As culinary innovation manager at Piada Italian Street Food, Dawn McClung is responsible for all the fast-casual’s menu R&D. Although she was skeptical at first to bounce ideas off an AI chat bot, ChatGPT has turned into a trusted advisor, she said.
“I don’t have anyone working under me, so I use ChatGPT as a sounding board and a brainstorming partner,” she said. “It’s never going to replace a culinary brain or hospitality, but it’s there as an aid and a gut check. But it has to be paired with a human mind.”
For a recent project in which McClung was tasked with developing Italian-style sodas for Piada, she began by researching beverage trends and consumer sentiment with ChatGPT.“I had it pull menus from bigger chains to see what flavors were trending, compare beverage prices, etc. As a smaller but growing concept, we don’t focus primarily on beverages like Starbucks and Dunkin’,” she said. She also mined trend data from market research firms and forecasters.
Once McClung had that intel, she shared about 30 beverage ideas with ChatGPT and asked “which of these make the most sense in an Italian fast casual.” She fed in some flavors gathered in her trend research and some profiles suggested by the chat bot, and watermelon, strawberry and cantaloupe rose to the top and turned out to be a good fit: “Fruits that grow together, go together,” said McClung. That gave her a starting point for recipe development.
AI’s ability to help “kickstart thoughts and ideas” is also what chef Jason Hernandez finds most worthwhile. The founder of Blade & Tine, a culinary consulting company based in Atlanta, he is sometimes asked to develop recipes with ingredients that are not as familiar. “Maybe we’re asked to tie into a global influence from Latin America, so we have to figure out how to merge that with a client’s ingredient or product,” he said. “ChatGPT is another tool in the toolbox.”
Hernandez may also be asked by commodity boards to work up ideas for a type of produce, like a fresh artichoke, creating menu items appropriate for different segments, such as a fast casual and a high-end steakhouse. “The more details you add, the better the answer, so I will ask for appetizers, entrees, pasta dishes, however it may fit, and then maybe add in a sauce that will tie it together,” he said. “It may kick back something we didn’t think about and we can see if any of those combinations work and are right for that particular menu.”
From idea to menu item
For the summer beverage lineup at Columbus, Ohio-based Piada, McClung worked with the suggestions generated by ChatGPT and prompted the bot to come up with some recipes. The recipes are never used “as-is;” she looks at them for ingredient ratios and a starting point for where she’ll take the flavor profiles. The beverages go through a lot more R&D and taste-testing before they’re finalized.
“ChatGPT is not great at tweaking recipes,” she said. “It’s not going to tell me, ‘Hey, it tastes too sweet.’ It doesn’t have taste buds … that would be a little scary.”
Piada researched trends on AI to pinpoint a flavor profile for its Summer Fresca. | Photo courtesy of Piada Italian Street Food
Piada’s Summer Fresca, launching June 2, features strawberry, watermelon, cantaloupe and a touch of lemon. To turn the seasonal sipper into a dirty soda, guests are guided to add Italian cream, a coconut-based product that the chain sourced for this LTO.
McClung also uses ChatGPT to help with menu descriptions that her marketing team can refine. “I can put 900 things in menu description, so I’ll also ask for help in simplifying,” she said. “It makes my descriptions feel cleaner and more cohesive.”
Both she and Hernandez also find AI image generators to be helpful during the R&D process. The bot can create a photo of what a recipe or dish may look like or come up with several different plate presentations, beverage garnishes, etc.
Fine dining pushes the boundaries
Award-winning avant garde chef Grant Achatz of Next restaurant in Chicago has taken AI into new frontiers. For a recent four-month menu series, he used artificial intelligence to create fictional chef personas—complete with detailed backgrounds and culinary training—who each “contributed” a course. One AI-generated chef, “Jill,” was a 33-year-old from Wisconsin who supposedly apprenticed with living masters as well as Auguste Escoffier, who died in 1935. Achatz uses AI prompts to generate recipes in these chefs' imagined styles.
At One by Spork, an upscale 16-seat tasting counter restaurant in Pittsburgh, Executive Chef Christian Frangiadis is also exploring new frontiers. Although he also uses AI to nail down ratios, times and temperatures in recipe development, he gets into more complex gastronomical challenges.
“We have a food lab in the restaurant, where we make vinegars, gels, miso, molasses, espumas and other ‘sciencey’ things,” he said. “We were searching for a better way to create beet molasses, for example. I had developed a recipe where I'm juicing the beets and adding some honey and a touch of lemon and running it through a roto evaporator. So I asked AI, ‘What would happen if I added koji [a fermentation starter] to it?’”
AI helped guide the R&D behind squash espuma at One by Spork. | Photo courtesy of One By Spork Facebook
The chat bot came back with the answer, “That would be perfect; you probably wouldn’t need any additional sweetener,” Frangiadis recalls. He went with it and it was a winner. “I bounce questions off AI like that, and it will lead me in a certain direction, but it’s necessary to have some basic knowledge of what I’m asking,” he added.
The chef dug even deeper into technical information with that challenge, asking the chat bot if he should use cooked or raw beet juice (it didn’t matter!) and whether he should put the distillate back into the molasses (only the first half; the second half would leave an earthy taste, according to the bot). The chat bot also suggested how fast Frangiadis should run the evaporator, which turned out to be spot on.
“Most times it’s very accurate when I ask about menu development, but at other times, AI can be inaccurate,” he cautioned. Writing a HAACP plan wasn’t its strong suit. “But it does help me get to a result more quickly with less experimentation in the lab,” Frangiadis added. Ultimately, the chef’s experience and expertise is key to success.
The restaurant purchased the paid version of ChatGPT, and the more Frangiadis and the team use it, the better its answers are as it gets to know One by Spork’s cooking style, pantry, equipment and menu.
Engineering the menu
When we asked Elysia, the proprietary chat bot developed for our parent company, Informa, about using AI in menu development, we learned that foodservice contract company Sodexo is on the leading edge.
Sodexo reports reducing recipe development time from three days to half a day using AI for seasonal planning and trend analysis. The technology analyzes their brand data—menu reports, ingredient lists, sales patterns—to suggest dishes with higher success probability. Their culinary team then focuses on actual preparation, testing and optimization.
For Sodexo, AI-powered menu engineering is delivering 10% to 15% profit increases by identifying which items to feature, reprice or remove. The technology models ingredient price changes instantly, allowing chefs to adjust before margins slip rather than reacting after the damage is done.
The chefs featured in this article agree that they’re just beginning to tap AI’s potential for menu development, supply chain management, pricing and countless other tasks. But while AI can generate the framework, human creativity, sensory expertise and the “culinary brain” remain irreplaceable. As one chef noted in our research, “AI can spit out a recipe in the style of José Andrés, but it doesn't taste, feel or sense ingredients, so it will never be great on its own.”
Members help make our journalism possible. Become a Restaurant Business member today and unlock exclusive benefits, including unlimited access to all of our content. Sign up here.