How often do entries for one product disagree on calories?
In 46.4% of name-and-brand groups with more than one entry, at least one entry differs from the shown row on calories: 124,534 of 268,578 groups in the Smidge food database.
| Entries in the group | Groups | With a differing entry | Share |
|---|---|---|---|
| 2 | 171,303 | 62,412 | 36.4% |
| 3 to 5 | 73,325 | 42,802 | 58.4% |
| 6 to 20 | 20,723 | 16,292 | 78.6% |
| 21 or more | 3,227 | 3,028 | 93.8% |
| All with 2 or more | 268,578 | 124,534 | 46.4% |
A group is not always one product
Groups form from the name and brand text alone, each reduced to lowercase letters and digits, so "Coca-Cola" and "coca cola" fall together. That is the only link between entries. A group can therefore hold several flavors, sizes, light versions or countries' recipes under one name, and some of the differences counted here are different products and not mistakes.
The data has examples of both kinds. The Danone Skyr group holds 20 entries between 51 and 81 kcal per 100 g, and its higher entries include one tagged fruit-skyrs (barcode 3033491998324, 76 kcal) and one tagged vanilla-yogurt (8001630012328, 70 kcal). The Chocapic group, shown at 384 kcal, includes an entry at 407 kcal (5900020027771) tagged as an extruded cereal bar. Those look like other products filed under the same name.
Other entries look like input errors. One Coca-Cola entry (7611223) says 594 kcal per 100 g. One entry in the Heinz Tomato Ketchup group (8715700124575) says 15 kcal beside a serving of 15 g. Scaled to 100 g that would be 100 kcal, the figure most ketchup entries carry, so it reads like a per-serving number typed into the per-100 g field. The counts here make no attempt to sort entries into errors and real differences.
Why entries for one product disagree
- More than one barcode. Packs of different sizes carry different barcodes, each entered separately. Brand spelling can also split a product: Nutella with the brand Ferrero is one group of 70 entries, and Nutella with the brand Nutella is another of 124.
- Entries that cluster by market. In the Heinz Tomato Ketchup group, 88 of 141 entries not tagged for the US sit between 95 and 105 kcal per 100 g, and 26 of 37 tagged for the US sit between 115 and 130. Different recipes in different countries could explain that, though the data does not say.
- Possible typing errors and unit mix-ups. A per-serving value can look like a per-100 g value, as the ketchup entry above may, and per-100 ml and per-100 g figures can be mixed up. The data cannot show which entries are such slips.
- Other products under the same name. In the Coca-Cola group of 324 entries, 243 (75.0%) sit between 35 and 45 kcal per 100 g, 28 are under 5 kcal (24 of them with sugar below 1 g) and 8 are at 150 kcal or more. Zero-sugar versions filed under the plain name would explain the entries under 5 kcal.
How Smidge picks the number it shows
The food database keeps every entry but shows one row per group, the row the food search lists. The build picks it in four steps:
- Each entry gets a signature: calories rounded to the nearest 5, and protein, carbs and fat each rounded to the nearest 2 g. A missing nutrient counts as its own value.
- The most common signature in the group wins. A tie goes to the signature with a USDA entry, then to a fixed order by the signature's text.
- Among entries with that signature, a USDA entry comes first, then one with a serving of 1 to 500 g, then the most scanned, then the lowest barcode.
- The shown row's calories are the reference. Another entry counts as differing when its calories are further from the shown row's than the larger of two allowances: 5% of the larger of the two values, or 3 kcal.
For a shown row of 100 kcal, an entry of 105 does not differ (5 is not more than 5.25, 5% of 105) and an entry of 106 does (6 is more than 5.3). For a shown row of 20 kcal, an entry of 22 does not differ, because 3 kcal is the larger allowance. The rule looks at calories only, and the shown row is the most common signature, not an average. The method page and the food database page describe where the data comes from.
