AI in the Kitchen: Formulating Questions for Culinary R&D

It is undeniable that a unique era is unfolding in the culinary world, one we should observe with care. A cook can open a laptop or phone, ask a question, and receive an answer almost immediately. A recipe. A substitution. A technique. A list of flavor pairings. A possible menu. A way to stabilize a foam. A rough formula for a sauce. A suggestion for what to do with an ingredient that has been sitting in the refrigerator for three days. Etc. Etc. Etc. This is new. Or at least, the speed of it is new.

Cooks have always looked for help. We have always asked questions. We have asked chefs, books, grandmothers, vendors, colleagues, encyclopedias, markets, suppliers, and the person working next to us on the line. Cooking has always been built through exchange. What changes with AI is not the existence of the question, but the immediacy of the response. And that immediacy can be useful. It can also be dangerous. Not dangerous because a cook asks for help. There is nothing wrong with asking for help. The danger is that the answer arrives so quickly that we may forget to ask whether the question was good.

In cooking, the quality of the question matters. Maybe more than the answer. This is especially true for a cook interested in creativity, authorship, and the development of a personal way of working. If you ask AI for “a great recipe,” it may give you something that looks complete. It may sound convincing. It may even work reasonably well. But there is a difference between receiving a recipe and developing an idea. The first gives you something to follow. The second teaches you how to think. And for the creative cook, that distinction matters. The goal is not to use AI as a machine that produces dishes for you. The goal is to use it, if you choose to use it, as an assistant inside a larger process that remains yours. Your taste. Your questions. Your research. Your tests. Your notes. Your failures. Your refinement. That is where authorship lives.

AI does not replace the creative process. At its best, it can help you organize parts of it. But only if you know where you are in the process. This is why I find it more useful to think about AI through the same structure I use for culinary development: research, experimentation, documentation, and refinement. If you are in the research phase, the question should sound like research. You are not asking for a finished dish yet. You are trying to understand a field. You might be looking into the history of an ingredient, common preparations, regional uses, possible techniques, similar ingredients, traditional pairings, or how a process behaves. You might ask for references, comparisons, terminology, or questions worth investigating further.

But even then, the answer is only a beginning. AI can point you toward areas of inquiry, but it does not absolve you from verifying, reading, tasting, visiting, asking, and observing. It can summarize, but it cannot stand in a market for you. It cannot smell the dried chile. It cannot tell you how the masa feels in the hand of someone who has made tortillas for thirty years. It cannot taste whether the broth has the right depth. It can help you map the territory. It cannot walk the territory for you. If you are in the experimentation phase, the questions change. Now you are asking about possibilities. What happens if this ingredient is roasted instead of boiled? What are three ways to create bitterness without overpowering sweetness? How could I introduce acidity into a creamy preparation without splitting it? What techniques might help me create a stable foam from a liquid with very little fat? What variables should I test if I am trying to improve the texture of a particular filling?

These are useful questions because they do not ask AI to replace your work. They ask it to help you design the structure of your R&D work. That is very different. A good experimental question gives you options to test, not conclusions to obey. It helps you see variables. Heat. Time. acidity. Fat. starch. protein. water content. temperature. resting. grinding. aeration. concentration. It helps you build a small field of trials. But the kitchen still decides. 

Something may sound elegant in language and fail completely in the pan. A combination may look logical and taste flat. A technique may be technically correct but wrong for the dish. A suggested ratio may be a decent starting point, but not the final formula. This is not a problem. This is cooking. Ideas have to pass through the effects of transforming matter. They have to pass through texture, heat, smell, appetite, service, cost, repetition, and the palate of real people. This is where AI reaches its limit. It can suggest. It can compare. It can organize. It can simulate logic. But it cannot taste.

That limitation should not make the tool useless. It should make the cook more responsible. If you are in the documentation phase, AI can become useful in another way. It can help you organize what happened. You can take rough notes from a test and ask for a clearer structure. You can ask what variables you forgot to record. You can ask how to turn observations into a testing table. You can compare Trial 1, Trial 2, and Trial 3 and identify what changed. You can ask for a template that helps you track ratios, temperatures, timings, sensory results, failures, and next steps. This is one of the places where AI can be genuinely valuable for a cook doing R&D.

Because many cooks test things and lose them. They cook, adjust, taste, get excited, move on, and then a week later they cannot remember exactly what happened. Was it 12 grams of salt or 16? Did the sauce rest before blending? Was the acidity added before reduction or after? Which version had the better texture? Why did the second one split? Memory is not documentation. And in creative cooking, undocumented work disappears. AI can help you build better records. But you still have to keep the records. You still have to write down what happened. You still have to photograph, label, date, compare, and return to the work with discipline. A tool cannot put the rigor in the work for you. It can only support the rigor you are willing to practice.

Then comes refinement. This is where the work becomes serious.A dish, product, sauce, formula, or menu idea may be interesting, but not yet ready. It may need better balance. A cleaner texture. A more stable structure. A more realistic cost. A clearer prep system. A shorter ingredient list. A better service flow. A stronger story. Here, AI can help you ask: what needs to be improved? What are the weak points? What assumptions should I test? What alternatives could reduce cost without losing quality? How might this be scaled? What part of this dish is essential, and what is only decorative? How could this recipe be written more clearly for another cook? These are refinement questions. They are not glamorous. But they are where many good ideas become usable.

The danger with AI is that it can make everything feel finished too early. It produces clean language. It gives structure. It creates the illusion that the idea has already been resolved. But a polished paragraph is not a polished dish. A convincing recipe is not a tested formula. A menu description is not a service-ready plate. The cook has to resist the seduction of completion. You are not finished because the answer looks complete. You are finished when the work holds. When the recipe can be repeated. When the texture is stable. When the flavor is clear. When the timing works. When another cook can follow the formula. When the dish survives service. When the idea has become real.

This is why I think the most important skill in using AI is not technical. It is not knowing a magic prompt. It is knowing how to ask from the right contextual position. Where am I in the process? Am I researching? Am I experimenting? Am I documenting? Am I refining? What do I need from this answer? What will I test after receiving it? Without that clarity, AI becomes a machine for producing more noise, basically your own noise. It gives you more ideas, more recipes, more possibilities, more names, more pairings, more directions. And at first, that can feel exciting. But too many directions can also weaken the work. You begin collecting instead of developing. You begin jumping from one possibility to another. You begin confusing quantity with depth.

But to be fair, this is a problem creative cooks already had before AI. AI simply accelerates it. A creative cook does not need endless ideas. A creative cook needs a way to choose, test, and refine the right ones. That is where formulating “the question” becomes central. A weak question asks for the whole answer too soon. “Give me a creative dish with carrots.” A stronger question gives context, intention, and constraints. “I am developing a carrot dish for a spring menu. I want the carrot to remain central, not become a garnish. I am interested in sweetness, smoke, acidity, and a contrast between soft and crisp textures. What are five possible directions I could test, and what variables should I document?”

This second question does something different. It does not surrender authorship. It frames the work. It tells the assistant where the cook is standing, what the cook cares about, and what kind of help is needed.That is the real craft of using AI well. Not asking it to be creative for you, but asking it to help you think more clearly about your own creative process. There is also value in asking it to show its path. Not because the answer will always be perfect, but because the reasoning can reveal useful references, assumptions, or gaps. You can ask: why are you suggesting this? What culinary principle is behind this technique? What traditions or preparations are similar? What could be wrong with this idea? What should I verify before testing it? What might fail?

These questions are important because they prevent passive use. You are not just receiving. You are inquiring with the purpose of expanding your knowledge. You are treating the answer as material. Like any ingredient, it has to be evaluated. Does it have structure? Does it make sense? Is it generic? Is it culturally shallow? Is it technically sound? Is it useful for the kitchen I am actually working in? This matters especially when dealing with culinary culture and heritage. AI can flatten traditions. It can produce confident summaries that miss nuance. It can combine ingredients in ways that sound plausible but ignore context. It can make food sound universal when it is actually specific.

So as the creative chef of the kitchen, you have to remain alert. If you are working with an ingredient, dish, or technique from a culture you do not know well, do not let AI be the final authority. Use it to begin research, to find better questions, to identify possible sources, to understand terms. But then go further. Read serious references. Talk to people. Taste the food where it is made with knowledge. Visit the shops. Ask the vendor. Go to the restaurant where the dish is not being explained for outsiders, but served to people who already know what it should be.

Respect requires contact. AI can assist inquiry. It cannot replace relationships. There is also the practical side. Sometimes you are in the middle of work and you need a quick starting point. You want to know how to create foam from a low-fat liquid. You want a basic ratio to begin testing a gel. You want to understand why an emulsion is splitting. You want to compare stabilizers, thickeners, acids, sugars, or cooking temperatures. In those moments, AI can function almost like a fast assistant.

That can be useful. But even then, the answer should be treated as a starting point, not a finished authority. If it gives you a ratio, test it small. If it gives you a method, check whether it fits your equipment. If it gives you a substitution, taste it. If it gives you a safety-related answer, verify it through reliable sources and professional standards. The kitchen is not only theoretical. Real ingredients behave in real ways. This is why side notes matter. If you are using AI as part of your R&D process, document the conversation itself. Not every word, necessarily, but the important parts: the question you asked, the context you gave, the answer that helped, the direction you chose, the test that followed, the result in the kitchen. Otherwise, you will lose the path that led to the idea.

And the path matters. Because creativity is not only the final dish. It is the route by which the dish became possible. If you do not record your route, you may not be able to return to it. This is one of the simplest ways to keep authorship intact. You are not hiding the fact that a tool helped you. You are showing how you used the tool inside your own method. The idea still has to pass through your judgment, your hands, your tasting, your refinement, and your standards. That is the work. AI can be one of the most useful tools available to a creative cook, but only if it remains in its proper place. It is not the chef. It is not the palate. It is not the culture. It is not the service. It is not the memory of your grandmother, the hand of the tortilla maker, the smell of the stock at six in the morning, the pressure of the line, the guest’s reaction, the body of the cook standing in front of the stove. 

It is an assistant. A very powerful one, perhaps. But still an assistant. And an assistant becomes useful when the cook knows what to ask. So before you ask for a recipe, ask yourself what you are really doing. Are you trying to understand? Are you trying to test? Are you trying to organize? Are you trying to refine? Are you trying to solve a technical problem? Are you trying to find a point of departure? The better your question, the better the work that can follow. Not because AI will give you the final answer. But because a good question places you in the right relationship to the process. And in the end, this is what matters. The creative chef does not collect answers. The creative chef develops them. AI can help. But the cooking still has to be yours.