SWITCH Miso
How can conversational agents improve literacy around healthy and sustainable eating?
The complexity of dietary information is a major barrier to adopting healthier, more sustainable diets. Miso is a research prototype developed by the EPFL+ECAL Center as part of SWITCH, a Horizon Europe project promoting this transition. Building on learnings from SWITCH, the team explored how conversational AI could change the way people engage with information about food, guiding them through a personalised journey to build their food literacy.
Food systems place growing pressure on the environment, and poor nutrition remains a major global health concern. Although dietary guidance exists, it is rarely adapted to individuals or their context. Different sources also offer conflicting recommendations, leaving people confused and unable to decide. Food literacy, meaning the knowledge, awareness and skills needed to make informed food choices, offers a way forward. Conversational AI could make this approach personal and accessible at scale, but little is known about how people experience it in everyday life, or how it can support eating that is both healthy and sustainable. Miso was designed to explore exactly that.
Encouraging self-reflection through microlearning and a proactive approach can promote user intentions to change their behaviour.
Miso is an AI-powered conversational agent designed to help people make healthier and more sustainable dietary choices.
Unlike many chatbots which offer users the chance to ask them anything, Miso is a structured five-day experience moving from foundational topics of healthy and sustainable eating to more nuanced details. Each day, the agent draws on three design principles:
- Reflection. Miso asks people to connect abstract principles to their own lives, for example by describing their last meal or imagining a new dish they could cook this week.
- Proactive guidance. Miso opens each session with a topic, moves through a planned sequence of activities, and delivers short PDF Tools (such as an overview of food labels or a calendar of seasonal produce). Miso takes the lead so people don’t need to know what to ask.
- Microlearning. Short daily sessions spaced over five days build knowledge progressively, with the Tools serving as takeaways between sessions.
We evaluated Miso in a five-day field study with 14 participants, combining daily diary surveys, questionnaires and interviews. The study showed that conversational AI can support food literacy by actively structuring reflection and learning over time. Participants particularly valued Miso’s personal reflection prompts, proactive guidance and short daily Tools, suggesting that a guided conversational experience can help people connect abstract principles of healthy and sustainable eating to their own habits. Participants also reported increased knowledge and intentions to adjust their eating habits. These early results suggest that conversational AI can play a meaningful role in supporting food literacy. They also point to next steps: balancing structured guidance with user autonomy, adding safeguards around trust in AI-generated advice, and integrating behaviour change expertise into a sensitive and complex domain like eating.
This project was presented as a full paper at the 2026 NordiCHI conference in Vaasa, Finland.
Miso was developed as a research prototype for the study and is not a public product. Its code is openly available on GitHub.
Design and Research
Emily Groves, Margherita Motta
Engineering
Delphine Ribes, Andrea Rovescalli, Gabriel Luthier
UX Research Lead