The most-scanned groups
Scans are Open Food Facts scans summed over a group's entries. The table lists the ten most-scanned groups with at least 20 entries.
| Product | Shown | Lowest | Highest | Entries | Differing |
|---|---|---|---|---|---|
| Nutella (Ferrero) | 539 | 500 | 556 | 70 | 1 |
| Tomato Ketchup (Heinz) | 102 | 15 | 173 | 178 | 87 |
| Nutella (Nutella) | 539 | 37 | 550 | 124 | 8 |
| Coca Cola (Coca-Cola) | 39 | 0 | 594 | 324 | 142 |
| Nocciolata (Rigoni di Asiago) | 544 | 157 | 582 | 83 | 3 |
| Skyr (Danone) | 57 | 51 | 81 | 20 | 8 |
| Chocapic (Nestlé) | 384 | 116 | 407 | 53 | 11 |
| Coke Zero (Coca-Cola) | 0 | 0 | 43 | 43 | 2 |
| Coca-Cola Original Taste (Coca-Cola) | 42 | 30 | 100 | 36 | 3 |
| Coca-Cola Zero Sugar (Coca-Cola) | 0 | 0 | 1 | 63 | 0 |
The groups range from near-agreement to wide spread. Nutella under Ferrero runs from 500 to 556 kcal with 1 of 70 entries differing, and Coca-Cola Zero Sugar from 0 to 1 with none of 63. The Coca-Cola group has 142 of 324 entries differing, running from 0 to 594.
How large the spread gets
Of the 268,578 groups with more than one entry, 69,209 (25.8%) have a spread (highest minus lowest) of more than 20% of the highest value, 36,106 (13.4%) have a spread above 100 kcal, and 4,781 (1.8%) have a lowest value of 0 beside a highest above 0. Counted by entries, 286,651 of the 683,584 entries other than the shown rows (41.9%) differ. Among the 67,797 groups with at least 10 scans, 22,136 (32.7%) have more than one entry, and 11,443 of those (51.7%) have a differing entry.
What published studies found
These studies measure something different from this post, which compares entries with each other.
- Labels against a laboratory. Jumpertz and colleagues measured 24 popular US snack foods by bomb calorimetry. After allowing for serving size, metabolizable calories were 6.8 kcal, or 4.3%, higher than the label statement, which the authors called "overall well within FDA limits".
- Restaurant and frozen foods. Urban and colleagues measured reduced-energy foods. Energy in 29 restaurant foods averaged 18% more than stated, and in 10 supermarket frozen meals 8% more. The authors report that these differences substantially exceeded laboratory measurement error but did not reach statistical significance, because the variation between items was large.
- The legal limit. Under 21 CFR 101.9(g)(5), a food can be deemed misbranded when its calories are more than 20 percent above the label value, with no action for differences within the usual variability of the lab method.
- Tracking-app databases. Evenepoel and colleagues had 50 participants keep a 4-day record in MyFitnessPal and compared the energy with the Belgian Nubel database. After extreme values (2.8% of the total) were removed, the correlation for energy was r = 0.96. Griffiths, Harnack and Pereira entered 24-hour recalls into five popular apps and compared them with a research dietary system. Correlations for energy and macronutrients ranged from 0.73 to 0.96, and app values tended to be lower.
What the numbers do and do not show
They show how often entries filed under one name and brand differ by the rule above, and that the share rises with the size of the group, from 36.4% at two entries to 93.8% at 21 or more. More entries also mean more chances for a different size, flavor or typing slip.
They do not show which entry is right, or how often the database is wrong. Most groups cannot be compared at all: 2,515,615 of 2,784,193 (90.4%) hold a single entry.
Where this data falls short
- Coverage and n. The comparison covers 268,578 of 2,784,193 groups, the 9.6% with more than one entry.
- Contributor-entered labels. Open Food Facts entries are typed in by contributors from packaging, and USDA Branded Foods rows are label data too. Label values can differ from measured energy, as the studies above show.
- Three category tags. Each product carries up to three tags, and 51% of groups have none, so the post does not split the result by category.
- One snapshot. The data is the build of 16 November 2026, with Open Food Facts data through 15 November 2026 and USDA FoodData Central release 29 October 2026. Entries change as contributors edit them.
- Grouping by name and brand. The key is the name and brand text only, so one product can split across groups and different products can share one.
Sources
Food data from USDA FoodData Central (public domain) and Open Food Facts (Open Database License). The ultra-processed share post uses the same build.
- Food label accuracy of common snack foodsJumpertz R et al. · Obesity, 2013
- The accuracy of stated energy contents of reduced-energy, commercially prepared foodsUrban LE et al. · Journal of the American Dietetic Association, 2010
- 21 CFR 101.9, nutrition labeling of foodUS Code of Federal Regulations
- Accuracy of nutrient calculations using the consumer-focused online app MyFitnessPal: validation studyEvenepoel C et al. · Journal of Medical Internet Research, 2020
- Assessment of the accuracy of nutrient calculations of five popular nutrition tracking applicationsGriffiths C, Harnack L, Pereira MA · Public Health Nutrition, 2018