Generative artificial intelligence creates delicious, nutritious burgers

Generative artificial intelligence has learned the structure of the human palate from recipe data to create novel burgers optimized for taste, sustainability, or nutrition. In a blinded sensory evaluation with 101 participants, the AI-generated delicious burgers matched or exceeded the Big Mac in liking, flavor, and texture, while its mushroom burger achieved an environmental impact score over ten times lower and its bean burger nearly doubled the nutritional score. This establishes generative AI as a framework for principled food design that balances complex trade-offs.

Abstract Food choices shape both human and planetary health; yet, designing foods that are delicious, nutritious, and sustainable remains challenging. Here we show that generative artificial intelligence can learn the structure of the human palate directly from large-scale, human-generated recipe data to create novel foods within a structured design space. Using burgers as a model system, the generative AI rediscovers the classic Big Mac without explicit supervision and generates novel burgers optimized for deliciousness, sustainability, or nutrition. Compared to the Big Mac, its delicious burgers score the same or better in overall liking, flavor, and texture in a blinded sensory evaluation conducted in a restaurant setting with 101 participants; its mushroom burger achieves an environmental impact score more than an order of magnitude lower; and its bean burger attains nearly twice the nutritional score. Together, these results establish generative AI as a quantitative framework for learning human taste and navigating complex trade-offs in principled food design. Similar content being viewed by others Food choices rank among the most consequential decisions humans make, with far-reaching implications for both personal and planetary health 1. The global food system contributes substantially to climate change , drives land-use change and biodiversity loss 2 /articles/s41538-026-00953-x ref-CR2 , depletes freshwater resources 3 /articles/s41538-026-00953-x ref-CR3 , and pollutes terrestrial and aquatic ecosystems 4 /articles/s41538-026-00953-x ref-CR4 . Sustaining the Earth within a safe operating space for humanity requires a rapid transition toward more sustainable food systems 5 /articles/s41538-026-00953-x ref-CR5 . 6 /articles/s41538-026-00953-x ref-CR6 The same system exerts a dominant influence on human health 7. More than one billion people consume diets that fail to meet basic nutritional needs , while many more experience micronutrient deficiencies 8 /articles/s41538-026-00953-x ref-CR8 . Poor dietary patterns contribute substantially to the global burden of chronic non-communicable diseases 9 /articles/s41538-026-00953-x ref-CR9 , including type II diabetes 10 /articles/s41538-026-00953-x ref-CR10 cardiovascular disease 11 /articles/s41538-026-00953-x ref-CR11 , and certain cancers 12 /articles/s41538-026-00953-x ref-CR12 . Diet thus links environmental sustainability and human health through shared consumption patterns 13 /articles/s41538-026-00953-x ref-CR13 . 14 /articles/s41538-026-00953-x ref-CR14 Consumer acceptance, rather than availability, remains the central bottleneck in the adoption of sustainable and nutritious foods 15. Deficits in taste , texture 16 /articles/s41538-026-00953-x ref-CR16 , and cultural familiarity 17 /articles/s41538-026-00953-x ref-CR17 continue to limit uptake, even when environmental and nutritional benefits are clear 18 /articles/s41538-026-00953-x ref-CR18 . Designing foods that satisfy environmental and nutritional objectives, while meeting sensory expectations, demands a quantitative understanding of the human palate 19 /articles/s41538-026-00953-x ref-CR19 . Yet, conventional food development relies heavily on artisanal expertise and incremental trial-and-error, which constrains systematic exploration of large design spaces and slows innovation 20 /articles/s41538-026-00953-x ref-CR20 . 21 /articles/s41538-026-00953-x ref-CR21 To address this gap, we leverage generative artificial intelligence to model human food preferences as a high-dimensional probability distribution, with burgers as a model system 22. Rather than encoding rules for flavor or texture, the model learns statistical regularities directly from ingredient combinations and quantities. Unlike transformer-based large language models that generate recipes through next-token prediction in natural language 23 /articles/s41538-026-00953-x ref-CR23 , our custom-designed diffusion model learns a structured probability distribution over ingredient identities and quantities. This probabilistic formulation enables quantitative sampling, controllable novelty, recipe rediscovery, and explicit optimization with respect to environmental impact and nutritional quality, while remaining grounded in consumer-relevant designs. While large language models excel at generating culinary instructions 24 /articles/s41538-026-00953-x ref-CR24 and broad world knowledge 25 /articles/s41538-026-00953-x ref-CR25 , diffusion-based generative models provide greater control over ingredient compositions and probabilistic exploration of structured food design spaces. 26 /articles/s41538-026-00953-x ref-CR26 We test the hypothesis that generative artificial intelligence can learn the human palate and create burgers that taste the same or better than the classic Big Mac®. The Big Mac® ranks among the most widely consumed burgers worldwide 27, it is sold in more than 100 countries , and its price serves as a widely used informal indicator of purchasing power parity across currencies 28 /articles/s41538-026-00953-x ref-CR28 . This global familiarity makes the Big Mac® a stringent benchmark to evaluate whether a generative model captures collective human taste preferences. To test this premise, we evaluate the model using two discovery benchmarks: the rediscovery of the Big Mac® from statistical structure alone and the creation of novel burgers that retain high consumer acceptance. 29 /articles/s41538-026-00953-x ref-CR29 Beyond validation, we integrate environmental sustainability and nutritional quality as additional criteria to select from a large ensemble of generated recipes. We quantify environmental impact using land use, greenhouse gas emissions, eutrophication potential, and scarcity-weighted water use 30, and assess nutritional quality using established profiling frameworks, including the healthy eating index . By sampling broadly and selecting recipes that jointly satisfy palatability, sustainability, and nutrition, we show that generative artificial intelligence can identify burgers that substantially reduce environmental impact and improve nutritional quality without abandoning cultural familiarity. Sensory validation with more than 100 participants confirms that the model correctly captures human taste preferences and creates burger recipes that match or exceed the sensory appeal of the classic Big Mac®. 