Background: Traditional dietary assessment methods face limitations including reporting bias and scalability challenges. Large language models (LLMs) offer potential for automated food classification, yet their validation in morphologically complex, non-English languages like Polish remains limited. Methods: We analyzed 1992 food items from a Polish long-term care facility (LTCF) cohort using three advanced LLMs (Claude Opus 4.5, Gemini 3 pro, and GPT-5.1-chat-latest) with two prompting strategies: a structured double-step prompt integrating NOVA and World Health Organization (WHO) criteria, and a simplified single-step prompt. Classifications were compared against consensus judgments from two human experts. Results: All LLMs showed high agreement with human experts (90.3–94.2%), but there were statistically significant differences in all pairwise comparisons (χ² = 1174.5–1897.1; p < 0.001). The structured prompt produced very high Recall for UNHEALTHY items at the cost of lower Specificity, whereas the simplified prompt achieved higher overall Accuracy and a more balanced Recall–Specificity profile, indicating a trade-off between strict guideline adherence and alignment with general human judgment. Conclusions: Advanced LLMs demonstrate near-expert accuracy in Polish-language dietary classification, enhancing workflow efficiency by shifting effort toward validation. Expert oversight remains essential, and multi-model consensus alongside language-specific validation can improve AI reliability in nutrition assessment.

Background: Unhealthy food consumption in long-term care facilities (LTCFs) contributes to poor health outcomes among residents. This study aimed to assess its prevalence, identify demographic and physical risk factors, and propose targeted interventions. Methods: A mixed-methods study (2017–2021) analyzed data from 1000 Polish LTCF residents (aged 35–105 years). Anthropometric measurements, bioimpedance analyses, dietary assessments, and physical activity records were collected. Food items were classified as “healthy” or “unhealthy” using an AI-based Large Language Model (LLM), applying WHO guidelines and the NOVA classification system. Logistic regression and chi-square tests assessed associations between unhealthy food consumption and marital status, education level, mobility aid use, and portion control. Results: Unhealthy food consumption prevalence was 15.6%. Married residents had significantly higher rates than unmarried individuals (22.6% vs. 14.3%, p < 0.01). Lower educational attainment correlated with increased risk (partial primary education: 34.7% vs. tertiary education: 8.1%). Mobility aid users exhibited elevated consumption (cane: 34.6%; walker: 22.6%). Poor portion control showed the strongest association (OR = 3.2, 95% CI: 1.8–5.7). Conclusions: Marital status, educational disparities, mobility limitations, and portion control were key modifiable risk factors. Findings suggest the need for targeted nutrition programs, caregiver education, and policy reforms to improve dietary literacy and meal portioning. Future research should validate AI-based food classification methods, assess long-term intervention outcomes, and expand studies to diverse LTCF settings. These findings align with Poland’s National Health Programme and provide actionable insights for global LTCF populations.