Brain Training Made People Better at Brain Games and Nothing Else. 11,430 Volunteers, Six Weeks, and Zero Transfer to General Thinking

The Promise You Paid For

Brain training sells a seductive syllogism: your brain is like a muscle, muscles get stronger with exercise, and therefore a daily workout of memory games and reasoning puzzles should make you sharper at everything, at work, in conversation, and as you age. The industry built on that syllogism promises a smarter you, not a better score on one game. Most buyers never ask the question that decides whether any of it is real. Does getting better at the game make you better at anything else?

That question has a technical name, transfer, and it is the whole ballgame. If practice only improves the practiced task, a brain-training app is an expensive way to learn a party trick. In 2010, a Cambridge team led by Adrian Owen answered it at unmatched scale, recruiting through the BBC program Bang Goes the Theory.

The Largest Test Ever Run

The design was admirably blunt. Volunteers aged 18 to 60 were randomized into three groups. The first group of 4,678 trained on reasoning, planning, and problem solving; the second group of 4,014 practiced tasks resembling commercial brain trainers, covering memory, attention, visuospatial processing, and mathematics; the third group of 2,738 served as an active control, spending their sessions answering obscure general-knowledge questions using the internet; and everyone took the same four benchmark tests before and after the six weeks, covering reasoning, verbal short-term memory, spatial working memory, and paired-associates learning. The minimum dose was ten minutes a day, three times a week; participants averaged 24 sessions.

On the trained tasks, the training worked. The first group improved with effect sizes from 0.73 to 1.63, and the second group from 0.72 to 0.97, gains any researcher would call large. Even the control group improved on its own assigned activity, at 0.33. Then came the benchmarks, and the entire promise collapsed. Improvements on the untrained tests ranged from 0.01 to 0.22 across all three groups, a range so small it is barely distinguishable from nothing, and the control group actually improved slightly more than the trained groups on two of the four measures. Lead author Adrian Owen put it without varnish: there were absolutely no transfer effects. Six weeks of regular computerized brain training conferred no greater benefit than simply answering general-knowledge questions using the internet, and the gains did not even generalize to similar tasks. Nothing transferred.

The Arithmetic Nobody Did

Those two ranges deserve to sit side by side. Take the midpoints: trained-task gains averaged 1.18, benchmark gains 0.12. Divide one by the other: a transfer efficiency of roughly 10 percent. About 90 percent of the improvement evaporated the moment the task changed, even when the new task drew on the same mental faculty, which is not a small leak but an empty bucket.

The dose data sharpen the point: sessions ranged from one to 188, nearly eight times the average, with no relationship between sessions completed and benchmark scores. Age did not matter either. The most diligent volunteers gained no more general ability than the dabblers. The funnel behind the headline number: 52,617 people registered, and 11,430 completed enough of the protocol to be analyzed, a completion rate of 21.7 percent. Four out of five starters never finished, which is itself a quiet verdict on daily brain drills.

The Strongest Counterargument

The fiercest challenge to this verdict does not deny the numbers. It attacks the dose. Critics at the University of Maryland's Center for Advanced Study of Language wrote an open letter arguing Owen's volunteers barely trained at all, since the minimum bar was 20 minutes across six weeks, the average participant logged only about four hours, and the studies that had previously shown transfer typically used roughly 20 hours of supervised practice over a similar period, a gap large enough to make the trial a test of a token dose rather than of training itself. On this reading, the trial refuted a homeopathic dose, administered unsupervised, not brain training itself. Their own evidence cut the other way: 47 participants who trained 20 hours over six weeks on adaptive working-memory tasks improved, and the gains lasted three months past the end of training.

That critique lands, and Owen's own data answer only part of it. The absence of any dose-response relationship in his trial is genuinely awkward for the dose story: if four hours were simply too little, the volunteers who logged 188 sessions should have pulled ahead, and they did not. But the critique exposes a real boundary: the intensive study measured working-memory tasks, not the broad everyday abilities the industry advertises. The ACTIVE trial, the largest cognitive-training trial in older adults, taught strategies rather than drilling games, and its reasoning training was associated with less everyday difficulty a decade later. Drilling a game and learning a strategy may simply be different interventions under one label, which means the honest verdict is narrower than the headline: game-style brain training, at doses people actually complete, does not make you generally smarter.

What We Didn’t Prove

This trial does not prove nothing can improve your thinking. It proves six weeks of game-style drills, done by healthy adults aged 18 to 60 online and unsupervised, produced no measurable general gain. It says nothing direct about children, older adults with cognitive impairment, or clinical rehabilitation. It cannot speak to very long interventions: the flat dose-response covers the range people actually completed, not a hypothetical year of practice. The sample was self-selected BBC viewers, and the dropouts may differ from those who stayed. The 2016 Simons consensus review reached the same verdict for commercial programs, finding extensive evidence for trained-task gains, less evidence for closely related tasks, and little evidence for everyday performance, which means the finding is not a one-trial fluke even though the boundary stays where Owen drew it: the tested claim failed, and other claims were not tested.

What You Can Do

Cancel the subscription that sells general intelligence; keep the games only if you enjoy them. A puzzle you love is a fine hobby, not a workout for your mind the way a run is a workout for your legs. If you want sharper thinking, spend the time on the skill itself. Want a better memory for names? Then practice names at introductions, and if you want faster arithmetic, do arithmetic. Transfer is narrow, so aim practice at the target.

Second, borrow the ACTIVE lesson: learn strategies, not just drills. Spaced retrieval, association, and deliberate note-taking change how you approach a task rather than how fast you click through it. When any cognitive product pitches you, ask Owen's question: show me the untrained test. If the only proof is a higher score inside the app, you are looking at a mirror, not a measurement.