NYT Cooking Introduces AI Recipe Customization Tool for App Subscribers
NYT Cooking has launched an AI assistant on iOS to help users customize recipes, adjust measurements, and substitute missing ingredients in real time.
The New York Times has introduced an AI assistant within its NYT Cooking app, offering subscribers on iOS a tool designed to customize recipes on the fly. Built to address common kitchen dilemmas like missing ingredients, dietary restrictions, or different pan sizes, the feature appears as a text prompt reading "Customize this recipe..." directly on recipe pages. Rather than modifying the publication's original tested recipes—sourced from NYT archives, cookbooks, and food columnists—the system leaves the original text untouched and generates an adapted version under a dedicated "Your Version" tab, highlighting changes in red.
How the assistant adapts recipes in practice
In real-world testing by tech journalist Jennifer Pattison Tuohy, the assistant handled complex kitchen modifications with practical accuracy. When adapting Yasmin Fahr’s "One-Pot Chicken and Rice With Ginger," the tool adjusted cooking times and steps after Tuohy substituted chicken tenders, red onions, green beans, and Thai basil for missing thighs, chard, cilantro, and scallions. The system renamed the dish to "One-Pot Chicken and Rice With Ginger and Green Beans," shortened the browning time for tenders, and instructed steaming green beans over the rice during the final eight minutes.
The assistant is trained on NYT Cooking's archive and user comments section, allowing it to provide contextual advice on ingredient swaps—such as suggesting grated lime zest when makrut lime leaves are missing while pointing out the loss of floral aroma. NYT Cooking editor Genevieve Ko noted the assistant is programmed to reject absurd or unpalatable requests. For instance, when asked to substitute minced beef for ladyfingers in a sweet dessert trifle, the AI explicitly declined.
The assistant also handles structural recipe changes, such as scaling and metric conversions. Converting Genevieve Ko’s "Cinnamon Roll Bundt Cake" into muffins yielded adjusted baking times from three hours down to 50 minutes. When asked to halve the recipe to make 12 muffins instead of 24, it logically rounded five eggs down to three rather than providing an impractical half-egg measurement. When compared against Alexa Plus on an Echo Show 11 smart display—which accesses NYT recipes through a partnership between the companies—the NYT app proved more reliable by providing visual text updates rather than relying solely on voice instructions, where Alexa became confused when explaining step-by-step alterations.
Limitations and UI quirks
Despite its utility, the assistant exhibits clear limitations that require user oversight. Because modified recipes cannot be saved separately within the app, switching between the original recipe tab and "Your Version" can lead to confusion. During testing on the bundt cake, asking the assistant for butter quantities via voice typing resulted in it quoting 6 tablespoons from the scaled-down modified version rather than the original 3/4 cup, without specifying which context it was pulling from. Deleting the modified version resolves the ambiguity, but it underscores that the tool functions as a sous chef rather than an independent head chef.
What it means for developers
For developers, the NYT Cooking AI assistant provides a practical case study in combining structured publisher data with fine-tuned conversational AI. By augmenting base model capabilities with real user comments and historical archives, the team created an assistant capable of giving nuanced advice on taste and technique rather than simple direct swaps. It highlights the importance of user interface design when displaying generative content alongside trusted source material, using dual tabs and visual diffs to maintain editorial integrity.
To build and test specialized tools like this, developers can try top AI models cheaply through one API at https://apixoai.online. Implementing similar guardrails—such as rejecting unpalatable combinations or managing complex scaling logic like egg rounding—requires iterative prompt engineering and model evaluation across various top-tier LLMs. As integrations like the NYT partnership with Amazon's Alexa Plus demonstrate, multi-platform deployment of structured recipe intelligence remains a key area of growth for domain-specific AI applications.
Source: NYT Cooking’s AI assistant saved my weeknight chicken, but almost ruined my cake — The Verge AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
