An AI cooking assistant can turn a half-stocked fridge into a workable dinner plan. In tests with NYT Cooking, the tool suggested ingredient swaps, adjusted cooking steps and converted measurements, while leaving the original recipe intact. It was useful in the kitchen, but one cake suggestion showed why its instructions still need a human check.
A recipe adaptation that worked
For a weeknight meal, the tester began with One-Pot Chicken and Rice With Ginger, but had chicken tenders instead of thighs and lacked several listed ingredients. They entered the ingredients they did have, including green beans and red onions, into the assistant on the recipe page.
The tool produced a new version, titled One-Pot Chicken and Rice With Ginger and Green Beans. It proposed browning the tenders for less time, cooking the red onion alongside the chicken, and steaming the green beans over the rice during the last eight minutes. The adjustments appeared under a Your Version tab, with the original recipe still accessible.
When asked about replacing optional cilantro, the assistant approved Thai basil and explained that it would change the flavor. The finished dish took the same amount of time to cook as the original and was described as delicious. That result made the assistant useful for more than brainstorming: it translated substitutions into steps that could be followed without leaving the recipe.
Swaps, conversions and sensible limits
The assistant also answered questions where the cook did not already have a substitute in mind. Asked what could replace makrut lime leaf in Thai red curry, it suggested grated lime zest and noted that it would miss the deeper floral, herbal aroma of fresh limes.
Other trials covered half-and-half instead of cream in a caramel apple Dutch baby, and Castelvetrano olives and smoked paprika in place of kalamata olives and chile pepper in roasted halibut. The app converted the Dutch baby’s metric weights to US measurements, too.
Some answers included a warning about the tradeoff. Fresh cayenne peppers could bring “significant, sharp heat” if used in place of chile. And when asked whether minced beef could replace ladyfingers in a sweet dessert trifle, the assistant rejected the idea outright. Those responses suggest a useful role for the tool: explain what a swap changes, and push back when a proposed substitution does not make sense.
Why the original recipe still matters
NYT Cooking’s existing recipes come from sources such as the New York Times archives, restaurants, cookbooks and food columnists, and people have tested them. Its reader comments are another source of practical advice, including answers to questions about unusual ingredients. But searching through many comments during dinner can be inconvenient.
The assistant offers a quicker way to ask about a specific recipe. It is available to subscribers on iOS at launch, and its prompt appears on the recipe page. Users can request changes such as scaling a recipe, making it vegan and gluten-free, or finding an equipment substitute. The app had also offered measurement conversion for a couple of years.
Crucially, a customized version does not overwrite the recipe it came from. The original remains available, and changes are highlighted separately. That gives cooks a way to compare the adaptation with the source and decide whether the new instructions fit their needs.
AI assistance still needs supervision
The chicken recipe was a success, but the tool was not consistently reliable. In a cake test involving a missing Bundt pan, it nearly led the tester to use half the required butter. That kind of error matters because cooking instructions depend on proportions as well as ingredient availability.
A comparison with Alexa Plus showed another limitation. The voice assistant gave similar guidance on the chicken and rice substitutions, though cooking times differed slightly. But it could not display the modified recipe on its screen; it showed the original while speaking the changes, and became confused when asked to walk through them.
For cooks, the practical approach is to treat recipe AI as an assistant that can save time on substitutions and explain likely effects, while reviewing its directions before cooking. It may reduce the need to search comments or leave the app, but a familiar recipe and a careful read remain valuable safeguards.