The New York Times officially integrated a generative AI recipe-modification tool into its NYT Cooking app in late October 2026. Designed to assist subscribers with ingredient swaps and portion rescaling, the tool has demonstrated a notable performance gap between the flexible improvisation of savory cooking and the rigid mathematical requirements of baking.
The feature is currently restricted to iOS subscribers as part of a gradual rollout phase that began on October 23, 2026. While the assistant is capable of adapting recipes to different pan sizes and suggesting substitutions for missing ingredients, its reliance on generative logic has led to documented errors in scaling and “context loss” during multi-step modifications.
The Scaling and Math Gap
The precision required for baking chemistry appears to be a significant hurdle for the assistant. In documented tests, the AI failed basic culinary math when attempting to scale down complex ingredient lists. For example, when tasked with halving a recipe that originally called for five eggs, the assistant suggested using three eggs. While such a rounding error might be negligible in a savory stir-fry, it can fundamentally change the structural integrity of a cake or custard.
Beyond simple division errors, the tool has also struggled with maintaining consistency between original data and user-requested modifications. This “context loss” often occurs when a user interacts with the app via voice query. In one instance, the AI suggested using six tablespoons of butter for a recipe that originally required three-quarters of a cup (12 tablespoons). The error was traced back to the AI confusing the original recipe’s data with a previously generated “Your Version” modification, effectively halving the fat content without user intent.
This “UI trap” suggests that while the app creates a visual “Your Version” card to help users track changes, the underlying AI may still pull information from the original recipe data during follow-up questions, leading to conflicting instructions. This distinguishes the NYT Cooking tool from competitors like Alexa Plus; while Alexa often provides text-only improvisation that can be difficult to track, the NYT assistant’s attempt to maintain a modified recipe card can create a false sense of security if the math behind the card is flawed.
The rollout follows a broader trend of integrating AI into niche service apps, though user reception has been cautious. According to Nieman Journalism Lab, some long-time subscribers have expressed hesitation regarding the reliability of automated features in a platform traditionally built on human-tested accuracy.
As the gradual rollout continues, the tool remains a mobile-only feature for paid subscribers. The New York Times has characterized this phase as a period to monitor performance and refine the tool’s ability to handle complex kitchen variables. For now, the assistant remains most effective for low-stakes substitutions—such as swapping kale for spinach—rather than the high-precision environment of the pastry kitchen.