Artificial Intelligence is Developing Products in the Kitchen: What Does Stanford's BurgerAI Experiment Tell Us About Restaurant R&D?

The development of new products in a restaurant often progresses through chef experience, repeated kitchen trials, tastings, and recipe adjustments. BurgerAI, developed at Stanford University, adds a different tool to the first stage of this process: AI can create options that meet specific goals by learning the relationships among thousands of existing recipes, even before entering the kitchen.

The study conducted by Vahidullah Tac, Christopher Gardner, and Ellen Kuhl was published in the peer-reviewed npj Science of Food journal in June 2026. The researchers trained the generative AI system they named BurgerAI on 2,216 burger recipes and 146 different ingredients found in those recipes. The data source consists of recipes created by users on Food.com.

This detail is important. BurgerAI learns not from a controlled database of professional restaurant recipes but largely from home-cooked recipes. Therefore, the results of the research should be viewed not as ready-made recipes that can be directly applied to a restaurant's menu but as a study demonstrating how AI can be used in product design.

The system learns relationships between ingredients rather than writing recipes

BurgerAI uses a diffusion-based generative AI approach instead of a large language model. The system learns which ingredients are used together in recipes, the relationships of quantities, and composition patterns to create new combinations.

The researchers first tested whether the system could rediscover a known product that it had not seen directly before.

For this, the Big Mac example was used. Since McDonald's original recipe is not publicly available, the research team created a reference recipe from four different publicly available similar recipes. After generating an average of about 7.3 million possible recipes, BurgerAI was able to create a burger that resembled this reference recipe in terms of ingredients and proportions.

This result does not mean "AI solved the Big Mac recipe." The system's ability to reach a statistically similar composition to the representative recipe created by the researchers shows that it learned the fundamental product structure in burger recipes.

Five burgers entered the kitchen from a million recipes

In the second phase of the research, BurgerAI was asked to produce one million new burger recipes based on different goals such as taste, nutrition, and sustainability.

The researchers selected five representative recipes from this pool. However, human intervention was critical at this point: AI was only generating the ingredient list and quantities; it was not describing the cooking technique.

Therefore, an executive chef transformed the combinations produced by AI into viable standard recipes. Then, the professional kitchen team prepared the burgers.

This detail may be the most important result of the research from the perspective of the restaurant industry.

AI did not develop the product alone. It generated possibilities; the chef turned this numerical design into an edible product.

101 people participated in the blind tasting

Five AI-designed burgers were compared to a real Big Mac in a blind tasting conducted in a restaurant setting in San Francisco. According to the peer-reviewed research, 101 people participated in the tasting. Participants evaluated the burgers based on criteria such as overall preference, taste, and texture.

The results showed that two burgers created by AI, focusing on taste, achieved the same or higher results than the Big Mac in overall preference and taste evaluations. In the tasting notes reported by Food & Wine, one of these burgers was described as meatier, juicier, and denser; the other was described as meatier and smokier.

Therefore, it is not possible to conclude that "AI made the best burger in the world."

The research involved only 101 participants, specific recipes, and a single comparison product. The findings indicate that some products developed by BurgerAI were liked as much as or more than the Big Mac under these experimental conditions. The entire research positions BurgerAI as a proof of concept for product development.

Taste, health, and sustainability are not easily solved simultaneously

The most interesting result of the BurgerAI experiment in terms of restaurant product development is that AI does not produce perfect products for every goal.

The mushroom-based burger, prioritized for sustainability, had a much lower environmental impact score compared to the Big Mac. However, it was rated lower in taste and texture in the tasting results.

The bean-based burger, focused on nutritional value, reached nearly twice the Healthy Eating Index score of the Big Mac and reduced its environmental impact. In contrast, participants rated it lower in taste and texture.

Another recipe that combined mushrooms and beef yielded a more balanced result. While the environmental impact decreased, taste and overall preference remained close to the Big Mac.

These results also highlight the fundamental problem of developing sustainable products in restaurants.