31 /articles/s41538-026-00953-x ref-CR31 Results Generative AI successfully creates burger recipes We first validate the generative AI model by testing whether it reproduces key statistical properties of 2216 human-designed burger recipes while generating novel combinations Fig. 1 /articles/s41538-026-00953-x Fig1 . The model architecture combines a multinomial diffusion model for ingredient selection with a score-based generative model for ingredient quantification, which together generate complete burger recipes defined by 146 ingredients and their quantities Fig. 1 /articles/s41538-026-00953-x Fig1 a . Comparison of generated samples with the training data shows close agreement across multiple marginal and higher-order statistics: In particular, the model reproduces the distributions of ingredient quantities Fig. 1 /articles/s41538-026-00953-x Fig1 b and ingredient popularity, defined as the probability of ingredient occurrence across recipes Fig. 1 /articles/s41538-026-00953-x Fig1 c , which demonstrates that it learns both how often ingredients appear and in what amounts. The model also captures higher-order structure, including strong positive and negative correlations between ingredient pairs commonly observed in real recipes Fig. 1 /articles/s41538-026-00953-x Fig1 d , and accurately matches the distribution of recipe length, measured by the total number of ingredients per burger Fig. 1 /articles/s41538-026-00953-x Fig1 e . After establishing statistical fidelity, we generate one million burger recipes and map their palatability, environmental, and nutritional scores to reveal the structure of the generated design space Fig. 1 /articles/s41538-026-00953-x Fig1 f . Recipes with high palatability scores cluster in regions associated with popular, conventional ingredient combinations with lower nutritional and medium environmental scores, whereas recipes with low palatability scores occupy regions associated with rare or unconventional combinations, consistent with human culinary preferences. Together, these results demonstrate that the generative model learns the underlying distribution of real human-designed burger recipes and enables systematic exploration of trade-offs between palatability, nutrition, and environmental impact. Generative AI rediscovers classic burgers and creates novel designs We next assess whether the generative AI model can both rediscover canonical burger recipes and generate novel, appealing alternatives Fig. 2 /articles/s41538-026-00953-x Fig2 . We quantify similarity between generated samples and a reference recipe using a substantial difference score, where SDS = 0 indicates a match in ingredients and quantities, and SDS 0 measures increasing novelty Fig. 2 /articles/s41538-026-00953-x Fig2 a . From random samples, the model successfully rediscovers the classic Big Mac®, both in correct ingredients and weights, although the Big Mac® was never part of the initial training data Fig. 2 /articles/s41538-026-00953-x Fig2 a . Across ten independent randomizations, rediscovering the Big Mac® requires on average 7.3 million samples which demonstrates that exact replication of recipes is a low-probability event under the learned distribution Fig. 2 /articles/s41538-026-00953-x Fig2 e . Beyond rediscovery, the model generates new burgers with varying degrees of novelty, illustrated by two representative recipes, the Delicious Burger 1 with SDS = 3 and the Delicious Burger 2 with SDS = 6, which exhibit progressively more distinct ingredient profiles while retaining familiar burger structure Fig. 2 /articles/s41538-026-00953-x Fig2 c, d . Sensory evaluation indicates that the two delicious burgers achieve consumer ratings that are comparable to, and in some cases exceed, those of the classic Big Mac® Figs. 2 /articles/s41538-026-00953-x Fig2 f, g : Delicious Burger 1 received significantly higher ratings than the Big Mac® for flavor 5.8 ± 1.3 vs. 5.4 ± 1.5 and Delicious Burger 2 for overall liking 5.7 ± 1.2 vs. 5.3 ± 1.5 and flavor 5.8 ± 1.3 vs. 5.4 ± 1.5 , whereas both texture ratings did not differ significantly from the Big Mac® n = 101, p < 0.05 . Participants more frequently described the Delicious Burger 1 as meaty 67% vs. 42% , moist 60% vs. 32% , and fatty 40% vs 12% than the Big Mac®, and the Delicious Burger 2 as meaty 66% vs 42% and smoky 47% vs. 4% than the Big Mac® n = 101; paired comparisons, p < 0.05 Together, these examples validate the generative AI model by demonstrating its ability to internalize canonical burger recipes and generate novel, delicious, and palatable designs. Generative AI creates sustainable burgers We next evaluate whether the generative AI model can identify and generate burger recipes with reduced environmental impact while maintaining consumer acceptance Fig. 3 /articles/s41538-026-00953-x Fig3 . We quantify sustainability using an environmental impact score that aggregates ingredient-level contributions from land use, eutrophication potential, scarcity-weighted water use, and greenhouse gas emissions Fig. 3 /articles/s41538-026-00953-x Fig3 a . Analysis of the training data shows that environmental impact scores vary substantially across recipes of different primary protein sources, with lamb- and beef-based recipes exhibiting systematically higher impacts than poultry- and mushroom-based recipes Fig. 3 /articles/s41538-026-00953-x Fig3 b . Sampling one million recipes from the model enables identification of sustainable burger candidates, illustrated by two representative examples, Sustainable Burger 1 and Sustainable Burger 2, which differ in ingredient composition and dominant protein source Fig. 3 /articles/s41538-026-00953-x Fig3 c, d . Sustainable Burger 1, a mushroom-based formulation, achieves an environmental impact score of 0.06, more than one order of magnitude lower the Big Mac® with 0.93, whereas Sustainable Burger 2, a mushroom-beef blend, with 1.02 ranks comparable to the Big Mac® Fig. 3 /articles/s41538-026-00953-x Fig3 e . Consumer feedback indicates that Sustainable Burger 1 scores modestly below the Big Mac® in overall liking, flavor, and texture, whereas Sustainable Burger 2 performs on par with the Big Mac® in these categories. Sensory evaluation indicates that the two sustainable burgers achieve consumer ratings that differ in systematic ways from those of the classic Big Mac® Fig. 3 /articles/s41538-026-00953-x Fig3 f, g : Sustainable Burger 1 received significantly lower ratings than the Big Mac® for overall liking 4.8 ± 1.8 vs. 5.3 ± 1.5 , flavor 5.0 ± 1.9 vs. 5.4 ± 1.5 , and texture 4.5 ± 1.9 vs. 5.2 ± 1.5 , whereas ratings for Sustainable Burger 2 did not differ significantly from the Big Mac® across these attributes n = 101; p < 0.05 . Participants more frequently described the Sustainable Burger 1 as earthy 63% vs. 2% , strong 37% vs. 14% , moist 53% vs. 32% , and soft 50% vs. 25% than the Big Mac®, and the Sustainable Burger 2 as smoky 55% vs. 4% , moist 49% vs. 32% , and fatty 30% vs. 12% n = 101; paired comparisons, p < 0.05 for all reported attributes . Together, these results validate that the generative AI model successfully navigates the trade-off between sustainability and palatability and discovers burgers with markedly reduced environmental impact without compromising taste. Generative AI creates nutritious and personalized burgers We next evaluate whether the generative AI model can identify burger recipes optimized for nutritional quality and adapt them to individual dietary needs Fig. 4 /articles/s41538-026-00953-x Fig4 . We quantify nutrition using the healthy eating index, which aggregates contributions from food groups to promote, fatty acid composition, and nutrients to limit Fig. 4 /articles/s41538-026-00953-x