Why the probiotic worked for your friend, not you

You are asking why the same capsule that seemed to rebuild your friend's month did nothing at all for you. It is the best question that reaches me, because the honest answer is not that one of you is imagining things. You may both be right — and there is published work that shows how.

"Probiotic" is a category, not a specification — a label, not a recipe

The working definition — live microorganisms which, when administered in adequate amounts, confer a health benefit on the host — was examined and reaffirmed by an expert panel convened by the International Scientific Association for Probiotics and Prebiotics ⁶. Read it closely and notice what it does not promise. It names no organism, no amount, and no person.

McFarland and colleagues went looking for the consequence. Searching the literature from 1970 to 2017 ³, they assembled 228 randomised controlled trials covering 25 identifiable probiotics — 15 single strains and 10 mixtures ³. Efficacy turned out to be strain-specific and disease-specific: 7 of 10 types tested for prevention, and 11 of 17 tested for treatment, showed significant benefit ³. The rest did not. Among the six Lactobacillus-containing probiotic types pooled for prevention of antibiotic-associated diarrhoea in adults, only four showed significant efficacy — L. rhamnosus GG was one of the two that did not ³.

So "my friend takes a probiotic" carries about as much information as "my friend takes a pill". If the code after the species name on their jar — the letters and digits, GG or R0052 or LRCC5282 — does not match the code on yours, the two of you did not run the same experiment. You ran two.

The gut you started with

Even an identical strain does not walk into an identical room. Zmora and colleagues gave healthy volunteers an 11-strain probiotic combination ² and then looked directly, by endoscopy, at whether it had settled on the gut lining rather than merely passed through. Some people were permissive, some resistant, and the pattern was person-, region- and strain-specific — predictable from host and microbiome features measured before the first capsule ². The detail that should end a great many arguments: whether the strain showed up in stool did not distinguish the colonised from the uncolonised ². Shedding is not settling.

The starting communities differ more than most people expect. Sequencing the gut bacteria of 292 Danish adults, Le Chatelier and colleagues found that roughly 23 % carried a low count of bacterial genes ⁷, and that this low-richness group carried more body fat, more insulin resistance and more inflammation than the rest ⁷.

The responder finding — who it did work for — and how big it actually is

The clearest test I know of arrived this year. In a 12-week randomised, double-blind trial, 120 overweight adults took either a heat-killed Lactiplantibacillus plantarum preparation or placebo ¹. Across the whole group, the answer was nothing — no clinical outcome separated the arms ¹. Then the authors went back to the stool samples taken before the first sachet and split everyone by how diverse their gut had been at baseline. In the low-diversity half, weight fell 1.98 kg against 0.95 kg on placebo, BMI 0.74 points against 0.31, and leptin 2.07 ng/mL against 0.24 ¹. In the high-diversity half, nothing coherent appeared ¹.

Now the caveats, out loud, because they matter more than the result. The primary endpoint — total body fat on a DXA scan — was not met ¹. The diversity split was a secondary analysis in subgroups of about 25 people ¹. And the sponsor developed the strain ¹. I wrote the trial up in full in the news item on baseline diversity. It is a hypothesis worth testing properly, not a finding to reorganise your shopping around.

Dose and duration are not decoration

Ouwehand reviewed what is known about how much you need. For antibiotic-associated diarrhoea, both the meta-analyses and the studies that compared two doses head-to-head found more organisms meant more protection ⁴. For necrotising enterocolitis, prevention of atopic dermatitis, slow intestinal transit and relief of irritable bowel symptoms, no dose–response could be identified at all ⁴ — which, as the review is careful to say, does not prove there is none; the data are simply too thin to conclude ⁴.

Duration is blunter. Thirty healthy adults took 30 × 10⁹ CFU a day for two weeks ⁵. The strains appeared in stool 1–2 days after the first capsule ⁵ and cleared between 3.4 and 8.5 days after the last, depending on which strain ⁵. Nothing here is permanent tenancy. Stop, and within roughly a week the strains are no longer detectable in stool.

What you can actually check about yourself

Four things, and none of them need a laboratory.

The strain code. Read the full designation on both jars. Same species, different strain, different evidence.

The dose and the schedule. Find the trial the claim rests on, and check that your amount and your duration match it. If the trial ran 12 weeks and you ran four days, you have not tested the product; you have tested your patience.

Your own transit. In the persistence study, whole gut transit sorted the volunteers into three groups averaging 28.2, 47.3 and 75.6 hours ⁵. Persistence did not simply rise with transit time: only one of the five strains, L. helveticus R0175, differed significantly between groups, and it lingered longest in the intermediate group, not the slowest one ⁵. So transit is a real variable, and one you can observe without a kit — but it is not a dial that predicts how long a strain stays.

What you were comparing. Your friend changed a capsule, and possibly also their sleep, fibre and week. So did you.

What I cannot tell you

Which half you are in. The 2026 trial cut its groups using a composite of five diversity indices, ranked within each treatment arm and split at that arm's median ¹ — a boundary no consumer stool test reproduces. And I cannot tell you that any product on a shelf will do for you what a dead L. plantarum did for two dozen adults in Korea.

What I can tell you is that a null result across a whole group — nothing at all separating the people who took it from the people who did not — can hide a minority who genuinely responded, and a glowing anecdote can hide a majority who felt nothing. Your friend is one data point. So are you. Neither of you is lying.

— Magnus