Gains:
- Ability to monitor BMH, TDEE and macro distribution with basic formulas (Mifflin-St Jeor, 4/4/9) and independently verify the artificial intelligence calculation
- Ability to catch calculation errors by checking critical assumptions such as activity coefficient, raw/cooked distinction and protein g/kg
- Being able to practice the habit of multiplying back macro grams to get the total on target and not falling for false certainty.
The numerical framework of the nutrition plan is based on two calculations: the person's daily energy needs (how many calories) and the distribution of this energy among macronutrients (how much carbohydrates, proteins, fats). If these two calculations are wrong, the entire menu, recipe and follow-up will be based on the wrong foundation. Artificial intelligence makes these calculations in seconds; But it is precisely this speed that is the source of danger. Because while AI can apply a formula correctly, it can also assume a coefficient incorrectly and produce an incorrect result with the same confidence. The aim of this unit is to make it a reflex to use AI as a "spreadsheet generator" and independently verify each result.
Fundamentals of energy calculation
Daily energy needs come from two components. The first is BMR — Basal Metabolic Rate, that is, the energy the body spends at rest (just to breathe, heart beat, produce heat). The second is to multiply this value by the activity coefficient (English activity factor or PAL - Physical Activity Level); This coefficient varies depending on the person's mobility, between approximately 1.2 (sedentary) and 1.9 (very heavy physical work). The result of the multiplication is TDEE — Total Daily Energy Expenditure.
The most commonly used formula for BMH is Mifflin-St Jeor:
- Male: BMH = 10 × weight (kg) + 6.25 × height (cm) − 5 × age + 5
- Female: BMH = 10 × weight (kg) + 6.25 × height (cm) − 5 × age − 161
For example, for a 78 kg, 165 cm, 38-year-old woman: BMH = 10×78 + 6.25×165 − 5×38 − 161 = 780 + 1031.25 − 190 − 161 = 1,460 kcal. If slightly active (1,375) is considered, TDEE ≈ 2,008 kcal. To lose weight, a reasonable deficit (e.g. 500 kcal) is deducted from here → ~1,500 kcal target. Being able to follow these steps manually is a prerequisite for being able to control the result of AI.
Macro distribution
Once the calorie goal is determined, we break it down into macros. The energy densities of macros are fixed and you should know them by heart: 1 gram of carbohydrates ≈ 4 kcal, 1 gram of protein ≈ 4 kcal, 1 gram of fat ≈ 9 kcal (alcohol ≈ 7 kcal). An example of a distribution: 40% carbohydrate / 30% protein / 30% fat distribution of 1,500 kcal → carbohydrate 600 kcal ÷ 4 = 150 g; protein 450 ÷ 4 = 112.5 g; fat 450 ÷ 9 = 50 g. Multiplying the sum back and seeing that it reaches 1,500 is the simplest and most powerful step of verification.
The following table shows rough distribution ranges for common targets (these are a starting point, adapted to the client):
Target/situation
carbohydrate
protein
oil
note
General healthy diet
45-55%
15-20%
25-35%
balanced
Weight loss (sparing protein)
35-45%
25-30%
25-35%
High in protein for satiety
Muscle gain / athlete
45-55%
1.6-2.2% g/kg
20-30%
Protein is calculated in g/kg
insulin resistance
40-45%
20-25%
30-35%
Priority of fibrous carbohydrates
Note: Protein needs are often calculated in grams per body weight (g/kg), not in percentage; because an athlete's protein need depends on his muscle mass, not his total calories. AI can sometimes confuse this distinction and give insufficient protein on a percentage basis. Recalculate.
AI account verification protocol
- Check the inputs: Which formula, which activity coefficient, which gender row did the AI use? Ask them to write these clearly.
- Request intermediate results: Show BMH and TDEE separately, not just "1,500 kcal". Let the steps be visible, not a black box.
- Multiply back: Multiply the macro grams by 4/4/9 to verify that the total meets the target calories.
- Logical filter: Is the outcome reasonable for the client? If an active man who needs 1,900 kcal gets 1,200 kcal, there is a mistake.
- Confirmation by independent tool: In critical cases, calculate a second time with a reliable calculation tool or manually.
three mini cases
Case 1 — Coefficient error. A dietitian has the AI perform calculations for a client who works on his feet for 8 hours a day and trains 4 days a week. The AI assumes an activity coefficient of 1.2 (no movement) and finds the TDEE to be 1,700 kcal. The dietician says "the client is very active" and recalculates with a coefficient of 1.725; actual TDEE comes out to ~2,450 kcal. The 750 kcal difference would inadvertently lead to an overly restricted plan. The verification prevented a month of weakness and muscle wasting.
