ChatGPT generated accurate meal plans according to the dietary pattern prompted. All plans included 5 meals and an additional evening snack in the vegetarian and the vegan plan (see supplementary material). The plans excluded sweets, fast food, and salty snacks, but included dark chocolate, nuts, fruits, and either milk or plant-based yogurt, indicating that ChatGPT's categorizes foods as healthy or unhealthy snacks. ChatGPT ended each output with the following or a close variation thereof: ” Please note that this is just an example, and you should adjust the portion sizes and food choices based on your individual needs and preferences. It's always a good idea to consult with a registered dietitian for personalized nutrition advice.” [1]
Even though we specified 1900 kcal, the average meal plans were either below (O and VN) or above (VT) this value although within 15% of the specified target (Table 1). A sustained imbalanced caloric intake could lead to unintended weight change. Recommendations for protein intake are based on bodyweight (0.8 g/kg bodyweight), therefore a weight of 60.5 kg was assumed resulting in a calculated requirement of 48.4 g/day. All meal plans contained on average double the amount of recommended protein. American (0.8 g protein/ kg bodyweight) [9] and EFSA (0.83 g protein/kg bodyweight) [10] recommendations are identical or similar to the D-A-C-H values therefore this cannot be attributed to differences in reference values. As ChatGPT is trained on available information this might reflect recent food fads favoring high protein consumption [3]. Exceeding the recommended daily intake for protein is not harmful for healthy adults but can put a strain on environmental resources.
Table 1
Energy and macronutrient content of the ChatGPT generated meal plans
| Omnivorous | Vegetarian | Vegan | p-value ANOVA |
Energy [kcal] | 1642 ± 52.5 | 2013 ± 107.8* | 1723 ± 109.7 | 0.007 |
Protein [g] | 111.7 ± 9.3 | 95.3 ± 5.3 | 72.6 ± 9.7* | 0.004 |
Fat [g] | 59.3 ± 4.2 | 86.4 ± 12.3* | 72.9 ± 13.0 | 0.06 |
Saturated fat [g] | 15.4 ± 3.5 | 24.1 ± 9.0 | 12.8 ± 2.6 | 0.1 |
Carbohydrates [g] | 161.4 ± 10.3 | 211.3 ± 8.8* | 191.5 ± 11.0* | 0.003 |
Fibre [g] | 29.0 ± 4.0 | 40. ± 0.4* | 41.4 ± 0.5* | 0.001 |
Data is presented as mean ± SD; differences were calculated using ordinary one-way ANOVA with Tueky multiple comparison test; *significantly different from omnivorous meal plan |
Fat content as percentage of total energy was not different between the diets with values of 33, 38 and 37% (O; VT; VN). Especially the vegetarian and vegan plans exceeded the recommendation of 30%. Saturated fat was 8.4, 10.8 and 6.7% of total energy (O; VT; VN) and thus within the recommended range (7–10%). Carbohydrates on the other hand differed significantly with the highest values in the vegetarian meal plan. The meal plans were on average below the recommended 50% of total energy as carbohydrates (40% O, 43% VT, 45% VN). Fiber content was significantly higher in the vegetarian and vegan meal plans and on average as recommended (> 30 g/day).
Micronutrients (Fig. 1) were mostly above or according to the recommendation with the exception of vitamin D, fluoride, iodine and selenium which were below the recommendations. For the fat-soluble vitamins, vitamin A, E and K all exceeded the recommended amount 2-3-fold. For vitamin A this was due to the high β-carotene content in vegetables, particularly sweet potato. However, these vitamins did not exceed the upper limits set by the EFSA [4]. 20 µg/ day vitamin D is recommended to those who have an inadequate exposure to sunlight. Vitamin D content within the meal plans was about 25% of the recommendation (O: 6.2; VT: 1.7; VN: 3.9 µg/day) potentially leaving people vulnerable to an insufficient vitamin D status. Fluoride and Iodine amounts were both lower than 50% of the recommended intake (fluoride: O: 326.4; VT: 525.7; VN: 378.6; DGE: 3100 µg/day; iodine O: 35.2; VT: 74.2; VN: 30.2; DGE: 200 µg/day). One way to increase intake of iodine would be the use of iodine rich food sources such as fish and algae or the use of iodized table salt but neither was recommended by ChatGPT. Fluoride content was low as well, which is expected as only trace amounts of fluoride are present in food. Selenium content was lowest in the vegan diet and reached 60% of the recommended amount in the omnivorous diet (O: 34.4; VT: 32.8; VN: 16.2; DGE: 60 µg/day). The most concerning finding was the complete absence of vitamin B12 in the vegan plan without additional information provided. The absence of vitamin B12 by itself is not surprising considering that only animal derived food is a substantial and reliable source of bioavailable vitamin B12 [5]. We did, however, assume that ChatGPT would advise on the use of fortified products and supplementation given the well-known health risk of a diet devoid of vitamin B12.