The best way to track calories is the one you are still using in a month. A consistent method can be calibrated against the bathroom scale, and an abandoned one cannot. I compared six methods on two tests: does the number mean anything, and are you still logging on day 30.
TL;DR
- Every method fails one of two tests: the number is meaningless, or you stop.
- The database is rarely the problem. The person logging is.
- A consistent bias can be corrected against the scale. An abandoned log can't. That is why consistency beats precision.
- I've used two of the six methods first-hand, and marked the other four as reasoning.
- Most of what a Malaysian eats in a week has no barcode, which is where scanner apps run out.
What is the best way to track calories?
Run every method past two questions.
Test one: does the number mean anything? The database behind the app is close enough. Evenepoel and colleagues compared MyFitnessPal's database with the Belgian database Nubel: 50 participants, each keeping a four-day record twice. After extreme values were removed, MyFitnessPal's energy figure was 1.3% high and its protein 7.8% low (JMIR, 2020). People do more damage to their own records. In a 1992 study of ten people who reported they could not lose weight on under 1,200 kcal a day, the group under-reported what they ate by 47% and over-reported exercise by 51% (Lichtman et al., NEJM). It is a small study of an unusual group, and I'm not claiming it describes you. My point is narrower: the big errors come from the person, not the database.
Test two: are you still logging on day 30? A method that is precise for a week and abandoned in a month produces a week of data. Most comparisons score only test one.
Which of these six methods can I actually speak to?
Two of them.
I tracked for about four weeks by describing each day to an AI chatbot: free text, often shorthand, often naming Malaysian dishes, sometimes a weight, sometimes two days in one message. Then I built the quick-log in Habitus Mind to reproduce that. Five things carried over, and they are the useful part of those four weeks:
- Estimates lean high, as a deliberate accountability choice.
- Oily, sauced or vaguely portioned food is flagged as uncertain, not given false precision.
- A weight jump is read against the previous days (a heavy, salty or late meal), not as fat gained overnight.
- A repeating pattern is named plainly, such as a third over-target day this week.
- The tone stays honest: a mediocre day is not inflated.
One thing I left out on purpose: memory. The chat could take “breakfast now, dinner later”. The quick-log is one form, one parse and one review, and it forgets. It is not a chat.
I have not lived with the other four: barcode-scanner apps (MyFitnessPal and Lose It are two), photo apps (Cal AI is one), the food scale with a database, and pen and paper. For those, what follows is reasoning about how the method works, not a report from someone who used it. I've marked them that way in the second table.
How accurate is each method?
Table A. What each method measures, its typical error, and where it breaks. Where I have no published figure, I say so.
- Barcode-scanner app
- What it measures: A packet's nutrition panel, or a database entry you pick by search
- Typical error: Small in the database (Evenepoel: energy +1.3%, protein −7.8%). The portion you enter is what it can't fix
- Where it breaks: Anything with no barcode
- Photo or AI camera app
- What it measures: A guess at the dish and its size
- Typical error: No published figure I can vouch for
- Where it breaks: What a camera can't see: oil, sugar, what is under the rice
- Food scale plus database
- What it measures: Grams of each ingredient, via a database
- Typical error: Smallest on paper: the portion is measured, not guessed
- Where it breaks: Meals someone else cooked or plated
- Pen and paper, or a spreadsheet, with portion estimates
- What it measures: What you write down, portions judged by eye
- Typical error: Set by your own eye. The documented failure is under-reporting
- Where it breaks: The meals you don't write down
- No counting: plate method plus a weekly weigh-in
- What it measures: Plate proportions, checked by the scale
- Typical error: None to quote: it isn't a calorie count
- Where it breaks: It tells you nothing about a single day
- Describe the day in text, AI estimate
- What it measures: A model's estimate from your description
- Typical error: Not validated against a measured meal, so I won't give one. Its direction is known: high on food, low on exercise burn
- Where it breaks: Only as good as what you typed
Which method are you still using on day 30?
Table B. The friction each method adds per meal, and what usually kills it.
- Barcode-scanner app (reasoning only)
- Friction per meal: Low for packets, high for everything else
- What kills it: A week of meals with no barcode
- Photo or AI camera app (reasoning only)
- Friction per meal: One photo, then corrections
- What kills it: The same wrong guess, corrected every meal
- Food scale plus database (reasoning only)
- Friction per meal: Highest: weigh, look up, enter
- What kills it: The first restaurant or shared meal
- Pen and paper, or a spreadsheet (reasoning only)
- Friction per meal: Low, if you allow guessing
- What kills it: Losing the habit of opening it
- No counting: plate method plus a weekly weigh-in (reasoning only)
- Friction per meal: None
- What kills it: Little, but nothing is measured either
- Describe the day in text, AI estimate (used it)
- Friction per meal: None at the meal; one text box a day
- What kills it: Forgetting what you ate by evening
For a Malaysian, the first row decides it. Most of what we eat in a week has no barcode and no label: mamak, hawker, kopitiam, economy rice. Someone else scoops the plate by eye, and it differs by stall. A scanner app falls back to a text search of a crowd-sourced database: a stranger's estimate of a different stall's portion.
The dish name doesn't settle the number either. The Ministry of Health's Malaysian Food Composition Database (MyFCD) lists nasi lemak at 165 kcal per 100 g and a medium plate at 306 g: about 505 kcal for the plate alone. Its ayam goreng entry lists 283 kcal and its beef rendang entry 253 kcal. So “nasi lemak” can differ by hundreds of kilocalories depending on what sits beside it. (Those are the 1997 tables MyFCD hosts.)
