The New York Times Cooking App’s AI Feature: A Risky Bet That Works (Mostly)

In a recent New York Times Cooking video, while Melissa Clark demonstrated her lemony white bean soup with turkey and greens, the esteemed food reporter and recipe columnist engaged viewers with a casual aside, inviting them to share their ingenious uses for kale stems in the comments section. This seemingly minor interaction underscored the unique and cherished relationship between The New York Times Cooking’s celebrated personalities and its dedicated audience—a bond characterized by a warm, casual, and deeply interactive vibe that has fostered a vibrant culinary community.

It was therefore with considerable surprise that, just a few days later, a segment of this very audience was observed in open revolt. The cause of this unexpected uprising was an announcement made on October 23, 2026: The Times was integrating artificial intelligence into its iOS-only beta application, allowing subscribers to modify recipes using AI. This news, far from being welcomed, ignited a firestorm of protest among the fervent "commenter army" of the Cooking section.

The article announcing this change was met with an almost immediate deluge of grievances. Commenters loudly aired their discontent, with many threatening to cancel their subscriptions, questioning the necessity of such a feature, and demanding instructions on how to disable it. The intensity of the backlash was striking; after merely four days, and a staggering 172 overwhelmingly angry comments, the commenting function for the story was abruptly disabled. While The Times does not publicly disclose subscriber numbers for its Cooking section, the platform’s considerable reach is evident: NYT executive communications director Jordan Cohen reported 514 million visits to the site and app in 2025, and its YouTube channel boasts over 1.4 million subscribers, highlighting the sheer scale of the community impacted by this decision.

Considering my own decade-long experience writing for WIRED, which has repeatedly showcased the inherent risks of integrating technology into kitchen products, my initial inclination was to side with the disgruntled commenters. I have witnessed countless companies attempt to shoehorn "smart" features into culinary tools, often with disappointing results. Bluetooth connections for every appliance, apps burdened with poorly conceived recipes, and even toasters fitted with optical sensors or touchscreens that failed to improve the fundamental act of toasting, have all contributed to a pervasive skepticism. The "smart kitchen" often proved to be anything but intelligent, frequently complicating simple tasks rather than enhancing the cooking experience. This history of tech-infused culinary missteps made the prospect of AI tinkering with expertly crafted recipes seem like a precarious venture.

Developing truly dependable and delicious recipes is an arduous and expensive endeavor, demanding a dedicated staff of culinary professionals, extensive testing, and meticulous refinement. It is this rigorous process that builds a brand’s reputation for quality and trustworthiness. Companies like The New York Times Cooking and America’s Test Kitchen have built their entire empires on this premise: their value lies in providing consistently high-quality, thoroughly tested recipes that deliver reliable results. Given that AI, in many culinary applications, has proven to be a "lousy chef," introducing it into the presentation and modification of NYT Cooking’s "gold mine" of meticulously developed content represented a significant risk to a hard-earned legacy of culinary excellence.

The Times, however, had previously introduced AI to assist with more straightforward functionalities earlier in the year, such as recipe scaling (adjusting ingredient quantities to serve more or fewer people) and converting measurements between imperial and metric systems (cups, tablespoons, and Fahrenheit to grams, milliliters, and Celsius). The new, more ambitious "customize this recipe" feature, available exclusively in the iOS-only beta, appears as a prominent field just above the ingredient list, inviting users to interact directly with the AI.

The capabilities of this new tool are surprisingly robust and practical. If a user wishes to make banana bread but lacks a loaf pan, the AI will prompt them for available pan types and then intelligently modify the recipe to suit. All changes made by the AI are clearly delineated throughout the recipe in a distinct red-orange text, offering full transparency. Over several days of testing, I observed its ability to handle both relatively simple requests, such as swapping ground pork for ground beef, and more complex demands, like tailoring a recipe to specific dietary needs. Furthermore, the system incorporates a "hard redirect" mechanism for requests deemed too far outside the spirit of the original recipe. For instance, asking it to transform Eric Kim’s gochujang buttered noodles into a cookie will result in the AI suggesting a more appropriate recipe, such as Kim’s gochujang caramel cookies, thereby safeguarding the integrity of the original creation while guiding the user towards a sensible alternative. Screenshots provided by The New York Times illustrate both the chat interface for AI interaction and the visual highlighting of AI-driven adjustments within a recipe.

Camilla Velasquez, the general manager of NYT Cooking, offered insight into the philosophy behind the feature. She emphasized the significant effort invested in ensuring the AI understands "when not to do something if it’s no longer in the spirit of the recipe," indicating a conscious effort to preserve the culinary intent of the original creators. The AI draws its recommendations from a multi-layered knowledge base: first, the specific recipe itself; then, The Times’ extensive ingredient substitution guide; followed by insights gleaned from user comments; and finally, what the team describes as AI "drawing on our expertise"—a broader application of the Times’ collective culinary knowledge. Velasquez clarified that the AI is intended to be a "helper, not the main act," stressing that users "shouldn’t be chatting with it" in a general conversational sense, but rather using it for targeted modifications. She also explained that recipe editors played a crucial role in creating the original rule set for the AI and subsequently evaluated its performance, ensuring a human-centric oversight in its development.

Addressing concerns about job security for the 60 editors, writers, video journalists, and other employees who constitute the NYT Cooking team, Velasquez unequivocally stated that NYT Cooking has been "hiring more than ever" and has "zero intention of publishing fewer recipes." This reassurance aims to quell fears that AI might diminish the need for human culinary experts. She noted that current user data indicates the AI is predominantly utilized for ingredient and equipment substitutions, adapting recipes for dietary needs, and scaling. For example, a user might ask the AI to adjust a recipe that calls for more oil than they have on hand, demonstrating its utility in practical kitchen scenarios.

