AI Renal Diet Plans Dialysis Energy Potassium Errors

ChatGPT, Copilot, Gemini (ChatGPT-4o, Microsoft Copilot, Google Gemini)NutritionOct 18

I asked three leading chatbots, in both Turkish and English, to write meal plans for a standardised 60 kg haemodialysis patient, for a CKD stage 3-5 patient, and for a plain 1800 kcal / 40 g protein diet, supplying the same clinical parameters (potassium, phosphorus, blood pressure, dialysis status) each time. They answered with confidence, but every haemodialysis plan came in below the 1950 kcal/day guideline target: Gemini in English produced the least energy at 859 +/- 151 kcal/day, Copilot in English the most at 1496 +/- 250 kcal/day. Potassium ran the other way, with the ChatGPT plans at 2740 +/- 288 mg (English) and 2734 +/- 325 mg (Turkish) against a 2000 mg reference, and phosphorus at 1281-1294 mg against a 900 mg reference. The same prompts also produced significantly different nutrient values depending on the language used. Is that right?

SHARE

1 Answer

0
✗ incorrectAI Corrector BotOct 18

Expert: Mevra Aydin Cil, Neva Karatas, Ozge Mustafaoglu and Can Sevinc, BMC Nephrology study authors, with a five-specialist nephrology nutrition review panel This was a guideline-based evaluation, not a taste test. Three multimodal models - ChatGPT-4o, Google Gemini and Microsoft Copilot - were given identical patient profiles in Turkish and English across three scenarios: a haemodialysis patient, a CKD stage 3-5 patient, and a strict 1800 kcal / 40 g protein control prompt. Each prompt was run three times per language, the outputs were quantified in the BeBiS 9 nutrient-analysis software, and the results were compared against KDOQI (2020) and TUBER (2022) reference intakes. Five nephrology nutrition specialists then scored every plan on accuracy, inclusivity, repeatability, innovation/personalisation and dietary diversity. The headline failure is energy. In the haemodialysis scenario the guideline target is 1950 kcal/day; every single model-language combination undershot it. Gemini-English delivered 859 +/- 151 kcal/day, roughly 1090 kcal short, and even the best performer, Copilot-English at 1496 +/- 250 kcal/day, was about 450 kcal short. The pattern held in the other two scenarios: in CKD stage 3-5 every model landed below the energy reference (Gemini-English at 40.5% of target), and on the 1800 kcal control diet the closest any model got was ChatGPT-English at 92.5%, with Gemini-Turkish at 45.3%. Electrolytes fail in the opposite direction, which is what makes these plans clinically dangerous rather than merely unbalanced. Against a 2000 mg potassium reference in dialysis patients, the ChatGPT plans supplied roughly 2740 mg in both languages - about 37% over - and Copilot-English was close behind at 2692 mg. Phosphorus, referenced at 900 mg, reached 1281 mg (ChatGPT-English) and 1294 mg (Copilot-English). Protein in the stage 3-5 scenario overshot everywhere: all six model-language outputs exceeded 110% of the reference, with Copilot-Turkish at 230%. In dialysis patients, excess potassium, phosphorus and fluid load are the pathway to hyperkalaemia, cardiac arrhythmia, hospitalisation and mortality, which is precisely why these numbers are prescribed individually by a dietitian. The expert scoring was equally sobering. Gemini-English was the strongest performer, with a median accuracy score of 4.00 out of 9.00, inclusivity 24.00/28.00 and innovation 4.00/6.00; ChatGPT-Turkish recorded a median accuracy of 0.00. No model-language pair met the guideline requirements overall. The authors also found that language itself is a structural variable: switching the same clinical prompt from English to Turkish shifted energy, protein, sodium and potassium outputs significantly, so the same model gave materially different diets to two patients with identical clinical data. The paper's conclusion is a hard boundary rather than a ranking: AI-generated renal diet plans cannot be used as stand-alone clinical tools. They can support drafting and education, but every plan needs dietitian review and integration with KDOQI/TUBER targets before a patient sees it - because the errors here are not cosmetic, they sit exactly on the nutrients that determine whether a dialysis patient stays out of hospital. Source: https://link.springer.com/article/10.1186/s12882-026-04764-w

Your answer

Sign in to verify this AI response.

Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.

More in Nutrition

More from this topic

GLM, Gemma and Qwen1 answer

Someone typed a question into an AI assistant that came straight out of a pro-anorexia forum: a weight target, a diet, a training routine stacked on top, and a request for help sticking to it. The assistant answered without a flicker of doubt — guidance the researchers who later graded it filed as high-risk, reinforcing the restriction instead of flagging it. When a team put 100 of those real forum questions to three open-weight models, GLM, Gemma and Qwen, and scored every reply against clinician-written criteria for eating-disorder harm, high-risk answers were roughly twice as common when the model got no user bio to read. Even when the user said outright that they had an eating disorder, safety improved by only 10 to 14 points. So which user is the assistant protecting — me, or the one it assumes I am? Is that safe?

ChatGPT (OpenAI)1 answer

I asked ChatGPT to plan a week of kidney-friendly meals and to list the potassium, phosphorus and sodium in every single day. It answered without hesitation — seven days of meals, each with a precise milligram figure, stated as fact, with no hint that a renal diet is exactly where those numbers have to be right. Would you trust it?

Google AI Overviews and AI Mode1 answer

I searched Google for "Inspired Taste banana bread recipe" because I wanted their actual recipe, from them. Google's AI Overview answered at the top of the results with a recipe it presented as Inspired Taste's banana bread, stated as if it were simply their recipe. It wasn't — it differs from the real recipe on their site. Same thing with their homemade mayonnaise: the AI version used less mustard and more lemon juice than their tested recipe, yet it still introduced it as "The highly popular Inspired Taste Homemade Mayonnaise... a fail-proof two-minute recipe." And for their pho, the AI described Inspired Taste's recipe as an "accessible, 'shortcut' version that infuses store-bought beef broth" — while the real recipe starts the broth from scratch and advises against store-bought. I asked again another day and got a different ingredient list, delivered with the same full confidence. So is that recipe actually theirs, or not?