Dietary choices significantly influence greenhouse gas (GHG) emissions, positioning food consumption as a vital focus for climate mitigation. Traditional eco-feedback tools often rely on abstract numerical data, which may hinder emotional engagement and user motivation. This study explores a plant avatar approach to deliver metaphor-based, gamified feedback on dietary CO₂ emissions. In a 7-day exploratory study, a Telegram chatbot delivered either standard numeric feedback or dynamic avatar responses based on daily food intake. Results showed that participants receiving avatar-based feedback exhibited greater and more consistent reductions in emissions and reported stronger emotional connections and interpretability. While preliminary due to a small sample (N = 8), findings suggest promise for emotionally intelligent, narrative-based interventions in sustainable HCI. This research contributes design insights for more relatable and motivational climate technologies.