Answers are grounded in this paper and cite their sources.
We compared 72 ChatGPT-generated gluten-free menus obtained across three sessions with 24 expert-reviewed reference menus for eight simulated adult profiles and three diet patterns. Primary paired analyses used eight profile-level observations per pattern, with Benjamini–Hochberg false-discovery-rate (FDR) adjustment across 90 exploratory tests. Mediterranean ChatGPT menus contained less carbohydrate and more protein and fat as percentages of energy than expert references. Low-carbohydrate ChatGPT menus had higher energy (+243 kcal/day, FDR q=0.025) and lower estimated cost (−52 TRY/day, q=0.035). Leave-one-profile-out analyses retained the direction of these effects. A 96-record audit identified one inconsistent source protein percentage; uniformly recalculated 4–4–9 percentages preserved the direction of principal contrasts but changed the FDR classification of one high-protein contrast. Soluble and insoluble fibre data were insufficiently validated and were excluded from headline conclusions. Retrospective adult-equation energy targets placed 21 of 24 unchanged expert menus within ±10% of target; 18 of 24 fulfilled their study-specific macronutrient ranges. These findings describe differences from an imperfect expert comparator, not clinical nutritional adequacy or gluten-free safety.
No takes yet. Share an insight, caveat, or question.
Araç et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: