AI enters a recipe library built on testing
The New York Times has begun testing an AI-powered recipe-modification feature in its Cooking product, adding a new automated layer to a service whose value rests heavily on carefully developed recipes. The feature is currently an iOS-only beta for subscribers and is intended to help readers swap ingredients or otherwise adjust recipes to suit their needs.
The rollout has already exposed a familiar tension around generative AI. Readers can see a practical use for a tool that helps them adapt a dish to what is available in their kitchen or to a particular dietary preference. At the same time, they may worry that an automated system will weaken the trust attached to recipes that have been tested by experienced cooks and editors.
That tension was immediate. After the Times announced the change on October 23, comments below the announcement filled with objections. Readers asked why the feature had been introduced, how to turn it off, and in some cases said they might cancel subscriptions. The comments were turned off after four days and 172 comments that were described as almost entirely angry.
A sensitive use case for generative AI
Recipe modification sounds straightforward: substitute an ingredient, change a serving size, or adapt a dish to a household constraint. But cooking is full of dependencies. An ingredient can alter texture, moisture, timing and flavor simultaneously. A small change may be reasonable in one dish and disruptive in another.
That is why the feature arrives in an especially sensitive part of the media business. High-quality recipes are not merely written instructions. They are the result of development and testing, work that can require staff, time and expense. The source material notes that both New York Times Cooking and America’s Test Kitchen have deep benches of people making cooking content, and that tested recipes are central to their brands.
An AI tool placed alongside that work therefore carries more reputational risk than a generic conversational feature. If it offers an unhelpful substitution or an adjustment that produces a poor result, a user may not separate the AI-generated suggestion from the publication whose recipe they chose. The practical question is not only whether the system can produce plausible language, but whether its changes preserve the result users expect.
Early reporting finds a more useful tool than expected
In a hands-on assessment, WIRED’s Joe Ray reported that the feature mostly worked and that he found it surprisingly good. That conclusion matters because the article began from skepticism. The writer expected to side with critics after seeing repeated attempts to put technology into the kitchen without clear benefit.
The skepticism was informed by a broader history of so-called smart kitchen products. The source cites Bluetooth-connected appliances, apps built around weak recipes, and devices such as toasters with optical sensors or touchscreens that still did not perform their basic job particularly well. The recurring failure in those examples is not the presence of technology itself. It is the mismatch between technical novelty and a genuine cooking need.
Recipe changes are a more grounded target. People routinely need to work around unavailable ingredients, personal preferences or other constraints. A feature that can help with those decisions without forcing the user into a separate search process has a clearer purpose than adding connectivity to an appliance for its own sake.
The audience reaction is part of the product story
The backlash nonetheless shows that utility does not settle every question. Cooking has a strong community dimension, and the source describes a warm, casual relationship between the Times’ food personalities and their audience. Readers comment on recipe videos, exchange techniques and bring their own habits to the material. In that setting, an AI modifier can feel less like a neutral convenience and more like a change to the character of the service.
The scale of the product makes that reaction consequential. A Times executive communications director told WIRED that the Cooking site and app were visited 514 million times in 2025, while its YouTube channel has more than 1.4 million subscribers. The company does not share Cooking subscriber numbers, but those figures underline the reach of the audience encountering the feature and the importance of audience trust.
The fact that the beta is limited to iOS also suggests that this is an early stage rather than a finished, universal rollout. The Times has previously used AI to assist with recipe scaling, according to the source. The new feature extends that assistance from quantity adjustments into more qualitative modifications, where judgment can matter more.
What the beta will need to prove
For an AI recipe modifier to earn a lasting place in a tested-recipe product, it will need to show that it is both useful and appropriately bounded. A good answer must respect the original recipe rather than present an improvised alternative with the confidence of a tested result. It also needs to make clear when a requested change is likely to alter the dish substantially.
The early assessment suggests there is potential. The strong reader response shows that potential alone will not settle the debate. This is a feature being introduced into a product where expertise, editorial labor and community trust are visible parts of the offering.
For now, the Times’ iOS beta is a small but revealing experiment. It tests whether AI can support the practical flexibility home cooks want while remaining compatible with the careful recipe development that made a service like Cooking valuable in the first place.
This article is based on reporting by Wired. Read the original article.
Originally published on wired.com
