The New York Times is testing NYT Cooking AI recipe tools that let subscribers modify its recipes on demand, and the reception has been anything but warm. The NYT Cooking AI feature, announced last Friday by Cooking senior editor Genevieve Ko, can recommend ingredient and equipment substitutions, scale any recipe to any number of servings, and, in Ko’s words, “generally make our recipes work for what a cook needs at that moment.” Some users aren’t exactly thrilled.
Within hours, the announcement’s comment section filled with subscribers calling the change a betrayal of the publication that built one of the most beloved recipe libraries on the internet. The NYT Cooking AI rollout is now a live case study in how even a gentle, optional feature can ignite a loyalty base that sees the product as human-made by definition.
Table of contents
- What the NYT Cooking AI Recipe Modification Tool Does
- How The Times Built and Guards Its AI Cooking Answers
- The Subscriber Backlash Against NYT Cooking AI
- Why “Can I Turn It Off?” Became the Real Fight
- NYT Cooking AI in the Wider Context
- What Product Teams Should Steal From This Whole Episode
- Where the NYT Cooking AI Rollout Goes Next
- Frequently Asked Questions
- References
What the NYT Cooking AI Recipe Modification Tool Does
Ingredient substitutions that actually change the steps
The headline capability is substitution advice that rewrites the method, not just the shopping list. In one example from the announcement, a cook asks about using baby spinach instead of kale. The NYT Cooking AI tool responds that baby spinach wilts faster, so it should go into the pan during a later step than the kale would have. That is a small, concrete difference — the kind a experienced cook carries in their head and a beginner usually learns through a ruined side dish.
Equipment swaps and tool-aware rewrites
Ko says the feature can also rework recipes around different pans and tools. No Dutch oven? The instructions can be adjusted for a deep skillet with a lid. The substitution logic treats equipment as another constraint to solve for, the same way it treats a missing vegetable. For the NYT Cooking AI rollout, equipment swaps are among the most technically interesting requests, because they change heat transfer, timing, and even doneness cues, not just ingredients.
Scaling to any number of servings
The third pillar is scaling: a recipe built for four can be resized to three, seven, or eleven, with quantities adjusted automatically. NYT Cooking already offered crude halve-and-double controls, but the NYT Cooking AI version scales to arbitrary counts and tries to keep the ratios sensible for real cooking. Privacy is a deliberate boundary: customized recipes stay private to the account and cannot be published or shared through the app.
A gradual, monitored rollout
The feature is being rolled out gradually while the Times monitors how it performs. It is currently available to all Cooking subscribers on the iOS app, and the company says it will be available more broadly soon. NYT product staff say they are watching speed, request patterns, and accuracy — a monitoring posture that matters a great deal to how the NYT Cooking AI experiment will be judged. Gradual rollout is also a humility move: the tool answers conversationally, and conversational systems fail in ways test suites miss, so the Times is effectively letting a few thousand careful cooks stress-test the NYT Cooking AI tool before it reaches everyone.
How The Times Built and Guards Its AI Cooking Answers
Years of quiet recipe engineering
Ko wrote that she “naïvely assumed A.I. wouldn’t affect” her work as a recipe developer before conceding that product and engineering colleagues saw its potential. The NYT Cooking AI work did not start this month. Two years ago the team used AI to convert imperial volume measurements into metric weights — answering the eternal “what is one cup of bok choy in grams?” — and last year it shipped serving-size conversion. Substitution advice is simply the newest rung on that ladder.
Human evaluation against culinary standards
Before launch, Cooking staffers “spent countless hours evaluating sample after sample to get the chatbot answers to closely align with how we would answer as culinary experts,” Ko wrote. “We applied the same rigor as we do to editing, grounded in Times standards.” That line is worth reading twice, because it is the NYT Cooking AI pitch in miniature: the model is the plumbing, the editors remain the standard, and the promise is that a bad substitution answer will be caught the way a bad sentence would be.
What the tool is not
It helps to be precise about the boundaries. The feature does not generate new recipes from scratch, does not replace the test kitchen’s catalog, and does not publish or share anything you customize. Every modified recipe starts from a human-written NYT recipe and departs from it under your instruction. Critics argue the NYT Cooking AI boundary is exactly the problem — that normalizing machine mediation inside a hand-curated library changes what the library is. Supporters argue the boundary is exactly the point.
| NYT Cooking AI milestone | Year | What it changed |
|---|---|---|
| Volume-to-weight conversion | ~2024 | Recipes gained metric gram equivalents automatically |
| Serving-size conversion | 2025 | Recipes could be halved or doubled in one tap |
| Recipe modification tool | Oct 2026 | Substitutions, equipment swaps, arbitrary scaling, on request |
| Broader platform rollout | “Soon” | Beyond iOS to all Cooking subscribers |
The Subscriber Backlash Against NYT Cooking AI
The comment section turned against the feature fast
The pushback was immediate, emotional, and remarkably consistent. “What a massive disappointment,” wrote Susan from New Jersey: “The internet is full of AI cooking content. I pay you because I prefer mine done by people.” Judah Mann of New York put it more bluntly: “PLEASE PLEASE PLEASE DO NOT DO THIS. Oh, wait, you already did. Ok, I’m out.” Another subscriber called it “unbelievably disappointing” that “the AI mandate has reached what used to be a wonderful and simple collection of recipes.”