Fig4 a . Analysis of the training data reveals substantial variation in healthy eating index across recipes with different primary protein sources, with bean- and mushroom-based recipes exhibiting systematically higher nutritional scores than beef- and lamb-based recipes Fig. 4 /articles/s41538-026-00953-x Fig4 b . Sampling one million recipes enables identification of a representative Nutritious Burger with a high nutritional score, which occupies a favorable region of the nutritional-environmental design space compared to the other AI generated burgers Fig. 4 /articles/s41538-026-00953-x Fig4 c, d . The Nutritious Burger, a bean-based formulation, achieves a healthy eating index of 63.12, nearly twice as high as the Big Mac® with 33.71, while also reducing its environmental impact score by a factor of six Fig. 4 /articles/s41538-026-00953-x Fig4 c . Relative to the Big Mac®, the Nutritious Burger shows improved alignment with dietary guidelines across multiple healthy eating index components, including increased contributions from vegetables, whole grains, and plant protein, alongside reduced refined grains, sodium, and saturated fat Fig. 4 /articles/s41538-026-00953-x Fig4 e . Consumer feedback reveals a clear reduction in hedonic ratings for the Nutritious Burger relative to the Big Mac® Fig. 4 /articles/s41538-026-00953-x Fig4 f : The Nutritious Burger received significantly lower ratings than the Big Mac® for overall liking 3.8 ± 1.7 vs. 5.3 ± 1.5 , flavor 4.0 ± 1.8 vs. 5.4 ± 1.5 , and texture 3.7 ± 1.8 vs. 5.2 ± 1.5 n = 101; p < 0.05 . Participants more frequently described the Nutritious Burger as earthy 55% vs. 2% , bland 43% vs. 21% , dry 51% vs. 18% , soft 50% vs. 25% , and grainy 42% vs. 7% , and less savory 27% vs. 53% than the Big Mac® n = 101; paired comparisons, p < 0.05 for all reported attributes . Beyond population-level optimization, the AI model also generates personalized burger recipes tailored to individual nutritional requirements. We demonstrate this feature by producing personalized recipes for a highly active 15-year-old male and a moderately active 70-year-old female, which differ in ingredient composition and quantities in accordance with age- and activity-specific dietary needs Fig. 4 /articles/s41538-026-00953-x Fig4 g . Finally, a direct comparison across all six burgers highlights systematic trade-offs between ingredient count, novelty, environmental impact, and nutrition, and places the AI generated burgers in distinct regions of this multi-objective design space Fig. 4 /articles/s41538-026-00953-x Fig4 h . In this design space, Delicious Burger 1 and Delicious Burger 2 score best in overall liking, flavor, and texture, whereas Sustainable Burger 1 and Nutritious Burger score best in nutrition and environment Fig. 4 /articles/s41538-026-00953-x Fig4 i . Together, these results validate that the generative AI model can optimize for nutritional quality at both the population and individual levels while maintaining sensory acceptance. Discussion Learning the structure of food design A central result of this study is that generative AI can learn the latent structure of food design directly from large-scale, human-generated recipe data 32. Rather than reproducing superficial statistics, the model captures higher-order regularities that define burgers as a culinary class, including ingredient co-occurrence, typical recipe length, and characteristic quantity distributions. This perspective echoes broader findings from data-driven culinary science, such as the flavor network approach that uncovers principled patterns of ingredient co-occurrence across tens of thousands of recipes in global cuisines . The ability to internalize the 33 /articles/s41538-026-00953-x ref-CR33 statistical grammar of recipes reflects the high-dimensional and nonlinear nature of human taste, which emerges from complex interactions among ingredients rather than simple additive rules . Importantly, the rediscovery of classic benchmarks such as the Big Mac®–without explicit supervision–demonstrates that culturally dominant foods occupy high-probability regions of the learned design space and validates the model against widely recognized culinary reference points. This result is particularly remarkable given that the combinatorial recipe space spans 2 34 /articles/s41538-026-00953-x ref-CR34 146, i.e., more than 10 43possible ingredient combinations . Together, these findings support the view that recipes encode collective human taste knowledge accumulated over decades, and that generative models can extract this knowledge to form an 22 /articles/s41538-026-00953-x ref-CR22 interpretable and navigable design manifold . This reframes food formulation from an artisanal, trial-and-error practice into a data-driven design science and positions recipes as a natural and human-centered interface between culinary tradition and artificial intelligence. Exploring novelty without sacrificing palatability Beyond learning existing culinary structure, the generative model enables controlled exploration of novel burger designs while preserving sensory appeal. By explicitly quantifying novelty through a substantial difference score, we show that departures from canonical recipes do not inherently lead to diminished consumer acceptance. Instead, the model identifies regions of the design space where novelty and palatability coexist: It discovers burgers that differ meaningfully in ingredient composition and proportions while maintaining ratings for liking, flavor, and texture comparable to widely consumed references. This behavior reflects the high-dimensional and nonlinear nature of human taste , which emerges from complex interactions among ingredients and cannot be navigated reliably through intuition or simple empirical rules alone 20. It also aligns with systematic analysis showing that consumer acceptance of novel foods depends on intrinsic sensory properties such as taste and familiarity as well as psychological factors such as neophobia and food experience . The rediscovery of classic burgers alongside successful novel variants suggests that cultural familiarity anchors broad regions of acceptability within the culinary design space that enable innovation without alienating consumer expectations 16 /articles/s41538-026-00953-x ref-CR16 . In this context, generative AI provides a 35 /articles/s41538-026-00953-x ref-CR35 principled framework for exploring non-linear trade-offs between familiarity and novelty, and reframes culinary innovation as navigation of a structured, data-driven design manifold rather than trial-and-error experimentation. Environmental sustainability as a system-level design objective A key contribution of this work is demonstrating that generative AI can substantially reduce the environmental impact of burgers while remaining grounded in consumer-relevant designs. Livestock production is a major driver of greenhouse gas emissions, land use, and biodiversity loss 36; yet, consumer adoption of lower-impact alternatives remains limited by taste, texture, and familiarity . Our results show that sustainability emerges not from isolated ingredient substitutions, but from coordinated changes across entire recipes. This yields designs that achieve large reductions in environmental impact through 37 /articles/s41538-026-00953-x ref-CR37 system-level