Nine saved entries cover half of everything you will ever log#
The fastest way to speed up food logging is not a better interface. It is to notice that you eat a startlingly small number of distinct things, build a permanent entry for each one, and never estimate them again. The size of that shortcut has now been measured, and it is larger than almost anyone assumes.
When 21,006 adults logged 2,655,718 food and drink entries over two weeks, the median person recorded about 50 unique items across the fortnight. But the distribution of how often each item appeared was extremely lopsided: for half of users, just nine unique items accounted for 50% of all their logged entries, 21 items accounted for 75%, and 35 items accounted for 90%1. Past roughly the twenty-second most frequent item, the rest of the list was made up of foods eaten exactly once.
So the work of building a fast log is finite and small. Nine carefully built entries — properly weighed, properly componented, done once — retire half your future logging. Twenty-one retires three-quarters. Everything after that is a long tail of one-offs that will never repay the effort of getting right.
The distribution is lopsided in a way that decides your strategy#
| Number of saved entries | Share of a median user's logged entries covered |
|---|---|
| 9 | 50% |
| 21 | 75% |
| 35 | 90% |
Data: Tran et al., 2026; 21,006 adults, 2.66 million logs over 14 days.
Two details in that dataset keep the shortcut honest. The first is that repetition is real but not rigid: participants averaged only 3.85 items consumed on at least seven of the fourteen days, and no single food was eaten every day by anyone. Only four things — coffee, milk, black coffee and tea — were consumed on all fourteen days by more than a hundred participants. A saved-meal library is a library of frequent items, not of daily rituals.
The second is that unique-item counts depend on how entries get named. Two people eating identical breakfasts can produce different counts depending on whether they log "oats, milk, banana" or "porridge." And these were users of a research logging app, who are not a random sample of eaters. Read the shape of the distribution rather than the exact integers.
An independent dataset lands in the same place from the other direction. Among 112 adults in a twelve-week weight-loss program who logged 105,422 food entries, only 41% of the entries (excluding drinks and condiments) were unique foods — meaning about 60% of everything they logged was something they had logged before3. Three-fifths of your typing is duplicated work.
Most food logging is not measurement. It is re-typing things you already measured, which is why the effort curve should fall over time and usually doesn't.
Build for breakfast, because dinner will never cooperate#
Repetition is not spread evenly across the day, and this is the single most useful thing to know before deciding which entries to build first.
An analysis of 587,187 logged days from 1,581 long-term food-diary users measured how strongly food items recurred at each eating occasion. Breakfast came out highest, at a food-item recurrence strength of 0.65; dinner came out lowest, at 0.24. The ordering was breakfast, then snacks, then lunch, then dinner, with breakfast recurrence running two to three times that of the other occasions. Recurrence was also significantly higher on weekdays than weekends across every meal (p<0.01)2.
The larger study agrees without having set out to: habitual foods were more likely to be consumed earlier in the day, and as the day progressed the share of novel items rose1. Two datasets, different populations, different methods, same gradient.
Which gives an ordering for the work:
- Weekday breakfast first. It is the most repeated occasion in the day, it is usually assembled at home from countable components, and one good entry can cover 200 mornings.
- Snacks second — higher recurrence than lunch, and the occasion whose items are most often eaten without a plate, which is where records leak.
- Weekday lunch third.
- Dinner last, and only for the specific dishes you cook repeatedly. At a recurrence strength of 0.24, a dinner library will cover a minority of your dinners no matter how much effort you put into it. Dinner is where estimation skill has to do the work instead.
That ranking runs against instinct, because dinner is the meal people feel least confident about and therefore want most to systematize. The data says the systematizable meal is the one you barely think about.
Repetition is not just cheaper to log — it tracks the outcome#
There is a second reason to lean into repeated meals, and it is not about typing speed. In that 112-person program, the proportion of a person's entries that were unique foods negatively predicted twelve-week weight loss (p=.004): every 10-percentage-point drop in the share of unique foods was associated with about 0.5 percentage points more weight lost. The complementary measure agreed — a 10-point rise in the share of foods logged more than ten times was associated with about 1.6 points more weight loss (p=.031). Steadier day-to-day calorie intake pointed the same way, with each extra 100 calories of average daily deviation associated with roughly 0.6 points less weight loss (p=.025)3.