The numbers behind the anger
Nieman Lab scraped the 171 comments left on the announcement and ran a sentiment check, excluding staff replies. Roughly 77% of reader comments were negative about the new AI features, versus 14% positive, with the remainder neutral or mixed. The most common complaints clustered around four themes: this does not need AI, AI is destroying the environment, AI is not why I pay the Times roughly $390 a year, and the adjustments the NYT Cooking AI produces are going to be bad.
Environmental and industry objections
Recurring comment threads blamed AI’s data-center energy appetite, echoing a wider frustration with infrastructure like the data-center operations expanding across the country. Others objected to the concentration of power among AI companies or warned about systems escaping human control. Ko’s announcement essay actually pre-empted these objections — she devoted space to the environmental impact of AI, the power of AI vendors, and the risk of runaway systems — before arguing that cooking is tactile, physical, and gloriously human, and that the tool exists to protect that pleasure rather than automate it away.
Why "Can I Turn It Off?" Became the Real Fight
No disable switch, but “completely optional”
When one subscriber asked whether there will be a way to disable the feature, Ko answered simply: “No, not at this time,” adding that using it is “completely optional.” That exchange — a user asking for an exit and being told one does not exist — became the argument many commenters said defined the whole episode. The NYT Cooking AI tool is opt-in by behavior: you do not have to use it. It is not opt-out by design: there is no setting that returns the app to its pre-AI state.
Dozens of people asked; the answer stayed no
Nieman Lab’s scrape found dozens of commenters specifically requesting a toggle. The NYT response leaned on the voluntary nature of the feature: don’t ask the tool to swap your chard, and the chard stays unswapped. Product teams reading this exchange should notice how unsatisfying that logic felt to subscribers who see the feature as a change to the artifact itself, not merely to the workflow. A setting that hides a feature is different from a feature you personally decline, and the audience was arguing about the artifact.
Then the comments closed
The announcement’s comment section has since been closed. NYT spokesperson Jordan Cohen told The Verge that comments remained open for four days, that Cooking editors answered questions directly during that window, and that closing comments on older articles is “standard practice at The Times so that our moderators can focus on new discussions.” Cohen added that many Cooking members are “interested in exploring this tool” and “reporting good results” on the iOS app.
| What subscribers asked for | What The Times said |
|---|---|
| A switch to disable the AI features | “No, not at this time” |
| Reassurance the feature is human-supervised | Editors evaluated “sample after sample,” Times standards applied |
| Continued human-only recipe development | Existing catalog unchanged; AI only modifies on request |
| A say in the decision | Comments kept open four days, then closed as standard practice |
NYT Cooking AI in the Wider Context
The recipe app nobody expected to argue about
NYT Cooking has roughly a million-plus paying subscribers and sits in the unusual position of being a subscription product people describe with affection bordering on devotion. That affection is precisely why the NYT Cooking AI fight got loud: the app marketed itself as the calm, human corner of the internet. As one commenter from California wrote, “I am sad that AI has breached the one online resource I truly love.” Food media had largely escaped the AI culture war until now.
A test case for consumer AI trust
Media analysts see the episode as a referendum on where readers will and won’t accept machine mediation. The Verge’s coverage framed the tool as straightforward utility — substitutions, scaling, equipment swaps — while noting that “some users aren’t exactly thrilled.” For every brand with a loyalty base, NYT Cooking AI is now the cautionary tale: utility does not automatically defuse identity-based resistance, and an optional feature is still a feature announcement.
Where the positive voices landed
The minority who liked the feature tended to describe practical wins. Anne Wagner from Kansas City praised how quickly and accurately the tool converted recipes to gluten-free. Andy from San Francisco captured the utilitarian camp: “I don’t cook for stress relief. I cook to put dinner on the table. Anything that makes the process a little faster or smoother is welcome.” A smaller group simply appreciated that the Times had finally answered accessibility needs, like one-handed cooking or low-energy days, that handwritten recipes never addressed.
Dietary-restriction cooks — gluten-free, dairy-free, low-FODMAP — were the most consistent supporters of the NYT Cooking AI tool, because substitution is exactly the labor those cooks perform manually every week.
