NYT Cooking AI has entered a place where mistakes do not merely look embarrassing on screen: the kitchen. The New York Times is testing a feature that lets subscribers ask for changes to one of its recipes, including alternative ingredients, different equipment and a new number of servings. A cook can keep the original version alongside the altered instructions. What comes out of the chat, however, is not a newly tested recipe from the publication’s food team.
That difference is the heart of this story. The recipes that made NYT Cooking popular are edited and tested; the personalised versions are generated for an individual request. The publisher’s publisher’s own instructions for the beta make this explicit, saying that the customised instructions have not been cooked, checked or approved by the original recipe creators. It is an unusually candid warning for a feature designed to make a trusted product more convenient.
What NYT Cooking AI can change
The most appealing use is the simplest one. You want to make a dish, but the cupboard contains something slightly different from the listed ingredient. Instead of searching for a separate recipe, you can ask whether that substitution works. The system can also adapt a recipe to the oven, pan or other equipment available and change quantities for a larger or smaller table. These are everyday problems rather than glamorous demonstrations of artificial intelligence.
The feature is not presented as a machine that invents an unlimited catalogue of new dishes. According to the publisher’s guidance, requests that move too far away from the original may be redirected towards a more suitable existing recipe. That boundary is sensible. An instruction to replace one leafy vegetable with another is a different proposition from turning a cake into a savoury casserole. The further the change, the less the original testing can tell you about the outcome.
You can compare the original and the changed recipe
The interface gives readers a way to check what AI altered. After making a request in the chat box above the ingredients, a user can open a personal version. The revised portions are highlighted, while a switch allows them to return to the untouched recipe. The feature also offers a history of previous customised versions. This is more useful than simply replacing the original text and hoping the reader remembers what changed.
Those safeguards are about transparency, not independent verification. A visible edit can still be a poor culinary recommendation. Replacing an ingredient might affect moisture, timing, texture or the chemical reaction that makes a dough rise. Changing a pan can alter heat distribution. The interface makes experimentation easier, but it cannot guarantee that the resulting food will taste good. Readers familiar with cooking know that a plausible instruction is not the same as a successful meal.
A trusted recipe and an AI suggestion are not equivalent
A recipe’s credibility comes partly from its development. Someone cooks it, identifies ambiguous steps and finds out whether the results match the promised dish. The AI system works differently. It proposes a variation from the instructions and context available to it. That can be helpful for ordinary adjustments, but the publication has not performed the same practical check on each individual’s altered version.
This matters most when a change touches safety rather than flavour. The UK’s Food Standards Agency’s cooking guidance explains why cooking time and temperature can make the difference between safe and unsafe food. A sensible user should retain authoritative food-handling guidance even when an AI suggests a different ingredient or piece of equipment. LiveAIWire’s earlier examination of the limits of AI food-safety advice raised the same underlying issue: fluent recipe advice can sound convincing without having been physically tested.
What the trial means for readers
The feature is being offered as a beta, so its current form should not be mistaken for an unchangeable editorial policy. Its role is to let subscribers adapt established recipes for immediate circumstances while keeping the professionally prepared original available. The company also warns users not to enter personal information they would not want stored or used to improve the system. This is worth considering when a dietary preference or health-related restriction is part of the prompt.
The product raises a question that extends beyond one cooking app. When a publisher with a reputation for meticulous human work adds a generator beside that work, the distinction between original and personalised material needs to remain unmistakable. The answer is not to treat every AI suggestion as dangerous. It is to know whether a particular instruction has been tested and to use the right kind of evidence for the decision. The integrity of underlying data also matters, as why the information behind food guidance matters has explored.
The useful place for AI is beside the recipe
A good adaptation tool would help a cook make small, explicit changes, explain the trade-off and leave the verified starting point intact. It would be less useful if it confidently rewrote delicate steps without explaining the uncertainty. In practice, asking why an ingredient can be substituted may be more valuable than accepting the first replacement automatically. The original recipe remains a valuable point of reference.
There is an appealing human idea behind this experiment: a recipe should work in the kitchen someone actually has. AI may make that easier, especially for substitutions and equipment questions. Whether subscribers embrace the feature will depend less on its fluency than on its results in a real saucepan. The final verdict still belongs to the person cooking, not the chatbot beside the ingredients list.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.