Is there such a thing as a balanced microbiome?
Not in any sense science has been able to define. The idea that a gut microbiome sits somewhere on a line between balanced and unbalanced is a marketing frame rather than a measurement, and the literature is unusually direct about it: the opening line of a 2022 Nature study of 8,208 people states that a universal signature of a healthy or unhealthy microbiome has not been identified [14].
That sentence runs through everything below. Without a defined healthy composition there is no reference point, and without a reference point "balanced" has no numerical meaning.
"Dysbiosis": a word the field cannot agree on
The technical term behind consumer balance language is dysbiosis, and its meaning is contested by the researchers who use it.
Two researchers collected and systematically analysed a large sample of contemporary dysbiosis statements. They identified problematic implications in how the term is applied, "particularly with regard to causal hypotheses and normal–abnormal distinctions," noted that its use continues undiminished despite the questions raised, and called on researchers to reflect carefully on how they discuss it [1]. A commentary in Nature Microbiology had already put the position in its title: dysbiosis is not an answer [2].
A longer critical review is blunter. It describes human microbiome research as, with the notable exception of faecal transplantation, still mostly in a descriptive phase, and attributes part of the difficulty to "logical flaws in terminology particularly concerning 'dysbiosis' that avoids circular conclusions" [3]. The circularity is worth spelling out: if a community is called dysbiotic because it was found in people who are unwell, then reporting that dysbiosis is associated with being unwell returns exactly what was put in.
The idea is not worthless. A constructive counter-argument accepts the criticism and proposes a route forward: it "remains problematic to define the term dysbiosis by cataloguing microbial species names," and what is needed instead is a functional definition built from ecology and immunology [4]. That is a research programme, not something a species list can deliver.
The missing reference point
A review in Gastroenterology concluded that using microbiome-based biomarkers for diagnosis, prognosis or risk profiling requires a definition of a healthy microbiome in different populations — which does not exist. It would need strain-level technology and far more knowledge of how composition varies among apparently healthy people by age, diet, medication, ethnicity, geography and lifestyle; whole categories of members — viruses, fungi, archaea — remain largely uncharacterised [5]. A review working from classical ecology reached the same place, replacing the ideal healthy microbiome with one advantageous for the host in a given context [7], and an earlier review had named defining a baseline as the open challenge it remains [9].
The most consequential document is an expert workshop report with more than forty academic, government and industry participants, which explicitly addressed regulatory considerations for this kind of claim. Its conclusions include that mechanistic links of specific changes in gut microbiome structure with function or markers of human health are not yet established; that it is not established whether dysbiosis is a cause, a consequence, or both; and that microbiome biomarkers need to be determined and validated, along with normal ranges [6].
Part of why a normal range is hard to build is that very little is common to everybody. A consortium analysis of 18,340 individuals across 24 cohorts found only 9 of 410 genera detected in more than 95% of samples [13]. Two deeply phenotyped population cohorts, integrated globally to 3,948 people, identified a core of 14 genera out of 664 genera found — and noted that even those 664 underexplore total gut diversity [15]. A reference range needs a population that shares something to be a range of.
Which way does the arrow point?
Reports that people with a given condition have a different microbiome appear continually, and the interpretation offered is almost always that the microbiome is doing something to the person. The literature does not support that assumption. A systematic review of the rodent-transplant literature — the design most often invoked as proof — put it precisely: human diseases are increasingly linked with an altered or "dysbiotic" gut microbiota, but whether such changes are causal, consequential, or bystanders to disease is, for the most part, unresolved [10].
How a disease signature gets manufactured
A Nature analysis identified low concordance between studies of the microbiota in human disease as a pervasive challenge, and traced much of it to host variables that differ between cases and controls for reasons unrelated to the disease. Alcohol consumption frequency and bowel movement quality emerged as unexpectedly strong sources of variance, distributed differently between healthy and unwell groups and therefore capable of producing spurious microbial associations; matching cases and controls on them reduced both the observed differences and the rate of spurious associations [11]. Matching does not always dissolve a difference — sometimes it confirms one. What it always does is tell you which of the two you are looking at.
Two further findings close off the reading that a signature points at a specific condition. The population cohorts found that genera proposed as disease markers associated with ordinary host covariates [15]; the 8,208-person cohort found that seemingly unrelated diseases share a common microbiome signature, independent of comorbidities [14]. A pattern common to unrelated conditions is not diagnostic of any of them. Even which organisms differ is method-dependent: fourteen differential-abundance methods benchmarked across 38 datasets identified drastically different numbers and sets of significant features from the same data [18].
What real causal evidence looks like
Causal evidence does exist, and it is narrow. Using bidirectional Mendelian randomisation — inference from genetic variants, the strongest human design short of a trial — a study of 952 people found that genetically driven butyrate production was associated with a better insulin response after an oral glucose-tolerance test, and that abnormalities in propionate production or absorption were causally related to type 2 diabetes risk [12]. Read what that is: evidence about specific metabolic pathways, in specific traits, through a specific genetic instrument — not evidence that a person's measured composition causes anything they are experiencing. A larger consortium using the same methods called its own microbiome findings merely suggestive [13]. The chemistry involved belongs to gut microbiome and digestion.
What actually explains a person's microbiome
If not balance, then what? The best-powered answers are consistently mundane. In the 8,208-person three-generational cohort, composition was shaped primarily by environment and cohabitation: only about 6.6% of taxa were heritable, whereas the variance of about 48.6% of taxa was significantly explained by cohabitation [14]. In the two-cohort population analysis, 69 host covariates were associated with compositional variation, replicating at 92%; stool consistency showed the largest effect size, and medication explained the largest total variance [15]. Even the most sophisticated statistical analyses, one critique notes, explain only a small percentage of the variance — and stochastic processes may matter more than assumed [3].
