AI
NYT Cooking's AI Recipe Modifier Actually Works, Despite Initial Backlash
After drawing fierce complaints from subscribers, the Times' new tool for customizing recipes proved surprisingly competent in testing—but raises questions about whether tinkering with expert recipes improves them.
WIRED — AI
When the New York Times Cooking section announced an AI-powered recipe customization feature in October, readers flooded the comments section with outrage. Subscribers threatened to cancel, questioned who asked for this, and demanded to disable it. After 172 almost uniformly angry remarks, the Times shut down commenting on the announcement post. Yet when put to the test, the tool proved far more thoughtful than the backlash suggested.
How the Feature Works#
The customization tool appears as a field above ingredient lists in an iOS-only beta. It lets users ask for ingredient swaps, equipment alternatives, and dietary adjustments. When you request a change—say, swapping ground beef for pork, or adapting a recipe for a loaf pan you don't own—the AI modifies the recipe and highlights all alterations in distinctive red-orange text.
Camilla Velasquez, general manager of NYT Cooking, explained that the underlying AI consults the recipe text, the Times' ingredient substitution guide, reader comments, and broader expertise. "We put a lot of effort into trying to tell you when not to do something if it's no longer in the spirit of the recipe," she said. When requests venture too far afield—like asking Eric Kim's gochujang noodles to become cookies—the system redirects to a better match, in this case his gochujang caramel cookies.
Testing in Practice#
Over several days of hands-on trials, the modifications proved both competent and thoughtfully integrated. Asking the system to substitute boneless, skinless chicken breasts for skin-on thighs in an Alison Roman chicken and dumplings recipe generated "Chicken Breast and Dumplings," complete with a disclaimer that this version was "made with AI. Not tested or endorsed by the original recipe's creators." The tool correctly flagged that breast meat cooks faster and requires less fat, automatically adjusting simmering time and adding butter to compensate.
Similar success followed with other experiments: transforming a roasted mushroom and broccoli grain bowl into a version with cauliflower and squash, adapting Yewande Komolafe's sesame waffles into a yeasted overnight version, and swapping canned tomatoes for whole in a pasta and chickpea stew. The system intelligently handled timing, proportions, and cooking sequences for each variation.
One misstep emerged when the algorithm suggested simmering shredded breast meat for 30 additional minutes while a sauce thickened—resulting in noticeably drier meat compared to what thighs would have yielded. The feature worked best for straightforward swaps and dietary needs rather than fundamental protein substitutions.

