Experimenting with AI has become a hobby for me, as well as a professional role. I’d used AI for quick recipe inspiration before, but this time, I wanted to see if it could handle a week of dinners tailored to when I’m cooking, either for my wife and me or just for myself. I know my way around a kitchen even if my talents fall far short of my wife (whose family once showed off their skills on a TV show). Nonetheless, I was impressed with how AI, in this case, ChatGPT, opened some delicious new culinary doors, albeit with more than one tripwire to watch for when stepping through. Here’s how my AI sous-chef performed and how you can do the same without any of the disasters I faced.

AI meal prep

I wanted to set the whole week up before beginning, so I could do the shopping all at once. I discussed with the AI some of my tastes and what my wife and I both liked. We keep kosher, so I made sure to emphasize that, which meant making it clear that the recipes couldn’t include ingredients like pork and shellfish, couldn’t combine meat with dairy, and so on.

I began by giving relatively specific recipe prompts to see how ChatGPT would do with narrow parameters before just asking for ideas without any specific requests. The AI did well at devising a kosher dinner recipe for two featuring chicken and a vegetarian recipe using rice and seasonal vegetables. As a further experiment, I even uploaded images from recipe books for visual reference. The AI impressed me with its adaptability, suggesting dishes like roasted herb chicken with glazed carrots and potatoes and rice-stuffed peppers with a spicy tomato sauce.

Unprompted, the chatbot even showed what it thought the meals might look like on the plate. They were mostly appetizing, though sometimes the details would be a little off, as with the chicken and carrots being hard to distinguish.

Virtual Michelin stars

AI turns out to be pretty good at putting together a varied week of meals. The recipe for herb-crusted salmon with couscous and asparagus was delightful. I’d never tried making a citrus zest crust, but the instructions were clear, and we ended up devouring the whole thing. On an evening alone, the AI walked me through making chicken shawarma wraps with tahini sauce and Israeli salad. The chatbot enhanced my experience with a few cooking and spicing suggestions that kept me warm even as snow fell outside.

Probably the most successful dish was the scratch-made mushroom risotto. I’d count it at the far end of my cooking technique, but the meticulous detail put me through the labor-intensive process without any breakdowns, while the truffle oil and white wine made the dish feel way fancier than it was. Those kinds of small tweaks throughout the week made doing the dishes after feel like less of a chore.

Luckily, another night was a basic grilled veggie platter with hummus and other dips, so I wasn’t totally exhausted every night. It’s not that these recipes were hard to find on their own or in many cookbooks, but the AI did make the process feel like it was customized to my preferences and overall schedule.

Kitchen nightmares

I committed to following every recipe that ChatGPT provided (barring any obvious poisons), but that almost led to me aborting the whole experiment on day two when my inexperience led to a disastrous sweet potato curry. The description of a hearty, spiced dish sounded great, but I ignored how the recipe called for way too much garam masala and cayenne. It seemed like a lot even to me at the time, but the success of the fish lulled me, and I went along with it. I like spice, but after one bite, my taste buds went dead, and I couldn’t speak for hours. It might have been okay after scraping off the sauce, but the cooking time for the sweet potatoes was also off. Biting down released shards of potato that scraped my mouth just in time for the cayenne to really get in there.

I recovered, and all seemed well until the final night when I thought a dairy-free baked ziti sounded easy and fun. The AI apparently didn’t understand how plant-based cheeses behave when baked. The result was a grainy, unappetizing texture that made the entire dish feel like a chore to eat. Adding to the disappointment, the AI recommended mixing the marinara sauce with nutritional yeast, which created a bland, oddly bitter flavor profile. It looked fine coming out of the oven, but that first bite was all I needed to know it was a failure.

Leftovers

The experiment cemented some of my opinions of AI in the kitchen for good and ill. As AI is now, you can definitely benefit from employing it as a helper. But only as a helper, not as the final word. Clear, detailed prompts are crucial if you have a specific idea of the kind of dish you want. But don’t throw away your recipe books. AI is not a replacement for human chefs making a menu. You can and should trust your instincts and make adjustments if you think it makes sense. Perhaps dialing down the cayenne or looking up how to best use substitute ingredients. Still, as inspiration and culinary tryouts, you could do worse than popping a chef’s hat on ChatGPT.

You might also like…

• ChatGPT coded a game for me in seconds and I am simply astounded – and coders should be very worried
• ChatGPT’s new ‘Canvas’ is the AI collaborator you didn’t know you needed
• ChatGPT o1-preview can solve riddles faster than me and I kind of hate it for it
• ChatGPT’s 4o-mini model just got a big upgrade – here are 4 of the best new features