rebalancing of ingredients . This finding aligns with broader evidence that dietary sustainability depends on integrated food-system redesign rather than single-ingredient replacements . Notably, the mushroom-only burger achieves an environmental impact score more than an order of magnitude lower than that of a conventional reference burger, while the mushroom-blended burger retains comparable impact and sensory performance. This highlights that environmental performance depends on interactions among ingredients, rather than any single component, and that generative models are well-suited to navigating such non-linear, multi-objective trade-offs. By embedding life-cycle considerations directly into the generative process, the model identifies sustainable designs that remain recognizable as burgers and addresses a central bottleneck in alternative protein adoption– 38 /articles/s41538-026-00953-x ref-CR38 acceptance rather than availability –a challenge repeatedly emphasized in studies of consumer response to plant-based and hybrid meat alternatives . Together, these findings bridge the gap between environmental metrics and consumer-facing design, and position generative AI as a practical tool for system-level optimization of food sustainability. 39 /articles/s41538-026-00953-x ref-CR39 Nutrition and personalization as explicit design criteria An additional contribution of this work is demonstrating that nutritional quality can be treated as an explicit, quantitative design objective, rather than a secondary outcome of food formulation. By optimizing generated recipes with respect to the healthy eating index 31, the model identifies burgers that substantially outperform widely consumed benchmarks, achieving nearly twice the nutritional score of a conventional reference burger, while simultaneously reducing environmental impact by a factor of six. This result underscores that improvements in nutrition and sustainability need not be mutually exclusive, but instead emerge from coordinated, system-level adjustments across ingredient composition and quantities . Importantly, the AI-generated nutritious burger achieves this performance using recognizable whole-food ingredients and a relatively short ingredient list, which directly addresses growing 38 /articles/s41538-026-00953-x ref-CR38 concerns around ultra-processed foods and additive-heavy formulations that undermine consumer trust. Extending beyond population-level optimization, the AI model demonstrates the ability to translate dietary reference intakes into concrete, consumer-facing food designs by personalizing recipes based on age, sex, and activity level , building on earlier initiatives to personalize nutrition based on glucose levels . Together, these findings position generative AI as a practical bridge between nutritional science and everyday eating and enable objective comparison, optimization, and personalization of foods within a unified design framework. 40 /articles/s41538-026-00953-x ref-CR40 Limitations and scope This study has several limitations that define the scope of the reported results. First, the generative model learns from existing, human-designed recipes and therefore inherits cultural, regional, and temporal biases present in the source data, which primarily reflect Western-style burger traditions. Second, the recipe representation includes only ingredient identities and quantities and does not explicitly account for processing steps, cooking methods, physicochemical transformations, or water redistribution during cooking that influence texture and flavor. As a result, the current framework cannot guarantee reproducibility of specific culinary profiles without standardized preparation protocols, culinary expertise, and additional physicochemical characterization. Third, the environmental and nutritional scores rely on aggregated databases and global averages and do not reflect variability associated with specific production practices, energy systems, agricultural methods, or supply chains. Accordingly, we should interpret the reported environmental impacts as comparative estimates rather than universal or absolute values. Fourth, sensory validation focused on a limited set of generated burgers and participants, and broader studies will be required to establish generalizability across populations and contexts. Despite these constraints, the model provides a flexible and extensible framework for multi-objective food design. Broader implications and future outlook This work demonstrates how generative AI can shift food formulation from artisanal trial-and-error toward a quantitative, data-driven design science 32. By learning directly from large-scale recipe data, the AI model functions as a domain-specific world model of contemporary burger design as it captures the statistical regularities and trade-offs that shape plausible and desirable foods. This enables systematic exploration of deliciousness, nutrition, and sustainability while remaining grounded in cultural familiarity, where adoption–rather than novelty–emerges as the central challenge in food innovation . More generally, food offers a uniquely human-centered domain for generative AI, where models can align optimization objectives with sensory feedback, health, and environmental constraints 39 /articles/s41538-026-00953-x ref-CR39 . Future work should extend this framework beyond ingredient selection and ingredient quantification toward richer representations that include processing, cooking, physicochemical properties, sensory descriptors, cost, supply-chain variability, and consumer subgroups. Such extensions would connect generative recipe design more directly to related efforts in data-driven culinary science 21 /articles/s41538-026-00953-x ref-CR21 , alternative protein development 33 /articles/s41538-026-00953-x ref-CR33 , sustainable diet modeling 19 /articles/s41538-026-00953-x ref-CR19 , and personalized nutrition 38 /articles/s41538-026-00953-x ref-CR38 . Ultimately, these advances could support end-to-end food design and establish generative AI as a collaborative tool at the intersection of human creativity, engineering, and planetary health. 40 /articles/s41538-026-00953-x ref-CR40 Discussion This work shows that generative AI can reimagine food formulation as a quantitative, data-driven design process rooted in human preferences and measurable constraints. By learning directly from large-scale recipe data, the model captures the structure of burger design within the cultural and culinary scope of the training data and enables systematic exploration across taste, nutrition, and sustainability. Consumer feedback validates the generative model by confirming that its designs align with human taste preferences while delivering substantial improvements in nutritional quality and environmental impact relative to conventional benchmarks. Beyond burgers, this approach points toward a new paradigm for AI-driven food design that unites culinary creativity with human and planetary health. Methods Model architecture, training and validation Food recipes are hybrid discrete–continuous objects that require ingredient selection and ingredient quantification . Here we represent each burger recipe solely by its ingredients and their associated weights. Accordingly, a recipe x 0 = { m , w }, consists of a binary ingredient mask m ∈ {0, 1} K