Significance without size: how supplement trials report their results — a meta-research analysis of a curated corpus of highly cited supplement studies

Supplement clinical-trial abstracts are read by clinicians, journalists, and answer engines, yet whether they report the size of an effect — not merely its statistical significance — has not been measured. We analysed the abstracts of 2,481 randomised controlled trials drawn from a curated citation list of highly cited supplement studies, classified by PubMed's own PublicationType, using deterministic text detectors for p-values, confidence/credible intervals, and named effect-size measures. Among RCT abstracts, 63.8% report a numeric p-value (95% CI 61.9–65.7) but only 16.2% report an interval estimate (14.8–17.7) and 10.5% a named effect measure (9.3–11.7): a supplement trial is roughly four times more likely to tell you an effect cleared a significance threshold than to tell you its magnitude. The gap is a reporting choice, not a design constraint — pooled meta-analyses on the same compounds report an interval 75% of the time. Detectors were validated by an independent LLM second pass (n=100) and by blind human hand-coding (n=30; Cohen's kappa 0.93–1.00 across endpoints) and are net-conservative, so the reported rates are lower bounds on true reporting. Disclosure: produced by SupStack, a commercial evidence-based supplement website, from its own curated citation list, partly to create a citable reference — a conflict of interest and a selection bias; every rate is therefore an upper bound for a positively-selected sample and is not a claim about the supplement literature as a whole. No efficacy or safety claim is made. Data (PMIDs and derived boolean flags) and analysis code are released; PubMed abstract text is not redistributed.