Randomly assigned workout buddies: only one pair in five kept in touch

This study put the assumption "give people a workout partner and they'll stick with it" to an actual test. The assumption is what broke — randomly assigned buddies mostly stopped talking, and what held the programme together sat somewhere else.

The participants were 10 women with polycystic ovary syndrome (PCOS), mean age 26.6 (SD 8.8). Why that group is the design's starting point: women with PCOS report low exercise motivation driven by stigma-related stress and a lack of social connection, which inhibits exercising in public and group settings.

So the programme moved indoors. Participants received a smart home video device plus bands and a stability ball, and personal trainers supervised two individual sessions a week in real time through the device's video feature, for 12 weeks. On top of that, each participant was randomly paired with a buddy and encouraged to connect weekly.

The programme itself ran well

Delivery metrics were strong: 100% recruitment, 74% attendance, 80% retention, and acceptability rated 4.8 out of 5. Pre-post comparisons showed signals of improvement in depressive symptoms, muscular endurance and waist circumference.

With a one-arm design and 10 participants, this is not the stage for effect sizes. The question was not whether it works but whether it can actually be delivered, and the answer was yes.

Only the buddy system failed

One design element clearly did not survive contact with reality. Among the randomly assigned pairs, consistent contact happened in 20% — one pair in five.

Participants explained the failure themselves. The buddy they wanted was not just anyone but someone at a similar life phase — student with student, mother with mother — layered with the shared identity of the same condition. What they needed was not a partner but a partner they did not have to explain themselves to.

Motivation came from the trainer's side

Participants named three reasons their exercise motivation rose: rapport built with the personal trainer, the trainer's knowledge of PCOS, and the trainer's genuine interest in their fitness progress.

The programme was built on self-determination theory to satisfy three psychological needs — autonomy, competence and relatedness. Reading the results, relatedness was met by the supervisor, not the peer. The data lands on the side of relationships being built from repeated contact and shared context rather than from assignment.

The design principle to take from it

This is data about a feature almost every fitness app ships. "We'll match you with a workout friend" is standard, and this study pins down the condition under which it fails — random matching does not work.

Two conditions did work. First, pair people who share context. Here that meant life phase and diagnosis; in lifting the equivalents are a similar strength-score band, a similar weight class, a similar number of years training. Pairing a beginner with an advanced lifter stalls for exactly the same reason.

Second, the other person has to know where you are. What moved these participants was not encouragement as such but the fact that someone knew how far they had come. The gap between encouragement and competition is covered in competition, not encouragement, is what keeps people training, and what disappears first once supervision ends is in strength fades before muscle when the coaching stops.

Put together, the study proposes a minimum unit for adherence: a record exists, someone sees that record, and that someone is positioned close enough to you to be comparable. Why knowing what to do does not translate into doing it is in that is not an information problem.