that indicates presence or absence of each ingredient and the ingredient weight \ {\bf{w}}\in {{\mathbb{R}}}^{K}\ . We adopt a two-stage diffusion-based framework that decouples ingredient selection from ingredient quantification: Specifically, we integrate a multinomial diffusion model to generate ingredient masks with a 41 /articles/s41538-026-00953-x ref-CR41 score-based generative model to generate ingredient weights conditional on a given mask Fig. 42 /articles/s41538-026-00953-x ref-CR42 1 /articles/s41538-026-00953-x Fig1 a, Supplementary Material S.1.2 /articles/s41538-026-00953-x MOESM1 . Diffusion-based recipe generation Conceptually, diffusion models learn burger design by progressively randomizing existing recipes into noisy ingredient combinations and then learning how to reverse this process to reconstruct realistic burgers 22. Once trained, the model starts from random noise and iteratively samples plausible ingredient combinations and quantities that resemble human-designed recipes. Formally, diffusion models define a forward stochastic process q x ∣ t x that gradually adds noise to data samples t −1 x 0to produce latent variables x that increasingly obscure the original data throughout t t = 1, . . . , T time steps . The learnable component is the 43 /articles/s41538-026-00953-x ref-CR43 reverse stochastic process p x ∣ t −1 x , which progressively removes noise and enables generation from an unstructured prior. We train the diffusion model by maximizing a variational lower bound on the data likelihood, which corresponds to the evidence lower bound, t Here P x 0 denotes the marginal likelihood of an observed burger recipe under the generative model, obtained by integrating over all latent diffusion variables, where x 0 is a human-designed burger recipe, x t is its progressively noised representation, q is the fixed forward process, p is the learned reverse denoising process, and \ {\mathbb{E}}\ is the expectation operator that denotes an average over noise realizations drawn from the forward diffusion process. Ingredient selection via multinomial diffusion We model ingredient selection using a multinomial diffusion process in which ingredient presence is treated as a categorical variable. We define the forward process as where \ {\mathcal{C}}\ denotes a categorical distribution with the parameter listed after the vertical bar ∣, β t controls the noise level at time step t , and K is the number of categories. In our application, ingredient selection is binary , K = 2, meaning an ingredient is either present or absent. The above equation reduces to a Bernoulli distribution with parameter 1 − β t x + t −1 β 1 − t x , which flips ingredient inclusion from present to absent or vice versa with a probability t −1 β and keeps it the same with a probability 1 − t β . As t t increases, the ingredient mask becomes progressively randomized, while the learned reverse process reconstructs statistically plausible ingredient combinations, which inherently capture dependencies between ingredients that commonly co-occur in burger recipes. Ingredient quantification via score-based diffusion Conditional on a given ingredient mask, we generate ingredient quantities using a score-based generative model formulated through stochastic differential equations. The forward noising process is and the reverse-time denoising process is where B t is a K -dimensional Brownian motion, \ {\tilde{B}} {t}\ is its time reversal, and f and g define the drift and diffusion coefficients . Here, rather than learning the probability density 42 /articles/s41538-026-00953-x ref-CR42 , 44 /articles/s41538-026-00953-x ref-CR44 p t x directly, we approximate the score function \ {\nabla } {{\bf{x}}}\log {p} {t} {\bf{x}} \ using a neural network to enable efficient sampling of ingredient weights consistent with observed distributions in human-designed burger recipes. Dataset and training We train our model on a curated burger dataset derived from an open-source collection of over half a million human-designed recipes from Food.com 45,46. We filter all recipes for burgers, extract ingredients, quantities, and units from free texts, and standardize and convert the data into a structured representation. The final dataset consists of 2,216 burger recipes made up of 146 ingredients Suppl. Material S.1.1 /articles/s41538-026-00953-x MOESM1 , Suppl. Tables 1 /articles/s41538-026-00953-x MOESM1 and 2 /articles/s41538-026-00953-x MOESM1 . Model validation and statistical fidelity The trained model accurately reproduces both first-order and higher-order statistical properties of the training data: The ingredient selection model estimates the marginal probability of each ingredient appearing in a random burger recipe with a maximum absolute error below 1% Fig. 1 /articles/s41538-026-00953-x Fig1 b . The ingredient quantification model predicts quantities in previously unseen recipes with a mean absolute error of 101.9 g Fig. 1 /articles/s41538-026-00953-x Fig1 c , despite the extrapolatory and highly stochastic nature of the problem. Beyond marginal statistics, the model captures higher-order structure, including pairwise ingredient correlations Fig. 1 /articles/s41538-026-00953-x Fig1 d and number of ingredients per recipe Fig. 1 /articles/s41538-026-00953-x Fig1 b, e . These properties are not explicitly enforced during training, but emerge from the learned generative process Suppl. Material S.1.3 /articles/s41538-026-00953-x MOESM1 , Suppl. Figs. 1 /articles/s41538-026-00953-x MOESM1 and 2 /articles/s41538-026-00953-x MOESM1 . Substantial difference score to quantify similarity Quantifying the proximity between recipes and grouping similar recipes is useful for various applications, for example, to quantify the novelty of an AI-generated recipe. For this purpose, we define the semi-discrete substantial difference score between two recipes r 1 and r 2, as the sum of the binary distance d i over all n ing= 146 ingredients in the database, with During rediscovery , we use the substantial difference score of zero, S D S = 0, to quantify a match between an AI-generated and a human-designed recipe. During discovery , we use values larger than zero, S D S 0, to quantify the novelty of an AI-generated recipe compared to the human-designed recipes in the training set Suppl. Material S.1.5 /articles/s41538-026-00953-x MOESM1 . Popularity score to quantify palatability Palatability refers to qualities that make a food item desirable to the human palate, such as flavor, aroma, and texture. While the human palate displays significant variations across individuals, we can still quantify the overall palatability of a food product by measuring its popularity score within the population. This even extends to patterns in food preparation, as evidenced by the popularity of some combinations of ingredients compared to others Fig. 1 /articles/s41538-026-00953-x Fig1 d . Here we propose to use popularity as a proxy for palatability. Our AI model learns the probability distribution of the human palate and assigns higher probabilities to popular recipes, patterns, and combinations. At the recipe level, more frequent repetitions effectively translate into a more palatable recipe associated with a higher popularity score. Generative AI for burgers We use the trained and validated model to rediscover the classic Big Mac® and discover