Two measurement facts sit underneath this, both covered in full under how gut microbiome testing works: sequencing reports proportions of a fixed total, and total microbial load differs by up to tenfold between healthy people, which relative percentages cannot show [16][17]. The most heavily marketed balance ratio of all — the trade-off between Bacteroides and Prevotella — was shown to be an artefact of relative profiling, and did not survive counting the cells [16]. See relative abundance.
Ecology gives the vocabulary, not the score
Stability, resilience and functional redundancy come from ecology, and they are the right vocabulary here. The landmark review that established the framing described the gut microbiota as a complex ecological community — immensely diverse, varying between individuals, fluctuating over time — and called the task of understanding what drives that change "formidable" [19].
The temptation is to convert those words into personal metrics. The evidence does not support the conversion. One widely cited discussion of resilience is labelled by its own authors as an opinion article [20]; a clinical review says the relevant properties "seem to be" critical [21]. Neither offers a validated per-person measurement. More importantly, resilience is not a health property: the same opinion article that describes resilient communities as protective also states that a resilient dysbiotic microbiota may cause disease [20].
The related idea that a gut community settles into an alternative, self-sustaining state is a hypothesis stated in the conditional: unbalanced or even unhealthy stable states can develop [21]. No verified study documents an individual occupying one, and the modal verb is load-bearing. The closest thing to a balance type is the enterotype, and a 28-author reappraisal in Nature Microbiology records that the proposal met both excitement and controversy and says enterotypes may become relevant in clinical practice — future conditional [22]. The daily-sampling study found the proposed "dysbiotic" enterotype showed increased variability both between and within people [27].
Normal here, unusual there
If a single balanced configuration existed, it should be recognisable across humanity. It is not. A cross-cultural study of 531 healthy children and adults from the Venezuelan Amazon, rural Malawi and US cities found pronounced differences in bacterial assemblages and functional gene repertoires between US residents and the other two groups, evident in infancy as well as adulthood [23]. Work in Peru found hunter-gatherers form a distinct subgroup among traditional peoples — there is no single traditional microbiome either — and that Treponema, a genus most readers would assume pathogenic, falls outside the pathogenic clades, is a carbohydrate metaboliser, is found in all traditional peoples studied to date, and is simply lost in urban-industrialised societies [25].
The most striking result comes from 350 stool samples collected over more than a year from Hadza hunter-gatherers in Tanzania: an annual cyclic reconfiguration of the microbiome, in which some taxa become undetectable only to reappear in a subsequent season. Across 18 populations in 16 countries, community membership corresponded to modernisation — and the taxa most seasonally volatile within the Hadza were the same taxa that differentiate industrialised from traditional populations [24]. Whole groups of bacteria disappearing and returning is, there, ordinary.
That is evidence there is no single correct configuration — not evidence that another population's is the target. The catalogue itself is also incomplete: across 9,428 metagenomes, 77% of reconstructed species-level genome groups had no genome in public repositories as of that 2019 analysis; reference catalogues have grown since, and the unknown groups were enriched in non-Westernised populations [26]. A Western-built reference range measures against an incomplete catalogue, disproportionately so for people outside those cohorts.
The thing you would measure does not hold still
Even setting the definitional problem aside, a single measurement could only locate you against a reference if the quantity were stable. Mostly it is not. In intensive daily quantitative profiling — 713 samples from 20 Belgian women over six weeks — day-to-day absolute abundance variation was substantially larger within individuals than between them for 78% of microbial genera, with shifts of up to 100-fold; diversity and especially evenness indicators fluctuated substantially, and stool moisture was the strongest host covariate of that variation [27]. The authors add a qualifier that matters for anyone reading a consumer report: relative-abundance profiles show similar but less pronounced temporal variation than the absolute measurements above. A one-year study of 75 disease-free Swedish adults found 23% of total compositional variance was within-person — larger for functional pathways than for species — and concluded that reliable quantification requires repeated samples [28].
The counterweight belongs here. Over five years in 37 US adults, stability followed a power law which, extrapolated, suggests most strains in a person are residents for decades [29]. Membership persists; proportions move — a distinction taken up in retesting and longitudinal tracking.
So what can a profile legitimately say?
It can describe a sample: which bacterial groups were present in the material you sent, in what relative proportions, and how that compares with cohorts profiled using comparable methods. Those are real observations — and method-dependent ones, as how gut microbiome testing works sets out.
What it cannot do is place you on a balance scale, because no one has built one. Independent bodies have said as much: an expert policy analysis in Science, co-authored by microbiome scientists and health-law scholars, states that direct-to-consumer microbiome tests lack analytical and clinical validity, requiring more federal oversight to prevent consumer harm [30], and a Lancet Gastroenterology & Hepatology editorial independently calls for regulation [31]. We quote that rather than argue with it.
There is a subtler failure than overclaiming, and it deserves naming: conceding that the science is unsettled while keeping the claim anyway — the definition is coming, and meanwhile here is your balance score. The literature does not support that either. One critique argues, invoking Popper, that the field's next job is to run experiments attempting to refute the role of the commensal gut microbiota in human health and disease [3]. The systematic review of rodent transplants warns that overstatement may undermine the field's credibility: 95% of published studies — 36 of 38 — reported transfer of pathological phenotypes into recipient animals, a rate the reviewers judged implausible and concluded overstates the role of the gut microbiome in human disease [10]. That is a finding about the rodent literature, not about people.
The honest position is the more interesting one. Your gut carries a large, individual, shifting microbial community, shaped mostly by where and how you live and who you live with. Sequencing can describe a sample of it precisely. What nobody can yet do is say whether that description is the right one.