Allergies, preferences, goals: cooking for a table that eats differently

Real tables are mixed. One nut allergy, one vegetarian, one person quietly trying to eat more protein, one child at war with mushrooms. Cooking for that table with internet recipes means running every candidate through a mental checklist — and the checklist never gets easier, because every recipe site starts from zero on who you are.

The cost isn't just effort. Mixed tables under checklist pressure converge on the same defensive rotation — the four dishes everyone can eat — and dinner gets smaller every month. Or worse, the household splits into parallel meals: one pan for the vegetarian, one for everyone else, twice the washing-up and half the point of eating together.

The unglamorous fix is a profile: say it once, and let every recipe that reaches you already know. But “the app knows my diet” only helps if it treats different kinds of dietary information with different levels of seriousness — because an allergy and a mushroom boycott are not the same kind of fact, and a tool that flattens them fails at both ends.

Three kinds of dietary truth

• Allergies are hard lines. In RecipeGPT, a saved allergy never appears in any recipe, in any form — and derived ingredients count, so a peanut allergy also rules out peanut oil and satay sauce. If you explicitly ask for a dish that conflicts, it asks once whether it's for someone else rather than silently including the allergen.
• Preferences are defaults. Vegetarian, halal, kosher, no red meat, “no mushrooms, ever”: recipes follow them automatically, but your own message can override them without a fight — say “chicken tonight” and the pescatarian default steps aside for the dish you actually asked for. No confirmation dialogue, no guilt trip.
• Goals are direction. “Eat healthier”, “high protein”, “less sugar”, “quick meals”, “budget-friendly” steer portions, swaps and ingredient choices — a thumb on the scale of every dish, not a straitjacket on any of them.

That grading is the whole trick. Flatten it one way — treat the mushroom boycott as seriously as the nut allergy — and the tool becomes a nag that blocks half of cooking, so you turn it off and lose the allergy protection with it. Flatten it the other way and it's dangerous. Graded authority is what lets you save everything about your table without the strict things and the soft things interfering with each other.

The derived-ingredients problem (where checklists fail)

The reason manual allergen-checking is exhausting isn't the obvious ingredients — nobody with a peanut allergy misses “peanuts” in a list. It's the derivatives and the disguises: peanut oil in the crispy shallots, satay in the dressing, groundnut oil labelled by its politer name. Sesame hides in tahini, which hides in hummus. Dairy travels as whey, casein and ghee. A useful tool has to reason about what ingredients are made of, not just string-match the shopping list — which is exactly what “derived ingredients count” means in practice, and why a saved allergy in RecipeGPT rules out the whole family tree of an allergen rather than one word.

The same reasoning runs through revisions, which is where ad-hoc caution usually slips. It's easy to check the first recipe; it's the “actually, make it a stir-fry instead” three messages later where a forgotten constraint sneaks back in. A profile applies to every turn, every revision, every suggestion — the tenth message is as protected as the first.

What honesty requires: the limits

Here's the line an honest tool won't cross: it will never tell you a dish is “safe” for your allergy. RecipeGPT excludes saved allergens rigorously and still won't make that claim — because the recipe can't see your actual shopping. The pesto jar's label, the “may contain traces” warning, the fryer the chips shared: those checks belong to you, and any tool that says otherwise is overpromising on your safety. A saved profile removes the re-typing and the silly mistakes; it doesn't retire your label-reading. For anaphylactic-grade allergies, your allergist's rules outrank every app, this one included.

Three mixed tables, worked through

The nut allergy + the vegetarian. Profile: peanuts and tree nuts as allergies, vegetarian as a preference. Ask for a Thai-style curry night and the answer arrives nut-free by construction — no satay, no peanut garnish, coconut doing the richness — and meat-free by default. When one of you wants the sticky chicken thighs, say so: the vegetarian default steps aside for that request, the allergy never does.

The high-protein half and the just-eat-dinner half. Goals do their best work here because they bend dishes instead of splitting them. “High protein” saved by one cook steers the shared dinner — chickpeas doubled in the quinoa bowl, an extra egg in the shakshuka — without turning the table into a supplements aisle. One dinner, nudged, beats two dinners, negotiated.

The child at war with mushrooms. File it as the preference it is and every suggestion simply routes around mushrooms — no nightly negotiation, no picking-out ritual. Years later, when the war ends, delete one line and the mushrooms come back. Preferences are meant to be cheap to hold and cheap to change; that's what makes people actually keep them accurate.

One table, one dinner

The quiet win is social: instead of cooking two dinners or defaulting to the one dish everyone tolerates, you ask for the family classic made vegetarian, or whatever the fridge suggests minus the allergen, and get one meal the whole table can share. It composes with everything else: an imported recipe link gets rebuilt around your table's rules, a scaled-up Sunday version keeps the same constraints at eight servings that it had at two. AI recipe tools earn their place not by inventing exotic dishes but by making the ordinary ones fit the people actually eating them.

It changes hosting, too. “Anything you don't eat?” texted before a dinner party used to mean an evening of recipe vetting; now the answers go into the request itself — “that lamb thing, but gluten-free and one portion vegetarian” — and the plan comes back already shaped for the guest list. Cooking for people who eat differently stops being a test of vigilance and goes back to being hospitality.