AI Meal Plan Generator

How an AI Meal Plan Generator Actually Works

Strip away the marketing and every AI meal plan generator does the same three jobs: decide how much you should eat, decide how to split it between protein, carbohydrate and fat, and then choose foods and portions that land on those numbers. The differences between tools are almost entirely in where the first number comes from and how honest the third step is.

Step 1: the calorie target

Most generators start with a formula. Mifflin-St Jeor estimates your resting metabolic rate from weight, height, age and sex, then multiplies by an activity factor you pick from a list. It is a reasonable population estimate and it is wrong for you by some unknown amount, typically ±10 percent before the activity multiplier adds its own error. A "moderately active" person can burn 400 calories a day more or less than the label implies depending on whether this week's sessions were easy or brutal.

Step 2: the macro split

The split is set by the goal. The generator above uses 40/30/30 for fat loss (protein high enough to protect muscle in a deficit), 30/40/30 for maintenance and 25/50/25 for performance, where half of energy from carbohydrate keeps glycogen topped up for back-to-back sessions. When your weight is known it also enforces a protein floor of 1.6 g per kilogram (1.4 g/kg for performance), which is the low end of the range most sports-nutrition guidance uses for preserving lean mass.

Step 3: foods and portions

This is where many tools cheat. They pick recipes with fixed serving sizes and report whatever the totals happen to be, or they quietly pad the day with a "snack" to close the gap. The preview on this page does the harder thing. Each meal has a small template of foods, and a solver adjusts the grams of each until the meal's protein, carbs and fat are as close as possible to its share of the daily target, with sensible bounds so you never get 900 g of chicken or 2 g of rice. It then scales the meal to hit its calorie share exactly, rounds to whole grams (eggs to whole eggs) and corrects the last few calories on the least calorie-dense item. Across 1,200 to 4,000 kcal, every diet style and every goal, the day lands within 1 percent of target; we tested 432 combinations before publishing the page.

Plait's full plan uses a language model for what language models are good at — writing varied, appealing meals from a large recipe set and respecting the allergies and dislikes you set once — and the same kind of deterministic arithmetic for what they are bad at, which is making numbers add up. That separation matters. An AI that free-associates a day of food and then guesses the macros will be off by hundreds of calories with total confidence.

Why Wearable Data Beats a Static Calculator

A calorie target from a formula is a photograph. Your actual energy expenditure is a film. WHOOP, Garmin and Apple Watch measure heart rate all day and convert it, with your profile, into a daily calorie estimate that moves when your training moves. That alone removes the activity-multiplier guess. But the more interesting signals are the ones a static calculator cannot see at all:

• Recovery / Body Battery. WHOOP's recovery score and Garmin's Body Battery both summarise how your body responded to yesterday, via heart rate variability, resting heart rate, sleep and stress. A well-recovered body stores glycogen efficiently and is more likely to train hard, so carbohydrate is useful fuel that day. A depleted body benefits more from holding protein steady and letting carbohydrate ease off. That is exactly what the slider above mimics, and what Plait does automatically each morning with your real number.
• Sleep. Short or fragmented sleep raises appetite and lowers glucose tolerance the next day. A plan that knows you slept five hours can lean toward protein-forward meals and an earlier, lighter dinner rather than the same day it would have written after nine hours.
• Strain and training load. Yesterday's long ride does not just add calories to yesterday; it raises the carbohydrate that today's session will run on. A rolling view of strain lets the plan set a higher baseline through a hard block and drop it in a deload week without you noticing the arithmetic.
• Steps and stress as cross-checks. When step count implies more walking energy than active calories show, something is off — a stale profile weight or a loose strap. A wearable-aware plan can flag that instead of quietly under-feeding you for a month.

For Garmin users the mechanics are spelled out in our Garmin meal plan guide, including a sample day for charged, moderate and drained Body Battery mornings. WHOOP users should read the WHOOP nutrition plan guide, which covers strain-based carb cycling and recovery-zone eating.

Generic AI Meal Plan Generators vs a Wearable-Aware One

None of this makes a generic generator useless. If you do not own a wearable, a formula-based day like the one above is still far better than guessing, and the food itself is the same. The difference is what happens on day 12, when your training has changed and your plan has not.

How to Get the Most From the Free Generator

1. Get the calorie number from your device if you have one. Add resting and active calories from Garmin Connect or WHOOP (7-day average), apply your goal offset, and type the result into the target field. The Garmin macro calculator and WHOOP macro calculator do this for you.
2. Run it at two recovery values. Generate the same day at 85 and at 30 and look at what moves. That difference — a few hundred grams of carbohydrate swapped for fat — is the adjustment a wearable-aware plan makes for you automatically.
3. Treat portions as cooked weights. Rice, pasta, quinoa, potato, lentils, chicken and beef are all listed as cooked. Oats are dry. Raw chicken loses roughly a quarter of its weight in cooking, so buy accordingly.
4. Check reality after two weeks. If your weight trend disagrees with your goal, adjust the target by 150–200 kcal and regenerate. No formula or watch is accurate enough to skip this step.
5. Do not go below 1,200 kcal. The tool will refuse, and so should you without professional supervision.

From a One-Day Preview to an Adaptive 7-Day Plan

The generator on this page is deliberately simple: one day, a fixed menu per diet style, no account. Plait's product is the same idea taken seriously. Connect your device and Plait builds a full week from your measured calories, with three or four meals a day, gram portions, a consolidated grocery list and per-meal macros. Each morning it reads your recovery or Body Battery, sleep and yesterday's strain, and re-fits the day. The dashboard and your first plan are free. Pro adds daily plan refreshes, Today's Focus recommendations, meal swaps, AI meal images, body-composition projections and PDF export for $5.99 a month (3-day trial) or $29.99 once.

Further Reading