In 2014 leading publishers withdrew more than 120 nonsensical publications automatically generated with the SCIgen program. Casual observations suggested that similar problematic papers are still published and sold, without follow‐up retractions. No systematic screening has been performed and the prevalence of such nonsensical publications in the scientific literature is unknown. Our contribution is 2‐fold. First, we designed a detector that combs the scientific literature for grammar‐based computer‐generated papers. Applied to SCIgen, it has a 83.6% precision. Second, we performed a scientometric study of the 243 detected SCIgen‐papers from 19 publishers. We estimate the prevalence of SCIgen‐papers to be 75 per million papers in Information and Computing Sciences. Only 19% of the 243 problematic papers were dealt with: formal retraction (12) or silent removal (34). Publishers still serve and sometimes sell the remaining 197 papers without any caveat. We found evidence of citation manipulation via edited SCIgen bibliographies. This work reveals metric gaming up to the point of absurdity: fraudsters publish nonsensical algorithmically generated papers featuring genuine references. It stresses the need to screen papers for nonsense before peer‐review and chase citation manipulation in published papers. Overall, this is yet another illustration of the harmful effects of the pressure to publish or perish.

Background: The increasing use of preprints to disseminate evidence on the effect of interventions for the coronavirus disease 2019 (COVID-19) can lead to multiple evidence sources for a single study, which may differ in the reported evidence. We aim to describe the proportion of evidence on the effect of interventions for COVID-19 from preprints and journal articles and map changes in evidence between and within different sources reporting on the same study. Methods: Meta-research study. We screened the Cochrane living systematic review and network meta-analysis (COVID-NMA) database to identify all preprints and journal articles on all studies assessing interventions for COVID-19 published up to 15 August 2020. We compared all evidence sources (i.e., preprint and associated journal article) and the first and latest versions of preprints for each study to identify changes in two evidence components: study results (e.g., numeric change in hazard ratio, odds ratio, event rate, or change in p value > or < 0.05 in any outcome) and abstract conclusions (classified as positive, negative or neutral regarding the intervention effect, and as reporting uncertainty in the findings or not). Changes in study results were further classified as important changes if they (1) represented a change in any effect estimate by ≥ 10% and/or (2) led to a change in the p value crossing the threshold of 0.05. Results: We identified 556 studies. In total, 338 (61%) had been reported in a preprint: 66 (20%) of these had an associated journal article (median time to publication 76 days [interquartile range (IQR) 55–106]) and 91 (27%) had > 1 preprint version. A total of 139 studies (25% of the overall sample) were reported in multiple evidence sources or versions of the same source: for 63 (45%), there was a change in at least one evidence component between or within sources (42 [30%] had a change in study results, and in 29 [21%] the change was classified as important; 33 [24%] had a change in the abstract conclusion). For studies with both a preprint and an article, a median of 29% (IQR 14–50) of total citations were attributed to the preprint instead of the article. Conclusions: Results on the effect of interventions for COVID-19 are often reported in multiple evidence sources or source versions for a single study. Evidence is not stable between and within evidence sources. Real-time linkage of all sources per study could help to keep systematic reviews up-to-date.