When you roll out chilled sugar cookie dough or top your green bean casserole with crunchy onions this holiday season, you might not consider how much AI went into the recipe — it might be more than you think.
AI involvement in food related activities is becoming increasingly common, according to Stacy Bevan, professional practice associate professor of dietetics at Utah State University. One study reported 74% of people aged 18-24 use AI-powered tools for meal planning, recipe suggestions and grocery shopping. And increasingly, AI is prompted to create the recipes themselves.
But can AI whip up a decent dinner plan? Recently published research from a team in USU’s Department of Nutrition, Dietetics and Food Science tried to find out, pitting human-authored recipes and text against an AI chef. The researchers created two sets of recipe blogs, one authored by students in a food literacy course and another with AI mimicking the students’ style. Researchers then surveyed people’s reactions to the recipe package.
Results showed a growing tolerance for AI-assisted recipe curation, with some important limits, Bevan said. Parts of the process still require a human touch — and a tongue, according to the survey.
AI-generated recipes were rated similarly to the human-written content on several metrics — perceived ease of preparation, use of common ingredients, and time requirements. There was a slight difference in the way participants rated the budget friendliness of recipes, with AI splurging on ingredients more than human authors.
But when people found out who was in the kitchen, the written comments about the experience revealed nuanced reactions. About 43% of participants said that knowing a recipe was AI-generated wouldn’t impact their willingness to try it, but there was serious pause among many based on practical and philosophical concerns.
The most prevalent was that AI wouldn’t have the ability to taste or optimize recipes, perceptions that the recipes might be a bland average of all options and questions about copyright. Many participants in the survey said they preferred human-authored content because AI felt less personal and took away from the “humanness” of working with food and serving a meal.
Many said that they could tell when the text was AI-produced. Some participants said they were fine with AI-generated content as long as the use of AI was disclosed.
“It makes sense that some AI-generated recipes turn out well as they are based on information from existing recipes,” said Katie Kraus, lead author on the research. “But using AI-generated recipes for more complicated dishes is a gamble.”
In a well-written recipe, the narrative and the technical accuracy are important, Bevan said. The narrative around the more technical parts of the process can increase confidence that there is personal experience behind it. It allows readers to know that someone has tested things out and that there is real evidence that it can turn out well, she said. We tend to want to build on other people’s real experience in the kitchen.
Then again, many cooks just jump straight to the recipe.
“My students tell me that they don’t read the narrative anymore,” Bevan said. “The quality of the recipe is more important than the writing these days. Many people, including me, rely more on how many people have reviewed the recipe and how they rate it, or read through a few of the reviews to know people’s real experience.”
The assistance of AI in the kitchen still has tremendous potential. It already does some tasks really well, Bevan said — creating a prep schedule for a big holiday meal, budgeting ingredients on a shopping list and reducing food waste by suggesting meals that use foods already on hand.
And it can be a crack search engine for a good human-written recipe, evaluating thousands and leading you to one that fits your parameters and has high ratings.
But AI can’t replace professional expertise, Bevan said. There are instances where AI has listed inappropriate foods for specific dietary restrictions like diabetic or renal diets. Or given straight up bad advice, like adding non-food items to a recipe.
“It just doesn’t have enough context,” Bevan said. “But it can look good to an inexperienced eye. It is increasingly important to train dietetic students on how to critically engage with and evaluate this kind of content.”
Experienced cooks will be able to identify errors in the recipe and adjust accordingly. This is what Bevan tells her students in her food literacy courses. Classes like this offer a solid foundation in food and nutrition, as well as basics in the kitchen that allow people to translate knowledge about nutrition into actual healthy eating, she said.
“AI is impacting many areas of our students’ lives and future professions,” Kraus said. “We can help them by paying attention to the changes and teaching them to navigate information from a variety of sources, including AI, to create great recipes, understand what they are eating and be healthy.”