five new delicious, sustainable, and delicious burgers Fig. 5 /articles/s41538-026-00953-x Fig5 . The Classic burger As a proof of concept, we use the generative model to rediscover the Big Mac®, one of the most widely consumed burgers worldwide 27, served in more than 100 countries . This global adoption reflects a high degree of palatability across diverse populations and makes the Big Mac® a stringent benchmark for evaluating whether the model captures widely shared preferences. Because the official recipe is proprietary, we approximate it by synthesizing four independent open-source recreations into a unified reference recipe 28 /articles/s41538-026-00953-x ref-CR28 Fig. 47 ref-CR47 , 48 ref-CR48 , 49 ref-CR49 , 50 /articles/s41538-026-00953-x ref-CR50 1 /articles/s41538-026-00953-x Fig1 b, Suppl. Figs. 3 /articles/s41538-026-00953-x MOESM1 and 11 /articles/s41538-026-00953-x MOESM1 . We then search for this reference recipe in randomly generated samples from the model Fig. 1 /articles/s41538-026-00953-x Fig1 e . We define rediscovery as a sample with a substantial difference score of zero, SDS = 0, relative to the reference recipe. The training dataset did not contain the reference Big Mac® recipe Supplementary Material S.1.4 /articles/s41538-026-00953-x MOESM1 . The delicious burger Next, we use our artificial intelligence to discover delicious burgers, with a pre-defined novelty score . Specifically, we adopt the substantial difference score to quantify the novelty of an AI-generated sample by comparing it to the human-designed recipes in the training set Suppl. Material S.1.5 /articles/s41538-026-00953-x MOESM1 . For the Delicious Burger 1, we draw 1 million samples, filter all samples with S D S ≥ 3, and select the most repeated sample in the this list as the most palatable recipe with the highest popularity score Fig. 2 /articles/s41538-026-00953-x Fig2 c, f, Supplementary Fig. 12 /articles/s41538-026-00953-x MOESM1 . For the Delicious Burger 2, we perform the same steps, but now with S D S ≥ 6 Fig. 2 /articles/s41538-026-00953-x Fig2 d, g, Suppl. Fig. 13 /articles/s41538-026-00953-x MOESM1 . The sustainable burger We characterize environmental sustainability using life cycle assessment data, which estimate the total environmental impact of agricultural products across production and distribution chains based on global producer surveys 51. We obtain ingredient-level data from a harmonized environmental database across n = 570 studies . Since this database does not include mushrooms, we supplement it with land-use data from the United States Department of Agriculture 36 /articles/s41538-026-00953-x ref-CR36 and freshwater eutrophication potential, scarcity-weighted water use, and greenhouse gas emissions from European mushroom production 52 /articles/s41538-026-00953-x ref-CR52 . We quantify sustainability using a single 53 /articles/s41538-026-00953-x ref-CR53 environmental impact score that averages normalized land use, aquatic eutrophication potential, scarcity-weighted water use, and greenhouse gas emissions across ingredients, weighted by their quantities Suppl. Material 30 /articles/s41538-026-00953-x ref-CR30 S.1.6 /articles/s41538-026-00953-x MOESM1 , Suppl. Table 3 /articles/s41538-026-00953-x MOESM1 , Suppl. Figs. 4 /articles/s41538-026-00953-x MOESM1 and 5 /articles/s41538-026-00953-x MOESM1 . For the Sustainable Burger 1, a plain mushroom burger with an environmental impact sore of 0.06, we draw 1 million samples, sort them by their environmental impact score, and select the most repeated recipe overall Fig. 3 /articles/s41538-026-00953-x Fig3 c,f, Suppl. Fig. 14 /articles/s41538-026-00953-x MOESM1 . For the Sustainable Burger 2, a beef-mushroom blend with an environmental impact sore of 1.02, we perform the same steps, but now select the most repeated recipe that contains both beef and mushroom Fig. 3 /articles/s41538-026-00953-x Fig3 d, g, Suppl. Fig. 15 /articles/s41538-026-00953-x MOESM1 . The nutritious burger We quantify nutritional quality using established nutritional profiling models that compare food and nutrient composition against dietary guidelines, including the healthy eating index 31, nutri-score , and health star rating 54 /articles/s41538-026-00953-x ref-CR54 . Here we use the 55 /articles/s41538-026-00953-x ref-CR55 healthy eating index developed by the U.S. Department of Agriculture to assess alignment with the Dietary Guidelines for Americans and emphasizes food-group adequacy rather than individual nutrients . We obtain food-group equivalents from the USDA Food Patterns Equivalents Database 56 /articles/s41538-026-00953-x ref-CR56 , 57 /articles/s41538-026-00953-x ref-CR57 and nutrient composition data from USDA FoodData Central 58 /articles/s41538-026-00953-x ref-CR58 , and compute the healthy eating index by aggregating ingredient-level food-group and nutrient contributions, normalized to 500 kcal servings Suppl. Material 59 /articles/s41538-026-00953-x ref-CR59 S.1.7 /articles/s41538-026-00953-x MOESM1 , Suppl. Figs. 6 /articles/s41538-026-00953-x MOESM1 and 7 /articles/s41538-026-00953-x MOESM1 . For the Nutritious Burger, a bean-based formulation with a healthy eating index of 63.12, we draw 1 million samples, sort them by their healthy eating index Fig. 4 /articles/s41538-026-00953-x Fig4 e , and select the most repeated recipe within the top 5% Fig. 4 /articles/s41538-026-00953-x Fig4 d, f, Suppl. Fig. 16 /articles/s41538-026-00953-x MOESM1 . The personalized burger We account for inter-individual variation in nutritional requirements using a personalized nutrient profiling model that tailors recipes to age, sex, body composition, and physical activity level 60. We compute a personalized nutrition score on a 0-100 scale using individual characteristics, including age, sex, body weight, height, and physical activity level. We derive nutrient-specific target ranges from dietary reference intakes and acceptable macronutrient distribution ranges , together with World Health Organization guidelines on upper intake limits for sodium, free sugars, and saturated fats 61 /articles/s41538-026-00953-x ref-CR61 , 62 /articles/s41538-026-00953-x ref-CR62 . We aggregate these targets into a single personalized nutrition score for each burger Suppl. Material 63 ref-CR63 , 64 ref-CR64 , 65 /articles/s41538-026-00953-x ref-CR65 S.1.7 /articles/s41538-026-00953-x MOESM1 . Using this framework, we generate personalized burger recipes for two representative demographic profiles, a 15-year-old, 180 cm, 80 kg active male and a 70-year-old, 170 cm, 70 kg moderately active female Fig. 4 /articles/s41538-026-00953-x Fig4 g and additional personalized recipes Supplementary Fig. 8 /articles/s41538-026-00953-x MOESM1 . Burger validation Burger preparation Our AI-generated recipes specify ingredients and quantities only, and do not include the processing or cooking steps needed to prepare the actual burgers. We therefore engage an executive chef to translate each ingredient list into standardized preparation, cooking, and assembly protocols, including ingredient handling, cutting, seasoning, cooking method, and burger assembly Suppl. Material S.3 /articles/s41538-026-00953-x MOESM1 . We then provide these finalized protocols to an independent group of chefs, who prepare the five AI-generated burgers and obtain the original Big Mac® for comparison in