Hold that at its actual strength. It is 112 people, it is correlational, and the analysis controls for how many days people tracked — which matters, because tracking frequency is itself a strong predictor and is correlated with repetition. People who repeat meals may simply be people with more orderly lives.
The study also produced one result its authors did not expect and did not bury: participants with larger weekend-versus-weekday calorie swings lost more weight, not less (p=.025). That is the opposite of the routine hypothesis, from the same preregistered analysis, and it is a good reason not to over-read the general finding into "never vary anything." What survives is narrower and still useful: repeating your food is at minimum free, and probably helps.
What to actually build, and in what order#
A library that pays for itself needs about an hour, spread over a fortnight:
Log normally for two weeks, then read your own frequency list. Nearly every app can sort your history by how often an item was used. You do not need to guess which foods repeat; two weeks of data tells you, and the answer will surprise you at least twice.
Build the top nine properly. Weigh them once. Include the oil, the spread, the milk in the coffee — the components you would skip when typing at speed are precisely the ones a permanent entry should capture, because you are paying that attention once and collecting it hundreds of times. This is the same economics that makes batch cooking such a favourable place to spend a scale.
Save assemblies, not ingredients. "My breakfast" as one entry beats four items you must re-select each morning. The recurrence data says whole meals repeat far less than individual items do — so save meals where they genuinely repeat, and items everywhere else.
Let the tail be rough. The foods beyond your top 30-odd are, in the median case, things you will eat once. A quick-add with a note is the correct treatment for those; building careful entries for them is effort that never returns.
Rebuild the list quarterly. Diets drift with the seasons and with jobs. A library assembled in January describes a person who no longer exists by June, and a stale entry is worse than no entry because you will reuse it without noticing.
Done this way the log stops being a daily estimation task and becomes a daily selection task, which is a different order of effort — measured at roughly two to five minutes a day for a stripped-back method against thirty-four for full tracking in one trial4. That drop is what keeps a record alive through a bad month, and it is the mechanism behind almost everything that makes tracking survive past week ten or through a week with no time in it. Restaurant orders deserve exactly the same treatment and have their own version of this library; the general method it plugs into is how to count calories.
FAQ#
How many saved meals do I actually need?#
About nine to cover half your entries and twenty-one to cover three-quarters. Across 21,006 adults logging 2.66 million items over two weeks, nine unique items accounted for 50% of the median user's logs, 21 for 75% and 35 for 90%1. Building more than about thirty-five careful entries is effort spent on foods you will eat once.
Which meal should I build a shortcut for first?#
Weekday breakfast. Measured across 587,187 logged days, food-item recurrence at breakfast was 0.65 against 0.24 at dinner — two to three times higher than any other occasion — and recurrence was higher on weekdays than weekends throughout2. One good breakfast entry covers more logged occasions than any other single thing you can build.
Is it a problem that dinner never repeats?#
It is a limit rather than a failure. Dinner has the lowest recurrence of any eating occasion, so no library will ever cover most of your dinners2. Plan for that: shortcuts handle the morning and the middle of the day, and dinner gets whatever estimation method you would use for an unfamiliar plate.
Sources#
- Tran T, Manoogian ENC, Hou ZJ, et al. The diversity and consistency of what and when people eat. Nat Metab. 2026;8(4):981-997.
- Pai A, Sabharwal A. Food habits: insights from food diaries via computational recurrence measures. Sensors (Basel). 2022;22(7):2753.
- Hagerman CJ, Hong AE, Crane NT, Butryn ML, Forman EM. Do routinized eating behaviors support weight loss? An examination of food logs from behavioral weight loss participants. Health Psychol. 2026.
- Patel ML, Cleare AE, Smith CM, Rosas LG, King AC. Detailed versus simplified dietary self-monitoring in a digital weight loss intervention among racial and ethnic minority adults: fully remote, randomized pilot study. JMIR Form Res. 2022;6(12):e42191.
