A chatbot may help reduce the likelihood a person develops an eating disorder, according to a new study. For women at a high risk for an eating disorder, going through a dialogue with a bot developed by researchers reduced concern over body weight and shape — a factor that contributes to their risk.

Previous research showed that digital prevention programs are more effective when they’re guided by a human moderator, says study author Ellen Fitzsimmons-Craft, an assistant professor of psychiatry at Washington University School of Medicine in St. Louis. But it’s hard to get funding in the United States for programs that prevent mental health conditions. The team thought a chatbot could be a cheaper, more scaleable version of such a program that offers “some aspects of moderation in an automated format,” Fitzsimmons-Craft says.

The team developed a chatbot-version of the StudentBodies eating disorder prevention program, which is usually delivered through a website and includes informational content, exercises, and journal prompts. Participants could use the chatbot through texts or through Facebook Messenger.

The study recruited female participants through online ads, fliers, and the National Eating Disorder Association online eating disorder screening test, which is available to anyone on the organization’s website. Women who did not have an active eating disorder but had risk factors for one, like negative body image or excessive concern about their weight, were randomly assigned to either engage with the chatbot or sit on a waitlist. The chatbot offered eight conversations about topics around body image and healthy eating, and women who used the bot were encouraged to have two of the conversations each week.

At three- and six-month check-ins, women who talked to the chatbot had a bigger drop in concerns on a survey about their weight and body shape — a major risk factor for developing an eating disorder — than women in the waitlist group. The study, published in December in the International Journal of Eating Disorders, also found some indication that women in the chatbot group were less likely to have developed a clinical eating disorder by the end of the six months than women in the waitlist group. “This could have a great public health impact if we can continue to find ways to disseminate it widely,” Fitzsimmons-Craft says.

This is one of the first studies testing the effectiveness of chatbots in healthcare, even though they’re widely used in medical settings to assist with everything from COVID-19 symptom screening to therapy. Research has lagged behind their use, but there’s a growing body of work highlighting the role they can play in some areas of healthcare.

Fitzimmons-Craft thinks chatbots like the one her team developed could be integrated with things like the National Eating Disorder Association screening tests, or in places — like colleges — where people are likely to start showing signs of an eating disorder. The chatbot isn’t designed for people who already have a clinical disorder, but people who show risk factors could be directed toward it.

It’s also possible to identify people who are at risk for an eating disorder through their web search history, Fitzimmons-Craft says, and likely through their behavior on platforms like Facebook and Instagram. Those platforms are linked to body image issues for teen girls, research and whistleblower reports this fall continue to show. Partnering with platforms like Google or Instagram could help flag behaviors that shows someone is at risk, and pair them with the bot. “I think that’s absolutely worth exploring,” she says. “It’s ripe for identifying individuals who could use help with these issues and intervening on the spot.”

Chatbots could also be useful for other types of mental health issues where prevention is important, Fitzimmons-Craft says. They’re not going to be right for everyone or for every issue, but they’re a cheap tool and the barriers to entry are low. People who don’t have an active disorder might not want to do an intensive program, and it might be hard to get funding for involved prevention tools. But at-risk groups could still benefit by being nudged away from a potential problem, she says. “For someone who doesn’t really want or maybe even actively need help — in our study folks were high risk, they didn’t have a clinical problem — it might be just the right thing.”