When we last wrote about this ('Before the Forgetting, Something Shifts in the Gut'), the story was a signal: something in the gut looks different in people whose memory is starting to slip. The obvious next question was not another cohort. It was whether the pile of existing cohorts adds up to anything. In January a team at the Frame-Corr Laboratory at the George Washington University School of Medicine and Health Sciences tried to answer that directly. Alison Warren, Zan Wynia, Patrick G. Corr, Leigh A. Frame and colleagues published a scoping review in Alzheimer's & Dementia on 22 January 2026, built on the Arksey and O'Malley framework and reported against the PRISMA extension for scoping reviews. They searched PubMed, Scopus and the Cochrane Library, screened 4,751 records, and found 58 human studies that met their criteria. Animal work was excluded by design, as were non-English papers and studies of other neurodegenerative conditions including Parkinson's disease, frontotemporal dementia, Lewy body dementia and ALS. This is an audit, not an experiment: it counts and characterises what has been done, and does not pool the numbers into a single effect.

The shape of the 58 is the first finding, and it is a quiet one. Of the 33 non-prospective studies, the review reports that 87.9 percent were case-control designs, snapshots comparing people who already have a diagnosis with people who do not. There were two cross-sectional studies and two Mendelian randomisation analyses. There were three prospective cohorts, one of which, by Jiang and colleagues, followed 12 people with early-stage Alzheimer's taking donepezil. There were eight randomised controlled trials in the whole literature the review captured: five testing probiotics, two testing dietary interventions, one testing a mindfulness awareness programme. Five of those eight reported gut microbiota composition at baseline or at the end. Just over half the observational studies came from China (51.6 percent), followed by Japan, with single studies scattered across other countries, and 84.8 percent used 16S ribosomal RNA sequencing. A field that talks constantly about causation has, on this map, almost no longitudinal architecture to carry it.

Then the taxonomy, where the trouble becomes vivid. At phylum level the review reports Pseudomonadota and Actinomycetota more often raised in Alzheimer's, with reduced microbial diversity in some studies but not universally, while Bacillota (formerly Firmicutes) and Bacteroidota were the phyla most commonly enriched in healthy controls. At genus level the list of organisms found enriched in Alzheimer's in more than one study runs to eighteen names. Several of them, including Bifidobacterium, Lactobacillus, Blautia and Eubacterium, also turn up as common across Alzheimer's, mild cognitive impairment and control groups alike, and Bacteroides and Prevotella were enriched in both MCI and controls. In other words, the same genera sit on both sides of the ledger depending on which study you read. Only one organism was consistent at species level: Escherichia coli was the sole species repeatedly reported as enriched in Alzheimer's across more than one study. One species, out of nearly five thousand records screened.

For a family, that inconsistency is not an abstraction. It is the reason a bottle on a pharmacy shelf carrying a Bifidobacterium or Lactobacillus strain cannot honestly claim to be correcting anything, because the review found those very genera reported in people with Alzheimer's, in people with MCI and in cognitively healthy controls, in different studies, by different laboratories, using different sequencing pipelines. The trials themselves were mixed: the review's account of probiotic and dietary interventions describes some trials reporting gains in memory, attention or executive function, or slower loss of grey matter, and others reporting no significant cognitive benefit. The authors' own framing, given to Psychiatric Times, is that such interventions belong as "adjunctive, not primary treatments" for MCI or Alzheimer's. Nothing in the paper tells any individual what to swallow, and the paper does not pretend otherwise.

Two caveats sit inside the review itself and deserve daylight. The literature search ran through February 2023, so a paper published in January 2026 is mapping a field as it stood nearly three years earlier, and the microbiome literature has not been idle since. And the review was corrected on 6 February 2026, though only for a typesetting matter: phylum-level names had been italicised in the abstract and highlights and should not have been. That the sole correction to a review of an entire evidence base concerned italics is, in its small way, the most reassuring thing about it.

What the authors ask for at the end is not a new hypothesis but plumbing: standardised protocols, larger samples, longer follow-up, trials designed to test efficacy rather than describe composition. That is the grammar problem. A field can accumulate sentences for years without ever forming a claim that holds, because each laboratory conjugates its verbs differently: different extraction kits, different sequencing depths, different definitions of who counts as having MCI, different diets in different countries feeding different bacteria. Fifty-eight studies is not a small literature. It is a literature that has not yet agreed on how to speak.