Start with what a healthy gut actually looks like.
Before asking what diet changes, it is worth asking what it changes from, and the honest answer is that the healthy baseline is harder to pin down than it sounds. The faecal microbiome of healthy dogs is co-dominated by three phyla, Firmicutes, Bacteroidetes and Fusobacteria, while the cat's is dominated in most studies by Firmicutes, followed by smaller proportions of Proteobacteria, Actinobacteria, Bacteroidetes and Fusobacteria. That much is settled. Below phylum level, agreement thins quickly.
Two large reference studies show how thin. Rojas and colleagues in 2024 set out to define a species-level core for healthy dogs using full-length 16S sequencing. They collected faecal samples from 3,754 dogs and, after applying strict health criteria, retained just 286 of them, 7.61 per cent of the samples they started with. Ganz and colleagues in 2022 ran the equivalent exercise in cats, screening 1,859 samples and qualifying 161, or 8.9 per cent. In both cases more than nine animals in ten failed to meet a reasonable definition of healthy once medications, antibiotics in the previous year, body condition and clinical signs were taken into account.
What survived that filter in dogs was a core of 23 bacterial species, defined as those present in at least a third of animals at a mean relative abundance above 0.5 per cent. The most abundant were Megamonas funiformis at 11.2 per cent mean relative abundance, Streptococcus lutetiensis at 9.4 per cent, an unclassified Fusobacterium at 6.5 per cent, Clostridium perfringens at 5.3 per cent, Collinsella intestinalis at 4.8 per cent and Faecalimonas umbilicata at 4.4 per cent. Both of those reference papers come from AnimalBiome, a commercial pet microbiome company, and the 2024 paper discloses that the funder was involved in study design, analysis and writing. That does not invalidate the work, and it is the sort of thing worth knowing when a company cites a reference range at you. We are a commercial laboratory too.
Fibre works, but not all fibre and not by one route.
The clearest dietary signal in dogs comes from fibre, and the mechanism is straightforward: most fibres act by enriching fibre-fermenting, short-chain-fatty-acid-producing Firmicutes. Different fibres do different things. In the dog studies collected by Pilla and Suchodolski, beet pulp raised Firmicutes and Clostridia while lowering Erysipelotrichi and Fusobacteria. Inulin lowered Enterobacteriaceae and raised Megamonas and Lactobacillus. Potato fibre raised Faecalibacterium, Lachnospira and faecal acetate, propionate and butyrate. Inulin-type fructans raised short-chain fatty acids and total faecal bile acids.
That Enterobacteriaceae result is worth pausing on, because an increase in this family is a hallmark of dysbiosis in dogs, and it is one of the few dietary levers shown to push a dysbiosis marker in the right direction.
The size of the dietary shift tracks the size of the macronutrient change. In a study of four dry prescription diets in healthy dogs, the weight-loss formulation produced the largest shift in composition, and it was also the most extreme in macronutrients at 28.1 per cent fibre against 8.6 per cent in the low-fat diet. Bacteroides, Prevotella and Faecalibacterium all increased significantly on the weight-loss diet.
Not every fibre does something. In cats, cellulose produced no changes, and a wool hydrolysate produced no changes. Prebiotic fibres in cats do raise short-chain-fatty-acid producers and Bifidobacterium, and the combination of fructooligosaccharides and galactooligosaccharides raised total short-chain fatty acids, butyrate and valerate. But the presence of fibre on an ingredient list is not the same as an effect, and the two null results above are the reason to ask which fibre rather than how much.
The butyrate story is not the one on the packet.
Butyrate is the compound most often used to justify fibre, and its importance is not in dispute: it is the preferred energy source for colonocytes. What is often left out is that butyrate does not require carbohydrate.
Butyrate is found in the faeces of all mammals regardless of what they eat, which means alternative production routes must exist in animals that consume little or no carbohydrate. The evidence in dogs supports this directly. In a study comparing high-fat with high-starch diets, acetate, butyrate and propionate levels were not different between the two groups, indicating that short-chain fatty acid production in dogs is not exclusively dependent on carbohydrate content. Adding minced meat to a dry diet produced a small increase in faecal butyrate.
The mechanism is now partly mapped. Clostridium and Eubacterium can produce butyrate from carbohydrate through the but pathway or from protein through the buk pathway, while Faecalibacterium, Ruminococcus and Blautia produce it through the but pathway only. In carnivores, Clostridiaceae and C. perfringens in particular are associated with the butyrate kinase route that makes butyrate from protein. Fusobacterium varium, another protein-to-butyrate producer, was more abundant in dogs fed raw diets for more than a year, which reads as adaptation to a long-term diet rather than damage from it.
The practical consequence is that increased fibre and increased protein may bring similar benefits by different routes, and that a bacterium being labelled beneficial in an omnivore does not automatically make it beneficial in a carnivore. The ideal fibre and protein intake for a healthy pet microbiome has not been determined.
Raw diets: what 46 dogs actually showed.
