Ito was one of the first students at Carolina in a field outside of computer science to receive a 2025 Google Ph.D. Fellow in Human-Computer Interaction.

In her earlier work with endometriosis patients, which drives her dissertation research, she noticed a recurring pattern.

“There are ways that patients talk about their pain, such as certain metaphors, that are repeated across cultures,” she said. “Patients have their own language for what they are experiencing, and AI systems are not always equipped to understand it.”

For her dissertation, Ito plans to test how patients and caregivers interact with AI health information through a neutral interface. Combining behavioral data with in-depth interviews, she hopes to develop practical safety guidelines for building AI information systems.

New school, new possibilities

In July 2026, Carolina will launch the UNC School of Data and Information Sciences, which brings together SILS and the UNC School of Data Science and Society. Ito’s work embodies the promise of the new school by focusing on the human element and asking critical questions of data science solutions.

“I know that a lot of the incoming data science faculty have a similar background in studying AI with a critical perspective, with safety and transparency concerns,” Ito said. “I’ll be very interested in exchanging ideas with them.”

That exchange of ideas, she hopes, will contribute to a body of research with real human stakes: serving patients with accurate information about conditions that medicine has long overlooked.

“The technology is out there, and everyone will use it at some point, so how can we make sure it’s well-built to serve a clear purpose?” she said.