A Chatbot Will Write You a Meal Plan. It Just Won't Remember Why.

You can get a meal plan out of a general AI chatbot in about ninety seconds. Tell it your age, what you are trying to achieve, what you will not eat, and it will produce seven days of food that looks entirely reasonable. It is genuinely useful, and pretending otherwise would be silly. So the fair question is not whether a chatbot can write a meal plan. It obviously can. The question is what happens on week two.

Because on week two you start again. The blank box does not know your ferritin was low in March, or that you are on a statin, or that you said you would not cook anything over thirty minutes, or that it suggested the same salmon and quinoa three weeks running. You can re-type all of it, and most people do for a fortnight before quietly stopping. What you are really buying from a chatbot is a very good one-off. What almost nobody has the patience to reconstruct by hand, every week, is the context.

Nutridly is built the other way round. Before it writes a single recipe it is working from twenty-seven things you told us once, and never have to say again.

Some of them are the obvious ones: your age, sex, height, weight and how active you are, which between them give a real energy and protein figure rather than a guessed one. Your goals, your conditions, your medications, your symptoms, your life stage. Your diet, your allergies, the foods you simply do not like, and the cuisines you would actually choose on a Friday night.

Then there are the ones you would never think to mention to a chatbot, and that turn out to matter a great deal. How often you eat oily fish. How much dairy you get. Whether you drink tea or coffee with your meals. How many hours a day you spend outdoors. How often you eat out. Which meals you actually eat, because a plan with a lunch in it is useless to somebody who does not eat lunch. None of that is small talk. Each one changes what the plan has to work harder at.

And then the practical ones, which are the difference between a plan you follow and a plan you admire: how confident you are in a kitchen, how long you will realistically spend cooking, roughly how you shop, whether you batch cook, and whether you would rather have maximum variety or maximum nutrient coverage when the two pull against each other.

On top of all that sits your blood work, if you have it, including the direction it has moved since your last test. That last part is worth pausing on. A chatbot can react to a number you paste in. It cannot tell you that the number is better than it was in March but still not where you want it, and therefore that the approach is working and should be pushed rather than changed.

Here is the part that is hardest to replicate in a chat window, and the reason we built our own engine rather than sitting a nice interface on top of a general one. Nutrition is not a list of good foods. It is a set of constraints that actively fight each other, and the whole job is resolving the fight.

Iron wants vitamin C alongside it. It does not want your tea or your dairy anywhere near it. Certain medications deplete certain nutrients, so those need replacing without going near anything that interacts. A recipe that nails your iron target is worthless if it contains something you are allergic to, and a plan that covers every vitamin beautifully is worthless if every meal takes fifty minutes and you told us thirty. Then there are the clashes that only show up across a whole week: the same dish twice in two days, lunch and dinner that are the same three ingredients in different bowls, a "new" recipe that is the thing you thumbed down a fortnight ago wearing a different name.

Our engine checks for every one of those before a plan reaches you. Allergies, diet and clinically excluded ingredients are absolute and get checked against every single recipe. Nutrient pairings get built into the same dish rather than left to chance, and nutrient conflicts are kept out of it. Prep time is verified against your answer rather than assumed. Repeats are checked against everything you have been served recently, and against the rest of this week's plan, so your Tuesday lunch and your Tuesday dinner are not quietly the same meal.

The last piece is the one that compounds. Mark off what you actually cooked, and the plan reads what you really ate rather than what was offered. A dish you keep skipping gets quietly deprioritised without being banned, because being out on Thursday is not the same as disliking something. A nutrient your logged meals keep missing gets more attention next week. Rate a recipe down and it does not come back, and neither do its close relatives. Every week the plan knows slightly more about you than the week before, and none of it required you to remember to say anything.

That is the honest comparison. A chatbot gives you a good plan built from what you can be bothered to type today. Nutridly gives you a plan built from everything you have ever told it, checked against constraints that a single prompt cannot hold in its head at once, and adjusted by what you actually did last week. The gap between those two is small on day one. By month three it is the whole thing.

None of which makes us a substitute for a doctor, and we will keep saying so. What it makes us is the difference between starting from scratch every Sunday and having something that already knows.