Case 2 — The macro total doesn't add up. In another draft of the plan, AI gives "220 g carbohydrates, 150 g protein, 90 g fat" for 2,000 kcal. The dietitian multiplies it back: 220×4 + 150×4 + 90×9 = 880 + 600 + 810 = 2,290 kcal. The total exceeds the target by 290 kcal. AI macros produced inconsistently. The dietitian corrects the values to meet the target.
Case 3 — Correct use. A dietitian first manually determines the protein requirement for a 90 kg strength athlete (2 g/kg = 180 g protein = 720 kcal), then gives the AI clear instructions to "distribute the remaining calories between carbohydrates and fat in this ratio." The AI distributes, the dietitian verifies the total by multiplying it back. AI did the arithmetic, the clinical decision (2 g/kg protein) was left to the human.
Copiable prompt templates
ENERGY ACCOUNT TEMPLATE (steps visible)Your role: assistant drafting accounts. The decision is up to the dietitian. Client (anonymous): [gender, age, height cm, weight kg, activity level]. Calculate BMH with the Mifflin-St Jeor formula, write CLEARLY the activity coefficient you used, show TDEE. Goal: [weight loss/maintenance].Show each intermediate step on a separate line so I can verify it.
MACRO DISTRIBUTION TEMPLATETarget calories: [X] kcal. Apply the following distribution: carbohydrate __%, protein __% (or __ g/kg), fat __%. Calculate the GRAMS of each macro, then multiply it back by 4/4/9 to show that the total adds up to [X]. If not, fit the values to the target.
VERIFICATION / BACKPRODUCT TEMPLATECheck these macro values: [K g carbohydrate, P g protein,Y g fat]. Multiply each by 4/4/9 to find the total calories and compare with the target [X] kcal. If there is a difference, state it as a percentage and explain where it is not consistent.
PROTEIN PRIORITY TEMPLATEClient: [weight] kg, goal [muscle maintenance/gain]. First calculate the protein in grams based on g/kg. Then subtract the calories from this protein, distribute the REMAINING calories to carbohydrates and fat [ratio]. Show steps.
Weak prompt / Strong prompt
Weak prompt:
Tell me how many calories this person should eat: woman, 78 kg.
Height, age, activity missing; AI makes up by guessing and does not show what assumption it is making. You cannot verify.
Powerful prompt:
Your role: account drafting assistant. Client: female, 38 years old, 165 cm, 78 kg, slightly active (coefficient 1.375). Calculate BMH, show TDEE, achieve your weight loss goal with a 500 kcal deficit with Mifflin-St Jeor. Show each number you use on a separate line.
The second prompt contains all the inputs, formula, coefficient and transparent step request; The output can be inspected line by line.
Common mistakes
- Not verifying the activity coefficient: The most common and expensive mistake; The coefficient should match the real life of the client.
- Not multiplying the macro total back: AI's grams often miss the mark; 30 seconds of backslash captures this.
- Protein percentage/gram confusion: In athletic and clinical situations, protein is calculated in g/kg, not percentage.
- Relying on the black box: Accepting an output that does not show intermediate steps. Steps should appear.
- Illusion of hyperprecision: The AI's decimal output like "1,847.3 kcal" looks scientific, but the margin of prediction is ±10%; Don't be fooled by false certainty.
In summary
Energy and macro calculations are the digital skeleton of the plan, and AI produces this skeleton quickly. But speed is no substitute for verification. Know the formulas for BMH, TDEE, and macro distribution yourself so you can control the output of the AI. Ask for intermediate steps to be visible in each account, match the activity coefficient with the client's reality, multiply the macros back by 4/4/9, pass the result through the logic filter. Don't be fooled by false precision (numbers with decimals); The nutrition calculation is an estimate with a margin of guesswork. AI speeds up arithmetic; Clinical judgment and final approval are yours.
Application task
Select a client profile (gender, age, height, weight, activity) and manually calculate BMR YOURSELF first with Mifflin-St Jeor. Then give the same profile to the AI with the "energy account template". Compare the two results; If there is a difference, find where it originates (coefficient, rounding, formula). Then verify that the total adds up by dividing the target calories by a macro distribution and multiplying back.
checklist
- [ ] AI clearly wrote which formula and activity coefficient he used.
- [ ] I saw the interim results of BMH and TDEE separately.
- [ ] I multiplied the macro grams back by 4/4/9 to verify that the total was on target.
- [ ] I calculated the protein requirement in g/kg when necessary.
- [ ] I filtered the result through logic according to the real life of the client.
- [ ] I am not fooled by false certainty; I considered the result as a prediction range.