What a meal plan generator gets right (and wrong)

The week it wrote

I gave FamilyPlate's generator a brand-new household with an empty meal library, the hardest case, since it had no meals of ours to draw from. Ten seconds later it returned seven complete dinners. Every one of them was food a family could eat. Read the third column, though: that is where the plan meets a real household.

Add up the ingredient lines and the week needs 41 items, because the seven dinners share almost nothing: salmon, beef, chicken, cod, mushrooms, tofu, chicken again. No night reuses another night's cooking. That is the shape of machine-written weeks in every tool I have tried, ours included: each dinner is plausible alone, and the week has no architecture.

What a generator does well

The blank page. On a Sunday when nobody can name one dinner, let alone seven, a generator ends the standoff in seconds, and arguing with a draft is easier than writing one. Four of its seven dinners survived into the week we ate, which is a better hit rate than our tired Sunday selves manage some weeks. It also breaks ruts: fajita bowls had never occurred to us, and they have been back twice since.

Constraints are the other strength. Tell it vegetarian, dairy-free, or five dinners instead of seven, and the draft respects the rule without complaint. If a diet change has thrown your usual rotation out, a generated week is a fast way to see what the new constraint leaves you.

Where it fails

Every failure in that table is the same failure: the generator plans food, and a household plans a week. It cannot see the calendar, so the slow ragu landed on a work night and the 40-minute salmon on soccer night. It cannot see the pantry, so it bought curry paste we own. It does not know the kids, so taco night and pizza night, the two dinners that defend themselves, got overwritten by strangers.

The subtler miss is structure. A human-planned week has load-bearing habits: a night that eats a double batch from earlier in the week, fragile food cooked before it turns, a cheap pantry dinner before the next shop. Generators skip all of it and write seven stand-alone performances, seven proteins, seven cleanups. Ours has the same blind spots when the library is empty; I am not grading the competition on a curve I exempt myself from.

None of this makes the draft worthless. It makes the draft a first pass, and first passes respond well to checklists.

Check 1: Put the plan against the calendar

Open the family calendar next to the generated week and move the slow dinners onto the slow evenings. This one swap saves more plans than any other: an ambitious dinner on a packed night becomes takeout, and a plan broken on Tuesday is abandoned by Thursday. In our week the salmon moved to Sunday and the fajita bowls took soccer night.

Check 2: Restore your standing slots

Write your repeating dinners back in before you judge anything else, because the generator cannot know Tuesday is taco night. Standing slots are the strongest part of a family meal planning system, decisions you made once that keep paying, and no generated dish beats a dinner the kids defend. The ragu and the risotto came out; tacos and pizza went back.

Check 3: Cut the week to three proteins

Trim the plan until two or three proteins cover all seven nights, so ingredients overlap and the grocery list shrinks. Generators optimize for variety; grocery bills reward repetition. Chicken twice plus one fish and one pantry dinner reads boring on paper and eats fine at the table. This is also where the cod went: one fish per week, cooked early.

Check 4: Give one night to leftovers

Delete one cooked dinner, double a midweek one, and let the extra half feed the following night. No generator I have used plans a leftover night unasked; they assume a cook shows up seven times. We doubled the roast chicken and Thursday became chicken wraps, which beat the cod it replaced on cost, effort, and fridge safety.

Check 5: Read the list against the pantry

Walk the grocery list against what you own before you shop, and cross off the staples the generator listed that are already on your shelf. Rice, oil, soy sauce, curry paste: machine lists re-buy the pantry every week. The grocery-list method covers the merge in detail; on a generated week the pantry pass alone cut nine items.

The same week after the checks

Fifteen minutes of editing later, this is what we shopped for and ate:

The score: four of seven generated dinners survived, two of them on different nights, one doubled. Three came out for the two standing slots and a leftover night. The grocery list fell from 41 items to 24. The machine did the imagining and the calendar did the editing, and that division of labor is the whole trick: ten seconds of generation, fifteen minutes of checks, one trip to the store.

Where FamilyPlate fits

The checks above work on any generator's output: a ChatGPT prompt, a web tool, or an app feature. The gap they patch is context, and context is also the fixable part. A generator that knows your meals needs less correcting than one that knows food in general, which is the case for keeping the plan and the list of meals you cook in the same place.

FamilyPlate's generate week is built on that idea: once your household library has meals in it, the generator plans from dinners you own rather than inventing strangers, and it reads your recent weeks so that Friday looks like your Fridays. On a brand-new account it writes a starter week from scratch, the mode this guide tested, and saves those dishes to your library so the next generation has something of yours to stand on. The library, the weekly plan, and the self-building grocery list are free on iPhone (iOS 17+); the generator and the other AI features are Pro, $4.99 a month or $39.99 a year after a 7-day free trial.