Sophia Puznowski; Rochester Institute of Technology
Cite as: Puznowski, Sophia. 2026. “AI Versus Homemade: A Take on Colombian Empanadas”. Food-Fueled. doi:10.57912/32180610.
Web address: https://edspace.american.edu/foodfueled/issues/volume-iii/ai-versus-homemade-a-take-on-colombian-empanadas/
Please click here to download the piece as a PDF. The text is also listed below.
Introduction
Artificial intelligence (AI) has become widely used in a variety of fields, with one in particular being the culinary world. According to WifiTalents, 76% of restaurants in the US have integrated AI use in at least one area of interest and 90% of the “Top 100” global restaurants use a cloud-based AI infrastructure (Gustafsson 2026, 1). Whether or not AI use should be utilized in recipe creation processes is a question that varies in opinions. After reading five different peer-reviewed articles, it was almost an even split of whether real-world users, such as chefs and other culinary experts, would approve of the use of AI in their kitchens, let alone recipe creation.
A majority of the articles argued a lack of trust, authenticity, and/or overall food safety risks in using AI. One article written from the perspective of a cottage food producer, “Are Recipes written by AI Safe?”. She gave an example of how Meta AI mixed up the recipe for a shelf stable pickle versus a refrigerator pickle, which could ultimately result in any users increasing their risk of botulism. (Johnston 2024, 1). Another article, “Would You Trust an AI Chef? Examining What People Think When AI Becomes Creative with Food,” did a deep dive into the trustworthiness of AI use in generating recipes/instructions with a Negroni cocktail and a traditional lasagna. Ultimately, the test subjects concluded that trustworthiness becomes an issue when adhering to the steps and ingredients, not necessarily the ability to come up with a recipe. The authors wrote, “This finding suggests that the critical factor influencing trust is the congruence of the recipes steps and ingredients with the users’ previous culinary experiences, rather than the complexity of executing the recipe itself” (Califano 2024, 3).
In a third study entitled “Artificial intelligence in the kitchen: can humans be replaced in recipe creation and food production?” the researchers questioned if the use of AI in the kitchen would be possible by pursuing three separate studies. In their conclusion, studies 1 and 2 discovered that humans were not in favor of ordering their meals if AI was used in creating the recipe, with study 1 using a hypothetical restaurant and study 2 using a real one. Study 3 discussed any negative of using AI in the kitchen and found that customers were less likely to display a negative willingness to order food when AI with uniqueness consideration cues is used in the recipe creation process than in the traditional restaurant scenario where humans perform two focal tasks (Kim 2025). In conclusion, they suggested that AI cannot be easily automated, especially in the creative process of recipe creation.
On the other hand, there were two articles that posed positive possibilities for the use of AI. In a study named “Culinary Innovation: Will the Future of Chefs’ Creativity be Shaped by AI Technologies?” an experiment was conducted tasking AI to generate 3 recipes using three ingredients with a panel of judges to rank them on a scale of 1-9. These rankings considered aroma, taste, and appearance. Although ‘Table 7’ showed that ChatGPT could not consider knowledge of cultural history, knowledge of healthy products, and ability to make products safe and hygienic (Şener 2024, 348). All three of these categories should always be considered when creating a recipe or following one. In the context of making empanadas, knowledge of cultural history is extremely important. As mentioned by Raices’s Colombiana Restaurant, “Colombian empanadas vary widely by region, each reflecting local traditions, ingredients, and cooking methods that showcase the country’s rich culinary identity” (Raices 2025).
Therefore, if ChatGPT cannot consider these aspects, as Şener’s study lacked, it would heavily influence the outcome of the prompt I ask. In the fifth article, “An Economic Perspective on the Implementation of Artificial Intelligence in the Restaurant Sector,” these two researchers weighed on the impacts of artificial intelligence in the gastronomic business. They included chefs, entrepreneurs, and gastronomic experts to give their opinions, specifically in Spain. They believe that a depletion of food resources, special nutritional needs, and culinary innovations are all affecting gastronomy, and AI could be a solution to this issue. In the end of their surveys, they found that respondents indicated AI would reduce food waste and improve the customer experience in restaurants by optimizing and recommending menus. On the contrary, weaknesses include the lack of a large amount of data in gastronomy that AI needs to be effective and provide reliable results, the lack of human intuition in creating recipes, and the cost of implementing AI in restaurant businesses (Millán 2024, 19). Ultimately, these two articles posed positives for AI use in the restaurant industry, but still showed there are gray areas in the recipe creation processes.
