What a Meal Plan Generator Can and Cannot Know

The problem a generator is genuinely good at

Strip away the presentation and a meal plan generator is a small optimisation. It has a list of foods with a composition attached to each, a target for the day's energy, targets or ranges for protein, fat and carbohydrate, and a set of structural rules — three meals and two snacks, no repeated main dish, portion sizes that round to something you can serve. It then searches for quantities that satisfy all of that at once. Doing this by hand means recomputing four running totals every time you change one portion, which is exactly the kind of bookkeeping people abandon after a week. Software does not get bored.

So the fair way to judge one is by the quality of its constraints, not by whether it produces a plan. Any generator will produce a plan. The question is whether the energy target came from a named equation, whether the protein figure is anchored in body mass or pulled out of a percentage, whether the fat has a floor, and whether the food composition data is a real database rather than round numbers. Everything below is those constraints, written out.

The energy target, and where the uncertainty actually lives

The table above runs four published basal metabolic rate equations on one person. Mifflin-St Jeor gives 1,668 kcal, the 1919 Harris-Benedict original 1,718, its 1984 revision by Roza and Shizgal 1,720, and the Schofield equations adopted by the WHO 1,734. The spread is 66 kcal, four per cent. That is smaller than most people assume, and it is worth saying plainly: arguing about which BMR equation to use is arguing about a rounding error.

The uncertainty lives somewhere else, in two places. The first is the activity multiplier. The Institute of Medicine's categories run from 1.0 to 1.39 for sedentary, 1.4 to 1.59 for low active, 1.6 to 1.89 for active and 1.9 to 2.5 for very active. Applying 1.4 and 1.9 to the same Mifflin-St Jeor figure gives 2,335 kcal and 3,169 kcal — 834 kcal apart, more than twelve times the disagreement between the four equations. Nobody can tell you which of those numbers describes you, and self-reported activity is famously optimistic. The multiplier is a guess wearing a decimal point.

The second place is the person. Any equation is a regression through a sample, and you are one point around that line. The systematic review that made Mifflin-St Jeor the default found it landed within ten per cent of measured resting metabolism in about 82% of non-obese adults and 70% of obese adults — meaning roughly one non-obese person in five, and nearly one obese person in three, is further out than that. The 2023 US reference intakes for energy state the same thing in kilocalories: the standard error of their predicted requirement is 342 kcal a day for men aged 19 and over, and 241 kcal for women, with about 68% of people falling inside one such error. That is the real width of the target.

Protein in grams per kilogram, fat with a floor

Expressing protein as a percentage of energy is the single most common defect in a generated plan, because it makes the protein target move when the energy target moves — which is exactly backwards. Protein requirement tracks lean mass, not appetite. Cut a plan from 2,700 to 1,900 kcal at a fixed 18% and the protein falls by nearly 40 g even though the body it is repairing has not changed. Anchor it in grams per kilogram instead. The reference value is 0.8 g/kg in the US intakes and 0.83 g/kg in EFSA's; sports-nutrition consensus for people in training is 1.2 to 2.0 g/kg per day, spread across the day in doses of 0.25 to 0.3 g/kg every three to five hours.

Fat needs a floor rather than a target, and the reason is not energy. Both the US and EU reference frameworks put total fat at 20 to 35% of energy, and the 2016 sports-nutrition position stand is explicit about why the bottom of that range matters: athletes are discouraged from chronic fat intakes below 20% of energy because the loss of dietary variety that usually accompanies such a restriction reduces intake of fat-soluble vitamins and essential fatty acids, especially the omega-3s. Those essentials have their own numbers. EFSA sets adequate intakes of 4% of energy for linoleic acid and 0.5% for alpha-linolenic acid; the US intakes give 17 g and 12 g of linoleic acid a day for adult men and women, with 1.6 g and 1.1 g of alpha-linolenic. At a 2,700 kcal target, EFSA's 4% is 12 g of linoleic acid and its 0.5% is 1.5 g of alpha-linolenic — small numbers that a very low-fat plan can still miss.

Carbohydrate then takes what is left, and the arithmetic is one line. Take a 2,700 kcal target for the 165 lb man in the table, whose 75 kg is what the g/kg targets need. Protein at 1.6 g/kg is 120 g, or 480 kcal. Fat at 30% of energy is 810 kcal, or 90 g. Carbohydrate is 2,700 − 480 − 810 = 1,410 kcal, which is 352 g at 4 kcal per gram — 4.7 g per kilogram, and 52% of energy. Both checks pass: it sits inside the 45 to 65% acceptable range and near the moderate training band. Move fat down to 25% and carbohydrate rises to 386 g, 5.1 g/kg. That is the whole model, and it is worth doing once by hand so you can tell when a generator has produced something odd.

