"Generated with artificial intelligence" has become a label that says nothing. It can mean a model wrote your entire menu, or that somebody bolted a chatbot onto a spreadsheet from 2014.
So here is the concrete version: what happens when SinExcusas generates a nutrition plan, what gets checked before you see it, and what it will never do.
What you give it
The plan does not come out of nowhere. It comes out of what you answered when you set up your account:
- Your physical data: weight, height, age and sex, which set your calories and macros.
- Your goal: lose fat, maintain or gain muscle, at the intensity you picked.
- Your activity level, which multiplies your daily burn.
- How many meals you want a day: two to five, with optional snacks.
- What you cannot or will not eat: allergies, restrictions such as vegetarian, vegan, gluten-free or lactose-free, and your favorite foods.
If any of that is wrong, the plan will be wrong. It is the most boring part and the one that changes the outcome most.
What the AI builds
Generation has three phases, and only the first one is "creative".
Phase 1: the menu
A language model writes the full week: seven days, with their meals, their ingredients and the gram amount of each. It does not pick from a closed recipe catalog: it composes meals inside the calories and macros each one gets, based on how you split your day.
It also gets variety rules, so the plan does not turn into chicken and rice seven days running and two consecutive weeks do not come out identical.
If what it returns does not match the expected structure, it is thrown away and generated again.
Phase 2: the numbers
This is where the plan stops being nice prose. Every ingredient is valued nutritionally against food databases, and when no database has a food, the AI is asked for an estimate and the ingredient is flagged as estimated.
That last case matters and we do not hide it: some food is in no database, and there the number is a reasonable approximation, not a measurement.
Phase 3: portion adjustment
With every ingredient valued, the system adds up the real day and compares it with your target. It almost never matches on the first pass, so it adjusts gram amounts in an orderly way and within capped ranges, so you do not end up with 900 grams of rice.
The goal is to land inside a narrow margin of your numbers, not to hit them exactly: in real nutrition, a few points of difference change nothing.
What gets checked before you see it
Before the plan lands on your phone it goes through several automatic checks:
- That all seven days are there, and that each meal has a name, ingredients and quantities greater than zero. A plan with blank amounts is rejected; nothing is filled in by guesswork.
- That there are no allergens. Every allergy you declared is checked against the plan’s ingredients. If something fails the filter, the plan is not published and gets regenerated.
- That your restrictions hold. A vegan plan with dairy, or a vegetarian one with chicken, is rejected no matter how insistent the model is.
- That the macros add up once every ingredient has been valued.
None of this is foolproof: it is a filter on food names, and an oddly written ingredient can slip through. If you are allergic to something, read your plan before you eat it. The app is a tool, not an insurance policy.
A new one every week
The plan has an expiry date: it covers the week. Before the next one starts, a new plan is generated with whatever numbers you have at that moment, and if a plan expires without being renewed, it regenerates on its own.
What makes those numbers change? You logging a different weight, editing your profile or changing your goal. At that point your calories and macros are recalculated, and the next week’s plan comes out with the new figures. That is the loop: you log, the math moves, the plan follows.
What it does NOT do
This is the part almost nobody writes.
It diagnoses nothing. It does not know whether you have insulin resistance, a thyroid problem or an iron deficiency, and it will not spot any of that from your food. It does not treat any condition either.
It does not replace a health professional. If you are pregnant, breastfeeding, diagnosed with something or under treatment, your plan needs review by someone who can see your medical history. No app can do that.
It does not know what you actually ate. The plan is a proposal. If you do not log, the system has no way of knowing whether you followed it, and it adjusts nothing on its own. The app does not punish you or "detect" a skipped plan: the only thing that moves your numbers is the weight you record.
It does not guess what you did not tell it. An allergy you did not declare does not exist for the system. An old weight means it calculates with that old weight.
It guarantees no results. No number can promise you kilos by a date. What a good plan gives you is a coherent starting point and something to follow without deciding it every morning.
How to get the most out of it
- Keep your data current. Log your weight regularly: it is the only thing that makes the system recalculate.
- Declare everything you do not eat, dislikes included. They are respected like an allergy, and you avoid plans you were never going to follow.
- Use the shopping list. The app collects the week’s ingredients, adds up repeated amounts and groups them by aisle. A plan without the shopping done is a plan abandoned by Tuesday.
- Change what does not fit. If a day does not work, swap it and log it. A plan followed 80 % of the time beats a perfect plan nobody follows.
This article is general nutrition and wellbeing information. It is not medical advice or an individual plan. If you have a health condition, are pregnant or breastfeeding, or are under treatment, talk to your doctor or a clinical dietitian before changing how you eat.
