What's in the bag.
We read 2,782 real dog-food recipes from 408 brands as one market. The useful story isn't which ingredient is most common. It's where the claim stops, which formula system a brand repeats, and who else is running it.
The salt line only tells you about the produce.
Ingredient panels are printed heaviest first. Salt is a small inclusion in a dog food, so the ingredients printed after it are smaller still. That does not give you a dose. What it gives you is a position, and position is only informative for some kinds of ingredient.
The corpus backs the reading. Salt lands at position 15 of the list in a typical recipe, with 16 ingredients printed after it. Structural ingredients almost never fall on the far side: chicken does in 0.6% of the recipes naming both, and corn and oatmeal never do.
What the rule is usually used for, though, does not hold up against the data. Most of what sits below the salt belongs there. Vitamins are below it 87% of the time, probiotics 98%, minerals 69%. A dog needs a few hundred milligrams of glucosamine a day, so finding it under the line is correct formulation, not a shortfall. Treated as a verdict, it convicts every complete panel in the corpus.
The line carries information for one category. Produce sits below it 36% of the time overall, and inside that category the spread is wide enough to be the finding:
Nothing separates the top of that list from the bottom nutritionally. The structural carbohydrates are essentially never below the line. The decorative ones usually are. 49% of salt-bearing recipes carry at least one fruit or vegetable below it, 3 of them in a typical case, and one panel carries 15.
The obvious next claim is that brands advertise those ingredients and then bury them. We tested it, because a product name is printed on the front of the bag: it does not hold. Across the 71 recipes that name one of these ingredients in the product name and also list it on the panel, base rates say 28.2 should fall below the salt. 3 do. Naming an ingredient makes it roughly ten times less likely to be buried, and the direction holds for every ingredient we could test.
So the honest version is narrower and more useful than the version circulating. A named ingredient is load-bearing 96% of the time. The produce below the line is the produce the bag never claimed in words, which is where imagery and blend language do their work. This capture reads text, not packaging, so that half of the question is still open.
The basis is the 1,846 captured recipes that name salt. Each row below counts only the recipes naming both salt and that ingredient.
Joint-support language is the sharpest example. Glucosamine is below salt in 191 of 198 recipes that name both. Chondroitin is below it in 152 of 159. Flaxseed is the counterexample: it sits above salt 94% of the time.
without doing much to build the formula.
The first five ingredients track the format. Everything after salt tracks the brand.
We split every complete, salt-bearing panel into two regions: the first five ingredients, where the consumer-facing formula story lives, and the ingredients after salt, where the nutrient and functional system lives. Then we measured how much of each region two recipes actually share. The within-brand half rests on the 63 captured brands with at least six panels complete enough to split, so read it as a finding about those brands rather than about the whole market.
In the first five ingredients, format is the stronger signal. Two different brands share 1.75× as much when they make the same format as when they do not. After salt it reverses. A single brand matches its own recipes across formats 1.75× as closely as it matches other brands making the same format.
The brand decides what comes after salt.
A panel can identify who shares a formula system.
We found 69 cross-brand recipe pairs sharing at least 75% of a full ingredient list, across 28 brand pairs. Every panel in this test lists at least 25 mapped ingredients. Then we tried to break it.
Every complete panel carries a vitamin and mineral pack, and brands buy those from a small number of suppliers. Two companies on the same premix look like two companies on the same formula. So we held the sample fixed, stripped the vitamins, minerals, preservatives and probiotics, and ran it again. 48 of the 69 matches disappeared — 70% of them were the premix, not the recipe.
21 recipe pairs across 15 brand pairs are still matching once the pack is gone.
Our own control failed the same test, which is worth publishing. Taste of the Wild is a Diamond brand, and we had been using the pair to show the method works. 1 of their 14 matches holds up once the pack is removed. They were matching on the vitamins.
Which is the argument for running the control at all. A test that never fails is not a test. The pairs left standing matched on food rather than on a supplier's premix, and they are a reason to look into co-manufacturing, private label or shared sourcing. A shared ingredient list is still not proof of a commercial relationship.
A brand's kibble carries ingredients the rest of its line does not.
Across the 61 captured brands that make kibble and at least one other format, these are the ingredients most often present in the kibble and absent from everything else that brand makes:
The pattern is not a trend list. It is a manufacturing constraint: meals and fats for the structure, plus an antioxidant system for shelf stability. Whatever else a company runs, its kibble is carrying a bill of materials the rest of the line is not.
This is a comparison between a brand's own formats, not a migration. Only 74 of the 408 captured brands make more than one format at all, and among those, kibble is usually where a company started rather than where it went: it is the largest line for 45 of them and a secondary line for 16. The panel shows what separates the formats, not the order a company built them in.
Two foods that look four times apart are the same food.
Guaranteed analysis is printed as fed — the food exactly as it leaves the can or the bag, water included. A canned food is mostly water. A freeze-dried food has had almost all of it removed. So the same protein, in the same amount, prints as a completely different number depending on how the food is preserved.
Put the two furthest-apart formats side by side. A shopper reads 8% on the can and 36% on the freeze-dried bag, and concludes one has 4.5× the protein of the other. Take the water out of both and they land in the same place: 40.9% against 40.9%, across 422 and 213 products. Not close. The same.
That holds across every format. The distance between these two columns is the water, not the formula:
Read the right-hand column and the market simplifies. Canned, fresh and freeze-dried all sit near 40.9%. Raw sits a little above. Only kibble is genuinely different, and it is the one format whose label number already looked reasonable.
Inside kibble the number stops separating anything at all. 70% of the 1,348 kibbles here sit within five points of 29% dry-matter protein, and 51% sit within three. That is the most crowded, most heavily marketed shelf in the category, and on the number brands compete on hardest, most of them are within a rounding error of each other.
The practical consequence runs backwards from what the shelf suggests. Every one of the 422 canned products here prints a protein figure below the median kibble's 26%. 368 of them — 87% — hold more protein than that kibble once the water is out. Compare the two numbers on a shelf and the ranking is wrong nine times in ten.
This finding is the only one here that does not come from the ingredient panel. It is read off the guaranteed analysis, which 2,198 products across 393 brands publish completely enough to convert — 78% of the corpus. Where our sources printed different numbers for the same product, we dropped the product rather than pick a winner.
but it is most of what you are comparing.
Five questions to ask of any formula.
- Does the hero ingredient sit above the salt line? If not, the claim may be differentiated while the formula is not.
- What repeats after salt? The brand's formulation system, and often the more durable signature.
- What is the protein figure once the water is out? As printed it measures the format. Converted, it measures the food.
- Which competitors share the same formula fingerprint? Similarity can expose a closer operating peer than category or price tier suggests.
- What does the target format force you to add? Model the format tax before treating a line extension as a packaging exercise.
These are ingredient-panel findings, not efficacy claims. We hold no trial data and infer no clinical outcome. Full methodology and limits: the methodology page.