Drinks are the hidden line: teh tarik, kopi with condensed milk, Milo ais and 100 Plus arrive sweet by default, and “kurang manis” is a decision you make out loud. Meals are shared too, from dim sum to steamboat, so nobody knows their own portion. That is the case for estimating consistently rather than pretending to count.
Packaged food is the exception: Malaysian packets carry a nutrition panel and a barcode, and scanning works.
The plate method has a local version. The Ministry of Health's Malaysian Healthy Plate, “Suku Suku Separuh”, introduced in 2016, is a quarter grains, a quarter protein and half fruit and vegetables. The 2019 National Health and Morbidity Survey found that 79.6% of Malaysian adults had never heard of it (Tan et al., Nutrients, 2023). It belongs in the no-counting row.
Why does consistency beat precision?
Because the scale is the check, and only a consistent log can be checked.
Here is the method. Log a consistent intake for 14 days while weighing yourself daily. If your weight average holds, that intake is your maintenance calories, in the units of your log. If it moves, the conventional rule of thumb is that a kilogram of change is worth roughly 7,700 kcal. It is a rule of thumb, not a law, but it turns drift into a daily correction. As arithmetic only: a 0.5 kg drift over the 14 days is 3,850 kcal, about 275 kcal a day. How long the method really takes is its own post. To run it on your own numbers, use the from-your-log calculator.
A steady bias barely matters inside this method. If your log runs high every day, that is a constant, and the 14 days absorb it. A log that is accurate on Monday and abandoned by Friday absorbs nothing.
That is why the quick-log leans high on purpose. I don't claim its estimates are accurate, and I haven't validated them against a weighed meal. The aim is a number that is consistent and leans in a known direction. Here is one entry, run through it for this article:
Weight 78.4kg. Breakfast 2 roti canai with dhal and a teh tarik at the mamak. Lunch roasted chicken rice with soup and a kopi o ais. Snack one karipap around 4. Dinner economy rice: rice, one fried chicken thigh, stir-fried kangkung, two pieces of tofu, and a can of 100 Plus. Walked 30 min around the taman after dinner.
That is one text box, about 60 words, submitted once and reviewed once. Here is what it returned, item by item:
- 2 roti canai with dhal: 620 kcal, 18 g protein, flagged uncertain
- Teh tarik: 180 kcal, 4 g protein, flagged uncertain
- Roasted chicken rice with soup: 680 kcal, 32 g protein, flagged uncertain
- Kopi o ais: 45 kcal, 1 g protein, not flagged
- 1 karipap: 140 kcal, 4 g protein, flagged uncertain
- Economy rice with fried chicken thigh, kangkung and 2 pieces of tofu: 520 kcal, 28 g protein, flagged uncertain
- 100 Plus, 330 ml can: 138 kcal, 0 g protein, not flagged
- The 30-minute walk: 120 kcal burned
That comes to 2,323 kcal and 87 g of protein. I ran the same text twice; the first run came back at 2,398 kcal, 3% higher. I'm not offering it as accurate. Five of the seven food lines carry the 15 to 20% uncertainty flag, and the two that don't are the kopi o ais and the can of 100 Plus. Every parse is shown for review before anything is saved. Nothing is scanned: no barcodes, no database search. It costs an AI Credit per parse. Whether 87 g is enough depends on body weight, which has its own post.
Habitus Mind is free to start. The quick-log lives in its Fitness module; to try it, create a free account.
How do I track calories without an app?
It works.
- Keep a notebook and write each meal as you eat it.
- Use fixed portions for repeat meals. A portion guessed the same way every time is a consistent bias.
- Add each day up once and work in weekly averages. A single day is noise.
- Weigh yourself every morning and average the week. That average checks the notebook.
So which method should I use?
The rule is about you, not the tool.
If you cook most of your meals, eat a fairly fixed rotation and don't mind weighing, the food scale is the most precise of the six. If you eat out or share meals most days, precision isn't available, and what survives is a consistent low-effort estimate you'll still be entering on day 30: a notebook, a fixed-portion habit, or describing the day in one go. If you'd rather not count, use the plate method and the weekly weigh-in, and accept that it says nothing about a single day.
In all three cases, keep the weigh-in. Without it, the log has nothing to be checked against.
FAQ
What is the easiest way to track calories?
The one with the least work per meal that you'll still do on day 30: for most people, a rough estimate written down once a day. A crude method you keep gives a trend; a precise one you drop gives two weeks of data.
Do I need a food scale to track calories?
No. A scale is the most precise option, but it has the most friction and stops working at restaurants and shared meals. A consistent estimate checked against a daily weigh-in works without one.
Can I track calories without an app?
Yes. Use a notebook, fixed portions for repeat meals and weekly averages, and weigh yourself every morning. The weekly weight checks the notebook.
How do I find my maintenance calories?
Log a consistent intake for 14 days while weighing daily. If your weight average holds, that intake is maintenance. If it moves, the rule of thumb is roughly 7,700 kcal per kilogram of change, spread across the days. It is an approximation, not a law.
Why are my calorie counts always wrong?
Mostly because of portions, not the database. In one study MyFitnessPal's energy figure was 1.3% off a national database; in another, a group under-reported intake by 47%. A count can be wrong and still useful, if it is wrong the same way every day.