Why food is different from search
Readers tolerated AI search summaries and AI writing aids with shrugs; they are reaching for a NYT Cooking AI feature in the app they associate with a human test kitchen, a specific columnist’s lemon curd, and the memory of a Christmas dinner that finally worked. In categories where the product is explicitly craft, consumers treat machine mediation as a downgrade of the artifact itself, not just of the workflow. That is a different contract than search, where users expect machines. The NYT Cooking AI episode is the first time that contract was tested at recipe-app scale, and the community verdict says the craft contract wins — for now.
What Product Teams Should Steal From This Whole Episode
Announce the boundary before you announce the capability
Ko’s essay tried to do this — she addressed objections before readers raised most of them — and it still did not land. The lesson is not that pre-emption is useless but that audiences parse pre-emption as persuasion. When a community’s loyalty rests on craft, a capability announcement reads as a values announcement. Teams shipping consumer features like the AI models and tools hub covers daily should study the order of operations: what stays human, what the machine may touch, and what the user controls.
Optional is not the same as invisible
The NYT position — “completely optional to use” — is legally and operationally true, yet it failed as communication because the feature changes the app’s surface for everyone who opens it. If a feature can be hidden behind a setting without harming its purpose, hiding it is cheap insurance; if it cannot, owning the change loudly beats soft-pedaling. This distinction matters for any product team automating workflows their users consider part of their craft — see also our breakdown of intelligent automation adoption patterns.
The NYT Cooking AI episode is the cheapest UX research anyone will run this year: a real audience, a real feature, and a transcript of exactly where the line is felt.
Instrument what people actually ask for
NYT staff say the most common substitution requests involve ingredients — the chard-to-spinach problem in a thousand variations. That usage telemetry is the real story of the first weeks. If the tool mostly helps dietary restrictions, budget swaps, and regional availability, the NYT Cooking AI experiment will look like accessibility engineering rather than a betrayal. If it mostly helps laziness, the critics will feel vindicated. Either way, the request log is the scoreboard.
Where the NYT Cooking AI Rollout Goes Next
Android and web are the pressure points
The feature currently ships only on iOS, which means the most ardent detractors — many of whom write comment sections from desktop browsers — cannot even test what they are angry about. When the NYT Cooking AI tool reaches Android and the web app, the argument will shift from principle to evidence: what the tool actually answers, how often it is right, and whether a bad answer appears in a shared screenshot at the worst possible moment.
Accuracy is the only durable defense
A substitution engine will be judged by its worst public output. Tell the tool you are out of buttermilk and it suggests a milk-and-vinegar stand-in at the wrong ratio, and the screenshot will outlive the apology. The Times’ “rigor… grounded in Times standards” framing sets a bar that every future update has to clear, forever, while competitors can improve on average. That asymmetry — humans forgive institutions slowly and screenshots quickly — is the core risk of consumer AI features sold on trust. Food adds a second multiplier: a wrong answer isn’t just misinformation, it’s a ruined dinner, and the NYT Cooking AI tool operates on dinners.
The industry is watching closely
Every food publication with a test kitchen now has a data point: you can ship an AI recipe assistant and survive it, but the comment section will be brutal for a week. Some will wait. Others will accelerate quietly, shipping measurement and scaling utilities — the unglamorous, verifiable ones — before touching substitution advice. The NYT Cooking AI experiment has effectively ranked the features by controversy, and most of the industry will work up the ladder in exactly that order.
Competing food apps will watch the NYT Cooking AI retention numbers for six months before deciding whether loyalty survives machine mediation, and their users will inherit whatever conclusion the data supports.
Frequently Asked Questions
Is the NYT Cooking AI feature free for all subscribers?
Yes. The recipe modification tool is available to all NYT Cooking subscribers on the iOS app at launch, with broader availability promised “soon.” NYT Cooking subscriptions are sold as part of New York Times plans; there is no separate AI charge, which is one reason subscribers argued they never asked for or paid for this.
Does the AI rewrite recipes automatically?
No. Modifications happen only when a cook asks for them. The original recipe text remains exactly as the test kitchen published it, and customized versions are private, unpublished, and confined to the requesting account. The tool suggests changes — to ingredients, equipment, and serving counts — and the cook decides whether to use them.
Can I turn the AI features off?
Not right now. When asked directly about a way to disable the feature, NYT Cooking senior editor Genevieve Ko replied “No, not at this time,” while stressing that the feature is “completely optional to use.” As of this writing, that answer has not changed.
What happens if the AI gives a bad substitution?
The Times says human evaluation during development, grounded in editorial standards, is designed to keep answers culinary-sound, and customized recipes are always clearly yours rather than the test kitchen’s. That said, any modification is presented for your judgment; the published recipe remains the authoritative version, and the app’s own guidance positions the tool as advice, not instruction.
Is this the same feature that converts cups to grams?
Related but distinct. The measurement-conversion work from two years ago and the last year’s serving-size converter were quiet utilities. The new recipe modification tool is the first NYT Cooking AI capability that rewrites cooking instructions conversationally — which makes it the first one people feel strongly about.
