The best canine dataset we have on this comes from 22 dogs and 241 samples.
Most of what owners are told about antibiotics and the gut is extrapolated from human studies. There is now a canine dataset that answers the question directly. Marshall-Jones and colleagues in 2024 followed 22 dogs prescribed a seven-day course of metronidazole at 25 mg/kg daily, between 400 and 500 mg per dog, for veterinary management of a suspected Giardia infection after signs including mucoid bloody diarrhoea and sporadic vomiting. The cohort was 8 neutered males and 14 females, aged 0.6 to 8.8 years, across Petit Basset Griffon Vendéens, Labrador retrievers and Beagles.
What makes the study useful is the sampling. Faeces were collected immediately before treatment, at the end of the course, two to three days later, seven to eight days later, then at two, four, six and eight weeks and monthly out to six months. In total 241 samples were profiled by 16S sequencing of the V4 region, yielding 1,131 amplicon sequence variants across a 25-week window. Very few veterinary microbiome studies follow animals that far past the intervention.
The work was funded by Mars Petcare and the authors were employed by Mars Petcare UK at the time, which they disclose, along with the statement that there are no products in development or marketed products to declare. We note it because a reader should weigh who paid for a study, including when the finding is unwelcome to nobody in particular.
What a week of metronidazole does.
Before treatment, mean Shannon diversity across the cohort was 3.66, with a 95 per cent confidence interval of 3.41 to 3.92. At the end of the seven-day course it had fallen to 2.69, a mean difference of 0.97 and p less than 0.001. Species richness fell by an estimated 103.74 observed variants over the same period, again p less than 0.001.
That is a large, fast reduction from a short, routine, entirely appropriate course of a common antibiotic in animals that needed it.
Three days after the course finished, diversity had climbed to 3.10 but was still significantly below baseline, p less than 0.001. By one week after cessation it reached 3.54, and the difference from pre-treatment was no longer significant, p equal to 0.93. From that point on, with one exception, diversity stayed at baseline for the remaining six months. The exception was Week 20, where a smaller dip reappeared, mean difference 0.47, p equal to 0.001, which the authors do not explain and which is a reminder that a single reading can move for reasons a study does not capture.
Diversity is not the same as recovery.
Here is the finding that matters most, and it is the one a diversity number on its own would hide. Alpha diversity, the head count, returned within a week. Beta diversity, which measures how similar the whole community looks to its earlier self, took much longer.
Comparing each timepoint against the pre-treatment sample, the community distance was largest during treatment, R squared 0.46, and immediately after, R squared 0.42, both significant. It was still significantly different at one week, R squared 0.20, at two weeks, 0.13, and at four weeks, 0.07. Only by six weeks did the difference stop being significant, R squared 0.04 with an adjusted p of 0.06, and by 24 weeks it was 0.03.
So the count came back in a week and the community took about six. If you sampled a dog seven days after its course finished and looked only at a diversity score, you would conclude nothing had happened. The composition would tell you otherwise for another month.
The organisms that rose are the ones in the supplement.
The taxa that increased during metronidazole treatment were Lactobacillus, Bifidobacterium and Enterococcus. The authors attribute this to antibiotic resistance commonly exhibited by these organisms. Twenty-seven sequence variants accounted for 70 per cent of the community change, clustering in six families including Lactobacillaceae, Bifidobacteriaceae and Enterococcaceae.
These are the same genera that most companion animal probiotics are built from. That does not make giving them pointless, but it does undercut the simplest version of the story, in which antibiotics wipe out the good bacteria and a supplement puts them back. During metronidazole, those particular organisms were not the casualties. They were the survivors, and their relative abundance rose because everything around them fell.
What actually declined and then recovered were organisms like Peptoclostridium, Peptostreptococcus and members of the Erysipelotrichaceae. One Erysipelotrichaceae variant dropped close to zero as soon as treatment began and only rebounded two weeks after the course ended.
The evidence that supplements shorten this is weaker than the marketing suggests. A mixture of probiotics and prebiotics was unable to prevent or ameliorate dysbiosis in healthy research cats given clindamycin at 75 mg orally once daily for 21 days. That is a direct test of the proposition, and it did not work.
What this does not tell you.
This was one antibiotic in one context. Metronidazole is not amoxicillin-clavulanate and is not tylosin, and a seven-day course for giardiasis is not a six-week course for a deep pyoderma. The dogs were kennel-housed at a research institute on a largely uniform diet, which controls variables an owner's home does not.
Five of the 22 dogs were excluded during follow-up because they required repeat antibiotics or a probiotic on veterinary advice, the earliest at four weeks after the course finished. Their data were retained up to the point of exclusion. A study of recovery necessarily loses the animals that do not have a clean recovery to observe, and that is worth holding in mind when reading a recovery curve.
None of this is an argument against antibiotics. Every dog in this study had a diagnosed infection and clinical signs, resolution of those signs was observed from Week 1, and the alternative to treatment was an untreated giardiasis. The point is narrower: a necessary course has a measurable cost to the gut community, that cost is larger and longer than a diversity score alone reveals, and it resolves on its own in a healthy animal on a stable diet.
Where BAARK fits.
A sample taken during or just after a course describes a disturbed community, not the animal's usual one.
On the evidence above, a faecal sample collected within a few weeks of an antibiotic course is measuring the disturbance rather than the baseline. If the question is what your pet's gut normally looks like, the sample is more informative once the community has had time to settle. If the question is whether it has settled, then two samples are worth more than one.
BAARK reports what was present in the sample. It is a wellness profile, not a diagnostic test, it does not detect, diagnose, stage or predict the progression of any disease, and it is not a reason to delay, stop or alter a course of antibiotics your veterinarian has prescribed.
The bottom line.
Seven days of metronidazole cut bacterial diversity in these dogs by roughly a quarter and removed around 104 sequence variants. Diversity returned within a week of stopping. The community composition took about six weeks. The genera that rose during treatment were the ones most probiotics contain, because they tolerate the drug, and a probiotic and prebiotic mixture failed to prevent antibiotic-associated dysbiosis when it was tested directly in cats. In a healthy animal, the most reliable intervention in the published record is time.