ChatGPT analyses a meal, Happy Assistant tracks your nutrition
Photographing a plate and asking "how many calories?" works very well with a generalist AI. The trouble starts with the next question: "and over the month?". Here is what changes when a meal analysis becomes a history entry instead of a message.
An ephemeral message versus a journal entry
In a generalist chat, a meal estimate lives for one exchange. Nothing is totalled, nothing is comparable, and finding last Tuesday's lunch means scrolling by hand.
- Each meal becomes a dated entry: meal type, detected foods, estimated quantities.
- Values are stored separately: calories, protein, carbs, sugars, added sugars, fat, saturated fat, fibre, sodium.
- Your restrictions trigger explicit alerts (allergens, sugar, salt).
- Manual correction stays possible: a photo estimate is a starting point, not a verdict.
Browsable history: day, week, month
Tracking only matters if you can go back in time effortlessly. Three views, each navigable period by period:
- Day: meals in order, daily total, gaps against your personalised targets.
- Week: averages over the last seven complete days, quality score, plant diversity.
- Month: calendar-month averages, trend and consistency.
Incomplete days are excluded from consolidated averages, with a message saying so: otherwise a week would look excellent simply because you only logged today's breakfast.
Targets computed for you
Requirements are not fixed values copied from a generic table. They are estimated from your profile (age, gender, weight if you provide it) and from your real activity over the last fourteen days taken from the sport journal. A heavy week raises the calorie and protein target, a quiet week lowers it.
- Calorie target adjusted to your estimated expenditure.
- Protein expressed per kilo of body weight, increased when you train a lot.
- Added sugar, salt and saturated fat thresholds used for alerts.
What the assistant can then answer
- "Did I eat enough protein this week?"
- "What did I eat the evening before Sunday's race?"
- "Have I had too much added sugar this month?"
- "Compare this month with last month."
These answers come from your data, not from a statistical average: the assistant reads your journal before replying.
Frequently asked questions
Is photo estimation reliable?
It is an order of magnitude, not a weighing. It is enough to follow a trend over several weeks, and you can correct quantities and foods by hand.
Can I log a meal without a photo?
Yes: a written or dictated description is enough, and the analysis works the same way.
Why does the week exclude today?
Because a day in progress is incomplete: including it would bias the average down in the morning and up in the evening. Averages cover the last seven complete days.