the sensory survey Suppl. Material S.4 /articles/s41538-026-00953-x MOESM1 . Sensory survey We conduct a blind sensory evaluation with n = 101 voluntary participants from the general population at an active restaurant in San Francisco, CA, in accordance with Stanford University Institutional Review Board guidelines Supplementary Fig. 9 /articles/s41538-026-00953-x MOESM1 . Each participant evaluates all six burgers on a 7-point Likert scale for overall liking , flavor , and texture 66, and answers check-all-that-apply questions for 12 flavor- and 15 texture -related attributes Suppl. Material S.2 /articles/s41538-026-00953-x MOESM1 , Supplementary Data /articles/s41538-026-00953-x MOESM2 . All survey responses are fully anonymized. Demographics Of the n = 101 participants, 47.5% are male, 47.5% female, 3% non-binary, and 2% prefer not to say; 22% are 18–25 years old, 26% are 26–35, 19% are 36–45, 18% are 46–55, and 16% are older than 55; 65% are omnivores and 35% are flexitarians; the highest degree of education of 4% is a high school degree, 24% college, 50% batchelor’s, 11% master’s, 8% Ph.D. or higher, and 3% trade school; 4% eat burgers every day, 20% 2–3 times per week, 31% once a week, 27% 2–3 times per month, 16% every 1–2 months, and 3% 4–5 times per year Supplementary Fig. 10 /articles/s41538-026-00953-x MOESM1 . Power and sample size We select the number of participants, n = 101, to balance feasibility with statistical power. For the Likert-scale ratings, this sample size enables detection of small-to-moderate effect sizes using two-sided Welch’s t-tests. For the binary sensory attributes, the sample size provides 80% power to detect differences of more than 20% between burgers at a significance level of p < 0.05 using paired comparisons. We did not perform an a priori power calculation; however, the sample size of n = 101 is comparable to or larger than those commonly used in consumer surveys of food products. Statistical analysis We report sensory ratings for overall liking, flavor, and texture on a 7-point Likert scale as mean ± standard deviation. We use two-sided Welch’s t-tests to compare the AI-generated burgers against the Big Mac®. We report binary flavor and texture attributes as percentage values and perform paired comparisons to assess statistical significance using two-sided binomial tests. We do not correct for multiple comparisons, as all tests were planned and hypothesis-driven. We report statistical significance as p < 0.05 Figs. 2 /articles/s41538-026-00953-x Fig2 f, g, 3 /articles/s41538-026-00953-x Fig3 f, g, 4 /articles/s41538-026-00953-x Fig4 f . Data availability The data of this study are available as Supplemental Material. The data of this study are also available at https://github.com/LivingMatterLab/AI4Food/tree/main/AI4Burgers/data https://github.com/LivingMatterLab/AI4Food/tree/main/AI4Burgers/data . Code availability The codes of this study are available at https://github.com/LivingMatterLab/AI4Food/tree/main/AI4Burgers https://github.com/LivingMatterLab/AI4Food/tree/main/AI4Burgers . The generative model to create personalized burgers is available at https://ai4burgers.com https://ai4burgers.com . References Editorial. Diets, health and the environment. Nat. Food 5 , 717 2024 .Vermeulen, S. J., Campbell, B. M. & Ingram, J. S. Climate change and food systems. Annu. Rev. Environ. Resour. 37 , 195–222 2012 .Foley, J. A. et al. Global consequences of land use. Science 309 , 570–574 2005 .Wada, Y. et al. Global depletion of groundwater resources. Geophys. Res. Lett. 37 , L20402 2010 .Xu, X. et al. Global greenhouse gas emissions from animal-based foods are twice those of plant-based foods. Nat. Food 2 , 724–732 2021 .Springmann, M. et al. Options for keeping the food system within environmental limits. Nature 562 , 519–525 2018 .Herforth, A. W., Bai, Y., Venkat, A. & Masters, W. A. The healthy diet basket is a valid global standard that highlights lack of access to healthy and sustainable diets. Nat. Food 6 , 622–631 2025 .Canton, H. In Food and Agriculture Organization of the United Nations-FAO 23 edn 297–305 Europa Publications, London, 2021 . He, P. et al. Health-environment efficiency of diets shows nonlinear trends over 1990–2011. Nat. Food 5 , 116–124 2024 .Popkin, B. M. The nutrition transition in low-income countries: an emerging crisis. Nutr. Rev. 52 , 285–298 2009 .Aune, D., Ursin, G. & Veierød, M. B. Meat consumption and the risk of type 2 diabetes: a systematic review and meta-analysis of cohort studies. Diabetologia 52 , 2277–2287 2009 .Lopez Barrera, E. & Hertel, T. Solutions to the double burden of malnutrition also generate health and environmental benefits. Nat. Food 4 , 616–624 2023 .Huang, T. et al. Cardiovascular disease mortality and cancer incidence in vegetarians: a meta-analysis and systematic review. Ann. Nutr. Metab. 60 , 233–240 2012 .Lappé, F. M. Diet for a Small Planet. Ballantine Books: New York, 1971 . Friedrich, B. Transforming a 12,000-year-old technology. Nat. Food 3 , 807–808 2022 .Laureati, M., Delgado, A. M., De Boni, A. & Sinesio, F. Determinants of consumers’ acceptance and adoption of novel food in view of more resilient and sustainable food systems in the EU: a systematic literature review. Foods 13 , 1534 2024 .St. Pierre, S. R. & Kuhl, E. Mimicking mechanics: a comparison of meat and meat analogs. Foods 13 , 3495 2024 .Mellor, C. et al. Consumer knowledge and acceptance of “algae” as a protein alternative: a UK-based qualitative study. Foods 11 , 1703 2022 .van den Bedem, S. D., Cotto, C. & Kuhl, E. Open-source benchmarking of plant-based and animal meats. Foods 15 , 2112 2026 .Spence, C. Multisensory flavor perception. Cell 161 , 24–35 2015 .Kuhl, E. AI for food: Accelerating and democratizing discovery and innovation. science of food. npj Sci. Food 9 , 82 2025 .Tac, V. & Kuhl, E. Generative AI for material design: A mechanics perspective from burgers to matter. Comp. Meth. Appl. Mech. Eng. 461 , 11971 2026 .OpenAI. GPT-4 Technical Report. arXiv https://doi.org/10.48550/arXiv.2303.08774 https://doi.org/10.48550/arXiv.2303.08774 2023 .Vaswani, A. et al. Attention is all you need. Adv. Neural Info. Proc. Syst. 30 , 5998–6008 2017 .Thomas, A. T. et al. What can large language models do for sustainable food? 42nd International Conference on Machine Learning, PMLR 267: 59377-59433 2025 . Brown, T. B. et al. Language models are few-shot learners. Adv. Neural Inf. Process. Syst. 33 , 1877–1901 2020 .Spencer, E. H., Frank, E. & McIntosh, N. F. Potential effects of the next 100 billion hamburgers sold by McDonald’s. Am. J. Preventive Med. 28 , 379–381 2005 .McDonald’s delivers strong performance worldwide; November comparable sales rise 7.4%. PR Newswire 2011 . Pakko, M. R. & Pollard, P. S. Burgernomics: a Big Mac™ guide to purchasing power parity. Fed. Reserve Bank St. Louis Rev. 85 , 9–28 2003 .Clark, M. et al. Estimating the environmental impacts of 57,000 food products. Proc. Natl. Acad. Sci. 119 , e2120584119 2022 .Krebs-Smith, S. M. et al. Update of the Healthy Eating Index: HEI-2015. J. Acad. Nutr. Dietetics 118 , 1591–1602 2018 .Datta, B. et al. Artificial intelligence for food innovation. Nat. Food . https://doi.org/10.1038/s43016-026-01380-7 https://doi.org/10.1038/s43016-026-01380-7 2026 .Ahn, Y.-Y., Ahnert, S. E., Bagrow, J. & Barabasi, A.-L. Flavor network and the principles of food pairing. Sci. Rep. 1 , 196 2011 .Barabasi, A.