Personal Experiment:
I wanted to evaluate the result of letting AI write a recipe and compare both a traditional version versus the AI one. Ultimately, I chose a recipe I believed could challenge AI in authenticity and cultural importance: a traditional beef and potato Colombian empanada. Colombia is the home country to my mother and brother, and is part of who I am. Not only culturally, but the food aspect is a huge part of my life and how it influences my personal styles of cooking. Therefore, I compared an AI-generated recipe with a family-known recipe by having family and friends blind taste both dishes to demonstrate AI’s capability to replicate the empanada. Prior to this experiment, I contacted both my grandma and great grandma in Colombia to locate the exact recipe I would use for the comparison.
What defines a traditional Colombian empanada can vary upon regions within the country. In the article “Regional Variations of Traditional Colombian Empanadas Explained” (Raices Restaurant, 2025), Raices Colombiana Restaurant, an authentic Colombian restaurant located in NYC, explains that the dough and fillings for empanadas can vary. The variances range from having an egg, fish, including cheese, and/or more. One important takeaway is the “Use of local spices like achiote, coconut, or mild peppers” (Raices Colombianas Restaurant, 2025). Spices are what enhance and bring uniqueness to each regional empanada. In a recipe published online by a native Colombian Erica Dinho, her spices included Sazón, salt, pepper, and cilantro and aromatics such as onions, tomatoes, and garlic (Dihno 2025). Delish Dlites also published a similar recipe, with minor exceptions to the type of beef (shredded stew beef) used (Delish Dlites, 2025). My family’s recipe aligned quite similarly with techniques, fillings, and spices. In the Raices’ blog post, the recipe I will be using would be considered “Antioqueña,” originating from the Paisa region of Colombia.
Upon asking ChatGPT for a recipe, I stated, “give me a traditional Colombian empanada recipe utilizing ground beef and potato as your filling” to align closely with the recipe I would be using from my family. Upon reading the AI-generated recipe, the biggest difference I noticed was the lack of spices and aromatics, as shown in Table 1, particularly the inclusion of sazón, cilantro, and garlic. The predominant spice throughout the dough and filling, aside from salt and pepper, was cumin. In Table 2, the assembly of the empanadas was almost identical to each other, with small variances within the AI recipe that required mashing the potatoes and including a cup of broth in the filling. During the cooking process of the AI generated recipe, the only step I omitted was the inclusion of the one cup of broth to the filling as mine was already loose without that liquid inclusion.
Table 1: Comparison of traditional and AI-generated recipe ingredients
| Ingredients and amounts | Traditional | AI-Generated (ChatGPT) |
| Dough: | 2 cups of masarepa | 2 cups of yellow cornmeal (masarepa) |
| 2.5 cups of warm milk | 2.5 cups of warm water | |
| 1 teaspoon salt | 1 tablespoon vegetable oil | |
| 1 packet of sazón (specifically with cilantro and achiote) | ½ teaspoon salt | |
| ½ teaspoon cumin | ||
| Filling: | 1 pound of ground beef | ½ pound ground beef |
| 3 medium potatoes | 2 medium potatoes | |
| 2 scallions or ¾ onion chopped | ½ small onion chopped | |
| 2 cloves of garlic chopped | 2 cloves garlic chopped | |
| 2 tomatoes chopped | 1 small tomato chopped | |
| Small bunch of cilantro, chopped | ½ teaspoon of Salt, Pepper, Cumin | |
| 1 packet of sazón | 1 cup water or beef broth | |
| Salt, Pepper, Cumin, Garlic Powder (to taste) |
Table 2: Comparison of Instructions
| Traditional | AI Generated (ChatGPT) |
| Filling: Dice the potatoes and boil them until tender, drain them and set aside. In a large pan, add the diced tomatoes, onions or scallions, cilantro, and garlic and add to a pan with oil. Add salt, pepper, sazón packet, and cumin. Sauté for a few minutes. Add the ground beef and cook until all incorporated. Add the diced potatoes. Prepare the dough: In a large bowl, combine the masarepa, salt, and sazón packet. Warm up the milk separately in the microwave. Add the warm milk slowly until the dough is dense enough but not super sticky. Assemble the empanadas: Take a sandwich baggy and cut the sides open. Place a small ball of dough on one side of the sandwich baggy and cover it with the other side of the bag and flatten with your hands or rolling pin. Add about 1-1.5 tablespoons of filling to the center. Fold the dough in half to form a half circle and use a mug to cut through to secure the edges. | Filling: Boil the diced potatoes until just tender (about 8–10 minutes). Drain and mash lightly. In a large skillet, heat oil over medium heat. Add onion, garlic, and tomato. Sauté 3–4 minutes until softened. Add the ground beef, cumin, salt, and pepper. Cook until browned. Stir in the mashed potatoes and 1 cup broth; cook until the mixture thickens slightly and holds together. Let cool completely. Prepare the dough: In a large bowl, mix masarepa, salt, cumin, and warm water. Add oil and knead until smooth and pliable (not sticky). Cover with a damp cloth and let rest for 5–10 minutes. Assemble the empanadas: Place a small ball of dough (about golf-ball size) between two pieces of plastic or parchment paper. Flatten into a 4–5 inch circle using your hands or a rolling pin. Add 1 tablespoon of filling to the center. Fold the dough over to form a half-moon and press edges firmly to seal. Trim with a small bowl or cutter if needed. |
*The instructions for frying were left out as they were identical to each other and did not have a notable variance.