Raw feeding is the dietary question owners ask most often, and one study answers it more completely than any other. Schmidt and colleagues in 2018 compared 27 dogs fed bones-and-raw-food diets with 19 fed commercial diets, using 16S sequencing, targeted qPCR and untargeted metabolomics. The diets were genuinely different: the raw-fed dogs took in 44.40 per cent protein and 28.40 per cent fat on a dry matter basis against 30.45 and 18.21 per cent in the commercial group, with carbohydrate at 15.75 per cent against 40.43, and fibre at 2.69 per cent against 3.36.
Four findings matter, and they do not all point the same way.
Diversity did not change. Chao1 richness was 1595 in raw-fed dogs against 1559 in commercially fed dogs, p = 0.655. Shannon diversity was 5.07 against 4.82, p = 0.148. Observed species were 800 against 759, p = 0.086. None of these reached significance. Whatever raw feeding does, it does not make the community more or less diverse.
Composition did change. The two groups separated significantly on beta diversity, ANOSIM p < 0.01, and 34 taxa differed between them. Proteobacteria and Fusobacteria were higher in raw-fed dogs, p < 0.0001 and p = 0.013, while Firmicutes were higher in commercially fed dogs, p = 0.001. At genus level Bifidobacterium and Faecalibacterium were both lower in the raw-fed group, p = 0.004 and p = 0.002.
The dysbiosis index was significantly higher in raw-fed dogs, p < 0.001, driven by more E. coli and less Faecalibacterium. This is the finding most often quoted and most often over-read. The authors are explicit about its limits: the index was trained against the microbiota of dogs with chronic intestinal inflammation, and at this stage it is unknown whether the changes observed will cause intestinal disease in the future. Every dog in the study had normal stool quality and no history of gastrointestinal problems on its current diet.
The metabolome mostly did not survive scrutiny. Of 233 metabolites identified, 33 differed between the diets on unadjusted p-values, and all of them lost significance after adjustment for multiple comparisons. Faecal cholesterol was higher in raw-fed dogs, p = 0.0065, but primary, secondary and total bile acid concentrations did not differ. A study that found a compositional difference did not find a matching functional one.
Two further details deserve mention. The study's own qPCR and its sequencing disagreed about Fusobacterium, which the authors flag as unexplained and as an argument for combining methods rather than trusting one. And the first author was employed part-time at a raw-feeding nutrition consultancy whose owner had published a book on the diet, which the paper discloses. A study that reports an unfavourable finding despite that association is more credible for having declared it, not less.
What diet cannot do.
This is the part usually left out. Diet-induced changes in the microbiome of healthy dogs are less marked than the changes associated with disease. In sick animals, and particularly those with chronic enteropathies, diversity falls quickly and core species including Clostridium hiranonis, Fusobacterium and Faecalibacterium prausnitzii decrease. Dietary manipulation is unlikely to generate changes comparable in magnitude to those seen in disease.
Diet effects also do not persist on their own. The microbiome is resilient, and dietary change is maintained only by keeping the diet in place. Healthy dogs fed nothing but purified amino acids and easily digestible starch for 32 weeks returned to baseline composition quickly once the control diet resumed. Hypoallergenic diets do not significantly affect the microbiome of healthy dogs at all, although they have been associated with improvement in dogs with food-responsive diarrhoea, which is a useful reminder that a diet can do something in a sick animal and nothing in a well one.
There is also a measurement problem that no laboratory can currently escape. Different sequencing and data analysis methods generate different results, which prevents the development of reference intervals, and no true analytical validation of 16S rRNA gene sequencing has been reported. Most studies compare a small group against its own baseline rather than against a large reference population, so the magnitude of any reported change is hard to place. When you see a percentage attached to a bacterial taxon, the method that produced it is part of the number.
Finally, association is not causation. Dysbiosis accompanies a long list of conditions, and a causal effect is yet to be proven; the dysbiosis may be a symptom of the disease process rather than its cause.
Where BAARK fits.
A description of the community as it stands, read against dog-specific and cat-specific reference cohorts.
BAARK profiles the faecal community by shotgun metagenomic sequencing and reports what was present in the sample, alongside a targeted qPCR panel for parasites. Where the evidence above supports a dietary suggestion, the report makes it and names the basis. Where it does not, the report says so.
What this means in practice: a single sample describes the community on the day it was collected, on the diet the animal was eating. Because dietary change is maintained only while the diet is, and because individual variation is large, a second sample after a sustained change is worth more than a single reading interpreted against someone else's average. What BAARK provides is a wellness profile. It is not a diagnostic test. It does not detect, diagnose, stage or predict the progression of any disease, and it is not a substitute for veterinary assessment.
The bottom line.
Fibre and protein content shape the dog and cat gut community more than ingredient lists do, and both can raise butyrate by different routes, so the choice between them is less stark than the marketing on either side suggests. Raw feeding produces a real compositional shift and a higher dysbiosis index in healthy dogs, on an index trained against sick ones, with no accompanying difference in diversity and no metabolomic difference that survived correction. Every one of these effects is smaller than what disease does, and none of them lasts longer than the diet that caused it.