It is not always possible to reduce a recipe's carbon footprint, increase its nutritional value, and simultaneously maintain the taste and texture that customers love. The potential of AI lies not in automatically solving one of these goals but in its ability to scan the trade-offs between them more quickly.

Where could it be used in restaurant R&D?

BurgerAI is not a completed product development system for commercial restaurants today. The tool made publicly available by the researchers clearly states that AI can make mistakes and may even overlook fundamental ingredients. It is particularly emphasized that users need to check the generated recipes before applying them.

However, the approach shows an important direction for restaurant product development processes.

If a restaurant group can use its recipes, ingredient costs, nutritional values, customer preferences, and sustainability data in a structured way, AI can pre-screen possible products according to different goals.

For example, the business could explore answers to questions like:

What types of recipes can be produced under a certain cost?

How can the texture that customers are accustomed to be preserved while reducing the amount of meat?

What new menu options can be developed from the same product family?

What ingredient combinations can be used when certain allergens are removed?

How can the fundamental taste structure of the product be preserved while increasing its nutritional value?

These still require evaluation by chefs, food engineers, and product development teams today. AI could be a helpful tool that can reduce the number of possibilities that need to be tested.

If cost data is added, the business side could become stronger

The focus of the Stanford study is on taste, nutrition, and sustainability. However, there is another variable for commercial restaurants: cost.

A recipe may be delicious, nutritious, and environmentally better; however, if the raw material cost does not align with the targeted selling price, it becomes difficult to enter the menu.

Similarly, the preparation time of the product, the need for kitchen equipment, employee skills, shelf life, and service speed also affect the restaurant's decision.

Therefore, if AI-supported product development is to be genuinely used in restaurants, business data such as product cost, labor, preparation time, waste, and supply continuity must also be included in the model alongside taste data.

The larger area that the Stanford research could open up for the restaurant industry may be precisely here: evaluating a new product not only with the question "does it taste good?" but also with multiple business objectives at the same time.

The chef's role may change rather than diminish

The role of the kitchen team directly affects the outcome of the system in the BurgerAI experiment.

AI selects the ingredients but does not decide how they will be processed. Executive chef Justin Schneider transforms the numerically generated recipes into products that can be prepared in a restaurant environment. Stanford also clearly states that the intervention of professional chefs turns AI design into a physical and consumable product.

Therefore, the study does not support the conclusion that "AI will replace chefs."

A more realistic possibility is the expansion of the research area in front of the chef.

A chef cannot individually test hundreds of ingredients and millions of ratios. AI can scan this area and suggest candidates that fit specific goals; the chef can then make selections based on taste, technical feasibility, serving style, and brand identity.

In other words, AI can be used more as a product development assistant rather than a decision-maker in the kitchen.

Which businesses in Turkey could benefit?

It cannot be said that systems like BurgerAI are necessary for every independent restaurant in the short term.

In a single-location business that develops a small number of products, the traditional trial process of the chef may be faster and more economical.

The real potential may arise in restaurant chains with centralized product development teams, hotel groups, ready-to-eat food producers, and businesses where many recipes are standardized.

For example, for a restaurant group developing a new sandwich to be sold in dozens of locations, even a few grams of component changes in the product can create significant differences in total raw material consumption.

At this scale, AI's ability to generate alternatives between cost, nutrition, supply, and customer preference can turn into real business value.

Stafflink's perspective

The value of the BurgerAI research for the restaurant industry is greater than the debate over whether AI surpassed the Big Mac.

The real change is that the product development process can increasingly rely on data.

The intuition and experience of chefs will continue to be one of the fundamental elements of the kitchen. However, in large-scale businesses, recipe development is a process that is too costly to leave solely to intuition.

AI can highlight a few viable options from millions of possibilities here. The human team's task will be to determine whether these options are truly delicious, producible, safe, and profitable from a business perspective.

In our view, the next stage to follow in restaurant technology may be for AI to enter real product decisions in the kitchen rather than just writing text for customers.

BurgerAI is still an early and experimental example of this. However, when product development, cost management, and sustainability data begin to converge in the same systems, the way restaurant R&D operates could change significantly.