When a Promising Training Fails to Beat the Control Group

The question here is not whether inhibitory-control training is theoretically interesting. It is whether stand-alone, behavior-specific ICT changes unhealthy behavior in the adult clinical and at-risk populations for whom the behavior is already a serious problem. The 2026 meta-analysis gives a cautious answer: the training did not establish a reliable advantage over its control conditions immediately after training or at follow-up. Understanding what that means requires keeping the population, comparator, outcome, and delivery context in view.

What the meta-analysis actually tested

The review focused on adults aged 18 or older with clinical or at-risk conditions associated with excessive food consumption, alcohol use, or cigarette smoking. Eligible samples included people with conditions such as overweight or obesity, binge eating disorder or bulimia nervosa, alcohol use disorder or heavy drinking habits, and nicotine or tobacco use disorders. The inclusion criteria therefore placed the question closer to treatment and prevention for high-risk behavior than to a general self-improvement exercise for healthy volunteers.

The intervention had to be an inhibitory-control task based on a Go/No-Go, Stop Signal, or Antisaccade paradigm. The included studies used behavior-specific cues: food-related stimuli for eating behavior, alcohol cues for drinking, and cigarette cues for smoking. The control conditions varied. Many were active computer tasks, including neutral ICT with unrelated stimuli or tasks that showed behavior-related images without requiring the same inhibitory response. A smaller number used treatment as usual, informational brochures, or a waiting list. This mix is central to interpreting the result because the question is not only “Did behavior change?” but “Did behavior change more than it did under this particular comparison?”

The abstract and detailed Results section report 15 articles, 16 studies, and 1,216 participants. They describe nine studies focused on eating behavior, four on alcohol consumption, and three on cigarette smoking. The Discussion section later reports 17 studies and 1,398 participants. The supplied article does not resolve that discrepancy. The earlier counts are more consistent with the abstract, selection results, and behavior breakdown, but the disagreement remains a limitation of the source. Published interpretation should therefore avoid pretending that the total is more settled than the report allows.

The review also deliberately excluded studies in which ICT was combined with another structured intervention such as cognitive-behavioral therapy or non-invasive brain stimulation. That decision has a benefit: it isolates the evidence for ICT as a stand-alone intervention. It also narrows the conclusion. A null result for training delivered alone cannot tell us whether the same task would add value to psychotherapy, medical care, inpatient treatment, or another established program.

The review used random-effects meta-analysis to compare standardized differences between experimental and control conditions. The primary outcomes were behavioral, not merely task performance. The authors examined effects immediately after training and, where data were available, at follow-up. They also tested whether the target behavior, training paradigm, setting, duration, or number of sessions moderated the result. This makes the review more informative than a collection of isolated task studies, while the small evidence base and varied designs still limit precision.

At post-training, the overall mean weighted effect was Hedges' g = −0.0896, with p = .121 and a 95% confidence interval from −0.2028 to 0.0237. The reported heterogeneity was low and non-significant, with I² = 0%.