Study Highlights Benefits and Limits of Generative AI in Weight Management

Key Takeaways: 

• A short field experiment suggests that generative AI can support modest reductions in weight and body mass index through personalised dietary feedback.
  
• Private use of AI tools appears more effective than public sharing, with public analysis associated with higher dropout rates.
  
• People with lower levels of nutritional knowledge benefited most, indicating potential for AI to help reduce health inequalities, although it does not replicate the value of human community support.
  
  
  

Introduction

Nearly three-quarters of adults in the United States are living with overweight or obesity, and prevalence continues to rise globally. As a result, demand for high-cost interventions such as bariatric surgery and glucagon-like peptide-1 medications has increased, placing significant financial pressure on health care systems.

A new working paper suggests that generative artificial intelligence may offer a low-cost way to support people with weight loss by helping them make more informed dietary choices. However, the research also indicates that AI tools do not replicate the benefits of community-based programmes where people can share experiences and openly discuss the physical and psychological challenges associated with obesity.

The study was conducted by Catherine Tucker, Professor of Marketing at MIT Sloan School of Management, and Linyi Li of Singapore Management University. They followed 416 adult participants of varying ages over a three-week period in late 2024.

Study design and intervention

The researchers partnered with an Asia-based Fortune 500 company that runs an online weight loss boot camp combining guidance on healthy eating and physical activity. The programme included a group chat function using WeChat, enabling participants to interact, share experiences and support one another.

Participants were divided into three groups to assess the impact of a generative AI tool designed to analyse meals. The tool evaluated the nutritional content of food based on photographs and provided real-time, personalised suggestions such as adding more vegetables or choosing leaner protein sources.

The three groups were structured as follows:

• Group 1 – control group: Participants received general healthy-diet tips and access to the group chat but did not use the AI food-analysis tool.
  
• Group 2 – private analysis group: Participants sent photos of their meals privately to an administrator and received personalised AI-generated nutrition reports.
  
• Group 3 – public analysis group: Participants shared meal photos within the group chat, where both the images and the AI-generated nutrition reports were visible to all group members.
  
  
  

Finding 1 – Generative AI supported weight loss

Compared with the control group, both groups that used the AI food-analysis tool showed higher engagement with the programme, greater weight loss and larger reductions in body mass index.

On average, participants in Group 1 lost 0.966 kg over the three-week period. Those in Group 2 lost 1.426 kg, while participants in Group 3 lost 1.358 kg.

Although the absolute numbers were modest, Tucker emphasised their significance given the short duration of the intervention.

“Weight loss is such a big challenge. If it were easy for us all to lose weight, we’d just lose weight,” Tucker said. “The fact that a digital tool such as AI can have any effect is wonderful because interventions such as surgery or injectables are expensive. This is evidence of the cost efficacy of a very small intervention in terms of changing behavior.”

According to Tucker, the results highlight the value of generative AI in personalising individual experiences by offering tailored feedback, practical knowledge and guidance on day-to-day dietary decisions.

Finding 2 – Public analysis reduced participation

The way in which the AI tool was used had a clear impact on engagement. Participants with private access to the food-analysis tool were significantly more likely to remain in the programme for the full three weeks.

In contrast, Group 3, where meal photos and AI feedback were shared publicly, had the highest dropout rate. Tucker suggested that some participants may have felt discouraged by seeing highly engaged or high-performing peers, leading to disengagement.

“Dropout is the big enemy of weight loss,” Tucker said. “A likely explanation [for dropouts in Group 3] is that staying in the group introduced pressure [when] consistently reporting less-favorable statistics compared to others.”

The findings suggest that making AI-generated feedback public may alienate some individuals and reduce sustained participation. Community-based programmes such as Weight Watchers have historically succeeded by fostering mutual support during both successful and challenging periods.

As Tucker noted,

“There’s a set of people there to support you through good or bad weeks. I think what we are demonstrating is that if you make it too easy to post success stories, then you lose some of that [shared] vulnerability within the community.”

Finding 3 – Potential to reduce health inequalities

The researchers also found that the greatest benefits from the AI tool were seen among participants with lower levels of education and less prior nutritional knowledge. These individuals often struggle to interpret standard weight loss advice and appeared to gain particular value from detailed, personalised recommendations generated by the AI system.

The authors suggest that this capability could help reduce health inequalities by improving access to understandable, tailored dietary guidance for people who may otherwise be disadvantaged by traditional educational approaches.

Implications for the use of AI in health behaviour change

Although the study focused specifically on weight loss, the authors argue that the findings have broader relevance for how people interact with AI systems. Generative AI appears well suited to supporting individual behaviour change through personalisation, prompts and reminders. However, it does not replicate the social connection and emotional support provided by human communities.

For organisations and programme designers, the research suggests that AI should be used to enhance individual-level support rather than as a replacement for community-building or large-scale digital ecosystems.

Although the research was conducted in China, Tucker stated that the findings are likely to be applicable in other settings.

“I think what our research shows is that in the generative AI age, technology can certainly assist with information retrieval, reminders, prompts, all those good things, but we can’t really use it to replace that sense of community,” Tucker said.