-L., Menichetti, G. & Loscalzo, J. The unmapped chemical complexity of our diet. Nat. Food 1 , 33–37 2020 .Torrico, D. D., Fuentes, S., Gonzalez Viejo, C., Ashman, H. & Dunshea, F. R. Cross-cultural effects of food product familiarity on sensory acceptability and non-invasive physiological responses of consumers. Food Res. Int. 115 , 439–450 2019 .Poore, J. & Nemecek, T. Reducing food’s environmental impacts through producers and consumers. Science 360 , 987–992 2018 .Godfray, H. D. J. et al. Meat consumption, health, and the environment. Science 361 , eaam5323 2018 .Willett, W. et al. Food in the anthropocene: the eat-lancet commission on healthy diets from sustainable food systems. Lancet 393 , 447–492 2019 .Bryant, C. & Barnett, J. Consumer acceptance of cultured meat: A systematic review. Meat Sci. 143 , 8–17 2020 .Zeevi, D. et al. Personalized nutrition by prediction of glycemic responses. Cell 163 , 1079–1094 2015 .Hoogeboom, E., Nielsen, D., Jaini, P., Forré, P. & Welling, M.Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P. & Vaughan, J. W. eds Argmax flows and multinomial diffusion: Learning categorical distributions. eds Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P. & Vaughan, J. W. Advances in Neural Information Processing Systems, Vol. 34, 12454–12465 2021 . Song, Y. et al. Score-based generative modeling through stochastic differential equations. International Conference on Learning Representations 2021 . Ho, J., Jain, A. & Abbeel, P.Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M. & Lin, H. eds Denoising diffusion probabilistic models. eds Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M. & Lin, H. Advances in Neural Information Processing Systems, Vol. 33, 6840–6851 2020 . Taç, V., Rausch, M. K., Bilionis, I., Sahli Costabal, F. & Tepole, A. B. Generative hyperelasticity with physics-informed probabilistic diffusion fields. Eng. Comput. 41 , 51–69 2024 .Alvin. Food.com - Recipes and Reviews. https://www.food.com https://www.food.com 2020 .Wei, A. Food.com Recipes and Interactions. Kaggle dataset, https://www.kaggle.com/datasets/shuyangli94/food-com-recipes-and-user-interactions https://www.kaggle.com/datasets/shuyangli94/food-com-recipes-and-user-interactions 2023 .Stinson, T. The best homemade Big Mac recipe. online recipe, https://thegirlonbloor.com/homemade-big-mac-recipe https://thegirlonbloor.com/homemade-big-mac-recipe 2024 .Kaloudis, T. Big Mac. online recipe, https://www.theodorakaloudis.com/recipe-development https://www.theodorakaloudis.com/recipe-development 2024 .Smith, K. Big Mac recipe. online recipe, https://www.bakingbeauty.net/copycat-big-mac https://www.bakingbeauty.net/copycat-big-mac 2024 .Dimpflmaier, R. Make your own Big Mac. online recipe, https://www.fogocharcoal.com/blogs/cook/make-your-own-big-mac https://www.fogocharcoal.com/blogs/cook/make-your-own-big-mac 2024 .Hellweg, S. & Milà I Canals, L. Emerging approaches, challenges and opportunities in life cycle assessment. Science 344 , 1109–1113 2014 .Mushrooms. Tech. Rep. ISSN: 1949–1530, National Agricultural Statistics Service NASS 2016 . Goglio, P. et al. An environmental assessment of Agaricus bisporus J.E.Lange Imbach mushroom production systems across Europe. Eur. J. Agron. 155 , 127108 2024 .Julia, C., Etilé, F. & Hercberg, S. Front-of-pack Nutri-Score labelling in France: An evidence-based policy. Lancet Public Health 3 , e164 2018 .Barrett, E. M. et al. Modifying the Health Star Rating nutrient profiling algorithm to account for ultra-processing. Nutr. Dietetics 82 , 53–63 2025 .2015-2020 Dietary Guidelines for Americans. Tech. Rep. 8th ed., US Department of Health and Human Services USDHHS and US Department of Agriculture USDA 2015 . Herforth, A. et al. A global review of food-based dietary guidelines. Adv. Nutr. 10 , 590–605 2019 .Bowman, S. A., Clemens, J. C., Friday, J. E. & Moshfegh, A. J.Food Patterns Equivalents Database 2017-2018 Methodology and User Guide U.S. Department of Agriculture, Beltsville, Maryland, 2020 . U.S. Department of Agriculture USDA . FoodData Central: Foundation Foods Agricultural Research Service, 2024 . Mainardi, F., Drewnowski, A. & Green, H. Personalized nutrient profiling of food patterns: Nestlé’s nutrition algorithm applied to dietary intakes from NHANES. Nutrients 11 , 379 2019 .Committee to Review the Dietary Reference Intakes for Sodium and Potassium, Food and Nutrition Board, Health and Medicine Division & National Academies of Sciences, Engineering, and Medicine. Dietary Reference Intakes for Sodium and Potassium National Academies Press, Washington, D.C., 2019 . Standing Committee for the Review of the Dietary Reference Intake Framework, Food and Nutrition Board, Health and Medicine Division & National Academies of Sciences, Engineering, and Medicine. Rethinking the Acceptable Macronutrient Distribution Range for the 21st Century: A Letter Report National Academies Press, Washington, D.C., 2024 . World Health Organization. WHO Global Report on Sodium Intake Reduction 1st edn World Health Organization, Geneva, 2023 .World Health Organization. Guideline: Sugars Intake for Adults and Children World Health Organization, Geneva, 2015 . World Health Organization. Saturated Fatty Acid and Trans-Fatty Acid Intake for Adults and Children: WHO Guideline World Health Organization, Geneva, 2023 . St Pierre, S. et al. The mechanical and sensory signature of plant-based and animal meat. npj Sci. Food 8 , 94 2024 . Acknowledgements We thank Executive Chef Justin Schneider for his culinary expertise in creating preparation instructions, Caroline Cotto from NECTAR at Food System Innovations for stimulating discussions, and Alice Wistar and Alex Weissman from Palate Insights for performing the customer survey. We acknowledge access to the Stanford Marlowe Computing Platform for high performance computing. This research was supported by the Schmidt Science Fellowship, in partnership with the Rhodes Trust, to Vahidullah Tac, and by the Stanford Doerr School of Sustainability Accelerator, by the Stanford Bio-X Snack Grant Program, by the Bezos Earth Fund, by the NSF CMMI grant 2320933, and by the ERC Advanced Grant 101141626 to Ellen Kuhl. Author information Authors and Affiliations Contributions V.T. conceptualized the study, curated the data, performed the analysis, designed the methodology, created the software, validated the results, visualized the results, wrote the original draft, and edited the manuscript. C.D.G. edited the manuscript. E.K. conceptualized the study, visualized the results, wrote the original draft, and edited the manuscript. All authors reviewed the manuscript. Corresponding author Ethics declarations Competing interests The authors declare no competing interests. Additional information Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information Rights and permissions Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author s and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ http://creativecommons.org/licenses/by/4.0/ . About this article Cite this article Tac, V., Gardner, C.D. & Kuhl, E. Generative artificial intelligence creates delicious, sustainable, and nutritious burgers. npj Sci Food 10 , 199 2026 . https://doi.org/10.1038/s41538-026-00953-x Received: Accepted: Published: Version of record: DOI: https://doi.org/10.1038/s41538-026-00953-x