Conclusion:
After the completion of cooking the empanadas, I had a small handful of family and friends blind taste both empanadas and tasked them to make notes about differences in taste, presentation, and more. Overall, a majority of the panel agreed that the AI-generated recipe lacked flavor in the dough and the filling compared to the traditional one. As seen in Table 1, it is predicted this would happen as the absence of sazón, cilantro, and garlic logically affect the result in taste. As far as looks, Figure 1 demonstrates the color variances for the dough; this is also due to the absence of sazón resulting in a more yellow appearance. The panel of friends and family ultimately concluded that the AI recipe could not replicate the traditional empanada recipe, specifically that of a Antioqueña or Paisa region empanada. ChatGPT’s recipe lacked traditional spices and therefore resulted in a bland empanada, therefore the lack of these traditional spices can lead to an assumption that ChatGPT did not follow a traditional recipe.
AI could be used for other aspects to a restaurant business, such as preparing menus or having an AI robot prep the ingredients as Millán’s article states “As opportunities, respondents indicated that it would reduce food waste and improve the customer experience in restaurants by optimizing and recommending menus” (Millán 2024, 19). Bt the lack of cultural adherence to traditional spices and flavors, as demonstrated by the AI, will result in a poor outcome if a user is trying to create something as authentic as possible. Perhaps telling ChatGPT, or other similar AI programs, to include traditional spices and aromatics will influence the outcome, but as of now it seems AI is not yet programed to create culturally traditional recipes to an exact replica without that specification.
Photos:
Figure 1:
Photo Comparison of the AI versus Traditional Empanada outcomes for filling and the corn dough shell. AI results on the left side, Traditional on the right.
References
Califano, Giovanbattista. “Would You Trust an AI Chef? Examining What People Think When AI Becomes Creative with Food.” Science Direct, September 2024. https://doi.org/10.1016/j.ijgfs.2024.100973.
Dinho, Erica. “Colombian Empanadas (Empanadas Colombianas).” My Colombian Recipes, March 29, 2025. https://www.mycolombianrecipes.com/colombian-empanadas-empanadas-colombianas/.
Gustafsson, Linnea. “AI Restaurant Industry: Data Reports 2026.” WifiTalents, February 12, 2026. https://wifitalents.com/ai-restaurant-industry-statistics/.
Jannese. “Colombian Empanadas.” Delish D’Lites, May 16, 2020. https://www.delishdlites.com/course/appetizer-recipes/colombian-empanadas-recipe/.
Johnston, Amy. “Are Recipes Written by Ai Safe?” UMN Extension, July 2, 2024. https://extension.umn.edu/cottage-food-safety-news/are-recipes-written-ai-safe.
Kim, Hyunsu, Sungwoo Choi, and Hyejo Hailey Shin. “Artificial Intelligence in the Kitchen: Can Humans Be Replaced in Recipe Creation and Food Production? | International Journal of Contemporary Hospitality Management | Emerald Publishing.” Research Gate, February 2025. https://www.emerald.com/ijchm/article-abstract/37/5/1641/1246747/Artificial-intelligence-in-the-kitchen-can-humans?redirectedFrom=fulltext.
Millán, Genoveva Dancausa, and Genoveva Millán Vázquez de la Torre. “An Economic Perspective on the Implementation of Artificial Intelligence in the Restaurant Sector.” MDPI, September 2024. https://www.mdpi.com/2076-3387/14/9/214.
Raices Colombianas Restaurant. “Regional Variations of Traditional Colombian Empanadas.” Raices Colombianas Blog, August 21, 2025. https://raicescolombianasrestaurant.com/blog/regional-variations-of-traditional-colombian-empanadas-explained/.
Ulu, Efe Kaan. “Culinary Innovation: Will the Future of Chefs’ Creativity Be Shaped by AI Technologies?” Research Gate, May 2024. https://www.researchgate.net/publication/380538245_Culinary_Innovation_Will_the_Future_of_Chefs’_Creativity_be_Shaped_by_AI_Technologies.