Calorie counting with Australian food data
Most calorie trackers were built in the United States and run on American food data. If you live here, that shows up the first time you try to log a Weet-Bix, a lamington, or a chicken breast from Coles.
The problem is not that the apps are bad
It is that a food database is a national object. What people eat, how it is labelled, which cuts of meat are sold, how much sugar goes into a standard yoghurt — all of it is local. An American database describes American food accurately and Australian food by accident.
Three things go wrong, and they compound.
1. The product is not there, so you log something similar
Search Vegemite, Milo, Weet-Bix, Tim Tams, or almost any Woolworths or Coles own-brand product in an app with a US-first database and you get either nothing or a guess. So you pick the nearest American equivalent, and now every day containing that food is quietly wrong.
2. The generic entries describe a different animal
This one is less obvious and worse, because it never looks like an error. “Beef mince” is not one food. Australian beef is predominantly grass-fed and sold by fat percentage; American beef is predominantly grain-fed and graded differently. Same words, different composition. The number that comes back looks perfectly reasonable and is not describing what is in your pan.
3. Crowdsourced entries are only as good as whoever typed them
Large apps let users add foods, which is how they built big databases quickly. It also means a meaningful share of entries are wrong: a decimal in the wrong place, a per-serve figure entered as per-100g, a barcode attached to the wrong product. You cannot tell by looking, and the app will not tell you either.
What Setpoint uses instead
The primary source is the Australian Food Composition Database (AFCD, formerly NUTTAB), published by Food Standards Australia New Zealand. It is the reference dataset behind Australian food labelling: foods measured in Australian laboratories, described the way they are sold here, with vitamin and mineral figures alongside the macros.
It is not the only source, because it cannot be. The order is:
| Source | Used for | Why it is in the list |
|---|---|---|
| FSANZ AFCD | Whole and generic foods — meat, produce, dairy, grains, cooked staples | Laboratory-measured, Australian, carries micronutrients |
| Open Food Facts | Barcoded packaged products, including Australian supermarket lines | The openly licensed database with real coverage of local packaged goods |
| USDA FoodData Central | Gaps the first two do not cover | Large, public domain, well curated — a fallback rather than a default |
What actually changes for you
- Micronutrients that mean something. AFCD carries vitamin and mineral figures for Australian foods, so iron, calcium, folate and the rest are read off local measurements rather than inferred.
- Generic foods that match what you bought. Log “beef mince, 5% fat” and the numbers describe Australian beef mince.
- Barcodes that resolve. Scanning a product from an Australian supermarket usually finds it, because Open Food Facts has genuine local coverage rather than a US catalogue with a few imports.
- Grams, and metric throughout. No cups, no ounces, no converting a recipe in your head.
Where this still falls short
Worth saying plainly, because every app claims a good database and most of them are describing the same three sources.
- Barcoded products are crowdsourced. Open Food Facts is contributed by the public, same as anyone else's user database. It is openly licensed and correctable, which is better than a closed one — but a packaged product entry is not a laboratory measurement and should not be treated as one.
- New and reformulated products lag. A line launched last month may not be in any database yet.
- AFCD is a snapshot. It is revised periodically, not continuously. Setpoint uses Release 3.
- Restaurant and takeaway food is an estimate wherever the chain does not publish figures, and often even where it does.
The part no database can fix
Even a perfect database only tells you what is in the food. It does not tell you how much of it you ate, and portion estimation is where most tracking error actually comes from. That is a separate problem, and the honest answer is a set of kitchen scales.
It is also why Setpoint keeps the two jobs apart. When you photograph a meal, a model identifies what is on the plate and estimates the portion — and that is all it does. Every calorie and macro figure is then looked up from the databases above. The AI never produces a number. It decides what you are eating; the database decides what that costs.
Attribution
Food composition data from the Australian Food Composition Database, © Food Standards Australia New Zealand, used under CC BY 4.0. Packaged product data from Open Food Facts under the Open Database License. Additional data from USDA FoodData Central (public domain).