What no generator can see

Micronutrient adequacy is the first blind spot, and it is a real one. Hitting four macronutrient numbers says nothing about iron, calcium, iodine, vitamin D, folate, B12 or zinc, and adequacy for those is judged across a week rather than a day. A plan can be perfect on energy and macros and still be short of iodine every day for a month. Generators that carry only energy and macro data cannot check this at all, and the ones that carry a full composition table can only check it against the foods in their own list.

Then come the things that are not nutritional at all. A generator does not know your allergies unless you tell it, and telling it removes foods without telling it what to put back. It does not know that a medication you take interacts with grapefruit, or with vitamin K, or with a high-protein load. It does not know that you cannot digest lactose, that beans give you trouble, that you have no oven, that you cook for four people with different tastes, or that the salmon it has cheerfully put in three lunches costs more than the rest of the week combined. It does not know what actually makes you feel full — and satiety per calorie varies enormously between foods that are identical on a macro sheet.

Adherence is the variable that beats all the others

Consider two plans. The first is optimal: the energy target is exactly right, the macros are exactly where you wanted them, and you follow it for nine days before quitting. The second is a little loose — the protein is 15 g under, one dinner repeats twice a week, the fat sits at the top of the range — and you follow it at eighty per cent for six months. The second plan wins by a distance that the first cannot make up, because the deficit or surplus you actually run is the product of the plan and the fraction of it you execute. A 500 kcal deficit followed at 80% delivers 400 kcal; a 700 kcal deficit followed for nine days delivers nothing at all.

This is why the most useful thing a generator can do is produce a plan you find boringly easy — short shopping lists, repeated meals, foods you already buy, portions that scale to whoever else is eating. Optimising the macros to the last gram while ignoring whether anyone can live inside the result is solving the wrong constraint. And it is why the last step of the how-to above matters more than the first six: after two or three weeks, correct the energy target from what the scale actually did. The equation predicted your requirement with a standard error of a few hundred kilocalories; your own body mass over a fortnight measures it.

Frequently asked questions

Which BMR equation should the generator use?

Mifflin-St Jeor, on the strength of the 2005 systematic review that found it landed within ten per cent of measured resting metabolism more often than Harris-Benedict, Owen or the WHO equations, with the narrowest error range. But do not expect the choice to matter much: across the four equations in the table the spread is 66 kcal, about four per cent. Choosing the activity multiplier moves the answer twelve times further.

Why anchor protein in grams per kilogram rather than as a percentage?

Because the requirement follows body tissue, not appetite. A percentage makes the protein target shrink whenever the energy target shrinks, which is exactly when you least want it to — during a deficit, protein is what protects lean mass. Take a 165 lb (75 kg) person: 1.6 g/kg is 120 g whether the day is 2,700 kcal or 1,900. Written as 18% of energy, the same person gets 121 g on the higher day and 86 g on the lower one, for no physiological reason at all.

Can a generated plan replace a dietitian?

No, and the reason is structural rather than a matter of quality. A dietitian takes a history, sees your bloods, knows your medications, notices that you have been losing weight without trying, and can change the plan when your life changes. A generator has four inputs and a food list. It is a good arithmetic assistant to a decision that has already been made properly — and it is the wrong tool entirely if you have a diagnosed condition, are pregnant, are managing an eating disorder, or are planning a diet for a child.

The plan says 2,700 kcal but I am not losing weight. Which number is wrong?

Most likely the activity multiplier, and possibly the portion sizes. The 2023 US reference intakes put the standard error of a predicted energy requirement at 342 kcal a day for adult men and 241 kcal for adult women, so being a few hundred kilocalories off is the expected case, not a failure. The fix is not a different equation. Keep the plan constant for two or three weeks, record body mass, and shift the target by whatever the trend says — that measurement beats any prediction.

Does the plan need to hit the targets every single day?

No, and for micronutrients it is the wrong question. Adequacy for vitamins and minerals is assessed across several days to a week, which is why a single day that is short of one nutrient means very little and a month of the same shortfall means a lot. Energy averages out over weeks too. Protein is the partial exception: distributing it across the day in doses of roughly 0.25 to 0.3 g per kilogram every three to five hours is what the sports-nutrition evidence supports, so protein timing is worth more attention than the exact daily totals of anything else.