As a food writer whose career has recently navigated significant shifts due to AI advancements, I wouldn’t claim to have been clamoring for such a feature. However, after engaging with it, I found The Times’ measured approach to be remarkably effective. My testing began with Alison Roman’s five-star chicken and dumplings recipe, which boasts over 6,000 ratings. I tasked the AI with swapping bone-in, skin-on chicken thighs for boneless, skinless breasts. After a brief "simmering" period—the AI Cooking patois for processing—it presented "Chicken Breast and Dumplings," with the word "breast" highlighted in the characteristic red-orange. Just below the title, a clear disclaimer read: "Joseph’s version made with AI. Not tested or endorsed by the original recipe’s creators." The app also allowed me to toggle seamlessly between my AI-modified version and the original.

During the cooking process, the modifications proved smooth and logical. As a recipe follower, not an improviser, I appreciated the clarity. The AI intelligently flagged that chicken breasts would require a reduced simmering time compared to thighs and, recognizing that skinless breasts render less fat, it appropriately added butter to the roux. These changes were seamlessly integrated into the correct procedural steps. The only notable issue arose when the AI suggested returning the shredded, cooked breast meat to the pot for another 30 minutes while the sauce thickened and dumplings cooked. Foreseeing potential overcooking, I added only a handful as directed to observe the outcome, reserving the rest. The meat that underwent the additional cooking indeed had a noticeably less succulent texture than what I had reserved, and certainly less than thighs would have. This represented the most significant imperfection encountered in my testing, prompting a question as to whether this specific substitution, while not uncommon, should be offered without further refinement to prevent texture compromise.

Beyond this, the testing proceeded remarkably smoothly. I challenged the AI to transform Francis Lam’s roasted mushroom and broccoli grain bowl into a roasted mushroom, cauliflower, and squash grain bowl. The swap of cruciferous vegetables was a straightforward one-to-one, but I also requested the addition of a pound of delicata squash. The AI intelligently instructed me to slice the squash into half moons to ensure even cooking times with the other vegetables and adjusted the total amount of olive oil accordingly. Impressively, Lam’s distinctive poetic voice, which describes heating olive oil "until it slips around the pan like water" before adding beaten eggs for "festive dish-topping egg ribbons," was preserved in both versions of the recipe. This made for an enjoyable and healthy pre-game dinner before a cribbage night.

The AI also demonstrated its competence with more complex culinary chemistry. It capably transformed Yewande Komolafe’s Toasted Sesame and Scallion Waffles into a yeasted version, which, unlike the original, required an overnight refrigeration period, allowing active dry yeast to develop bubbles and an appealing tang.

For Colu Henry’s pasta and chickpea stew (Pasta e Ceci), I swapped canned whole tomatoes for diced and asked it to "add more vegetables." The AI provided relevant options—carrots, celery, onion, zucchini, mushrooms, and extra greens—and seamlessly integrated my choices into the procedure, ensuring each vegetable was added at the appropriate cooking stage for a healthy and hearty dinner.

Inspired by a recent visit to the Portland Fermentation Festival, I sought guidance on incorporating additional ingredients into Lidey Heuck’s sauerkraut recipe. Requesting "add carrots, garlic, and kale," the AI suggested grating the carrots, thin-slicing the kale, and mincing the garlic, all in proportions that felt intuitively correct. It also increased the salt content to account for the larger volume of kraut. As I write, the large jar is fermenting, and a preliminary sniff test suggests a promising batch.

Finally, I undertook a multi-faceted modification of Sue Li’s peanut noodles with spicy tofu, aiming for peanut udon with tofu, whitefish, and cucumber. I initially asked for a less spicy version, with added vegetables, keeping the tofu, and incorporating some fresh rockfish from the grocery store. The AI’s impressive logic was evident in its sequential instructions: it first had me cook the fish, then set it aside and wipe out the pan before proceeding to cook the tofu. Furthermore, it customized the noodle-cooking guidance specifically for udon, replacing the original bucatini instructions.

It seems entirely natural that The New York Times Cooking section encountered significant resistance upon launching this AI feature. The Times employs some of the most talented and respected recipe creators in the culinary world, whose meticulously tested and thoughtfully designed recipes are a hallmark of the brand. While using AI to facilitate substitutions is undeniably helpful—saving a trip to the store, utilizing ingredients on hand, or accommodating dietary restrictions—this tinkering with the original recipe is fundamentally a subtractive or adaptive process. It prompts the question: how many of us, making substitutions via AI, are likely to land on something genuinely better than what culinary masters like Melissa Clark, Yewande Komolafe, or Francis Lam set out to create by design?

Instead, the AI serves as a powerful utility, enabling incremental improvements to the user experience without necessarily "improving" the core culinary genius. It stands as a rare and commendable example of how technology can genuinely enhance a product’s practical utility without compromising its core values. The concerns raised by the commenters have clearly been heard and addressed. Camilla Velasquez promptly informed me that users will have the option to hide the AI feature within the app if they prefer. Furthermore, The Times is maintaining the feature in its iOS-only beta phase for the foreseeable future, with no immediate plans for broader platform rollout, demonstrating a cautious and responsive approach to user feedback.

The unique bond between the NYT Cooking team and its users—a connection built on trust, shared passion, and human expertise—is a truly valuable asset. Deep down, the upset commenters likely feared losing some of that cherished human connection and authenticity. My hope is that, despite the advancements in technology, this invaluable human element will endure and continue to define the New York Times Cooking experience.