This randomized parallel trial evaluated whether recommending simple food swaps in a naturalistic online grocery store changes the healthfulness and carbon footprint of grocery selections. In May–June 2025, 1,201 US adults completed three weekly shopping visits: a baseline no-intervention visit followed by two visits in a randomized trial arm (climate swaps, health swaps, climate + health swaps, or control). Primary endpoints were average Ofcom Nutrient Profiling Model score (1–100) and average kg CO2-eq per kg of selected items. Compared with control, healthfulness improved in all swap arms; reductions in carbon footprint occurred only in arms that targeted climate. The sample and online setting limit generalizability to the broader US population and to in-store shopping.
Poor diet quality contributes substantially to global mortality, and food production accounts for a large share of human-caused greenhouse gas emissions. Experts recommend diets higher in fruits, vegetables, and legumes and lower in red and processed meat, but consumers can find these guidelines difficult to implement because food choice involves comparing many items across multiple dimensions. Recommending one-for-one substitutions, or “swaps,” offers a potentially scalable, simple approach: swaps present a similar product with better nutritional characteristics or lower production emissions (for example, a turkey burger in place of a beef burger).
The trial enrolled 1,201 US adults aged 18 years or older in May–June 2025. Participants completed three shopping visits in a realistic online grocery platform. The first visit served as a within-subject no-intervention baseline. After that visit, participants were randomized in parallel to one of four conditions: climate swaps (n = 301), health swaps (n = 300), climate + health swaps (n = 300), or control (n = 300). Participants were not masked to trial arm.
Swap recommendations were presented in the online shopping task during the second and third visits. The three intervention types were: climate-focused swaps recommending items with a lower carbon footprint, health-focused swaps recommending more healthful items based on the Ofcom Nutrient Profiling Model, and combined climate + health swaps recommending items that aimed to improve both outcomes. On average participants received between 2.3 and 3.6 swap recommendations per shopping visit and accepted roughly 25%–27% of recommended swaps.
Two primary endpoints were prespecified and operationalized as follows:
Outcomes compared change from the baseline (first visit) to intervention visits (second and third visits) among participants assigned to a swaps arm versus control, analyzed using intent-to-treat methods.
Analyses used intent-to-treat comparisons of within-subject change from baseline to intervention visits across trial arms. Effect estimates are reported with 95% confidence intervals and p-values. The study assessed interactions to examine whether effects differed by age, health consciousness, or environmental consciousness.
Compared with control, all three swap interventions produced statistically significant improvements in the healthfulness of grocery selections. Reported effect estimates (change on the 1–100 Ofcom scale) were:
All p-values for these comparisons versus control were < 0.001. The health and climate + health conditions increased healthfulness more than the climate-only condition (p ≤ 0.02), while health and climate + health did not differ significantly from one another (p = 0.34).
Reductions in the selection-level carbon footprint occurred for the climate-targeted interventions but not for the health-only intervention. Compared with control, the reported changes in kg CO2-eq per kg were:
The climate and climate + health arms produced larger reductions in carbon footprint than the health-only arm (p-values < 0.001) but did not differ from one another (p = 0.43).
Participants accepted about one quarter of recommended swaps, indicating partial uptake in a naturalistic online shopping environment. The trial found no evidence that the effects of swap recommendations on healthfulness or carbon footprint varied meaningfully by age, health consciousness, or environmental consciousness; interaction p-values were all ≥ 0.10.
The authors noted two principal limitations. First, the study sample differed somewhat from the US population, which may affect external validity. Second, the naturalistic online store setting may not generalize to brick-and-mortar grocery shopping contexts. The source article did not provide additional unreported limitations beyond these points.
In this randomized online grocery trial, recommending simple substitutions to more healthful or lower-carbon foods improved the healthfulness of simulated grocery selections. Recommendations that explicitly targeted lower production emissions also reduced the carbon footprint of participants’ selections. These findings suggest that targeted swap recommendations in online retail environments could be a scalable strategy to improve diet quality and, when climate-focused, to reduce food-related greenhouse gas emissions. Further research is needed to test generalizability to in-store shopping, to other populations, and to real purchasing behavior.
The trial is registered at ClinicalTrials.gov (NCT#06648226), first submitted October 16, 2024. Individual-level selection data and analytic code are publicly available at the repositories cited in the source article.