Research paper · RP38

Can You Bond With a Bot?

Working adults in India are already talking to software about stress, sleep and self-talk. The open question is whether that conversation forms a working alliance — agreement on goals and tasks, plus a felt bond — and whether the alliance predicts change the way it does with a human coach. Across large observational cohorts of adult users of conversational agents and unguided apps, mean scores on a 12-item working-alliance short form sat between 3.36 and 3.75 on a 1–5 scale within five days of first use, close to published human outpatient means of about 3.8. Bond scores often arrived first and highest; task agreement lagged. Four independent samples linked later alliance to better outcomes, with small coefficients (standardised betas about −0.13 to −0.27 for distress; one partial r of −0.18). The human-coach benchmark remains r = 0.28. Alliance measured in week one did not predict change; alliance measured in weeks three or four did. No CTRI-registered trial has yet measured software-alliance in Indian working adults. Industry-affiliated samples dominate the largest means. Certainty is low for formation and speed, and very low for the outcome link. For an Indian workplace programme the practical reading is this: a bond with an agent can appear in days, not months; it is not a substitute for a human hand-off; and task-and-goal agreement, not warmth alone, is the piece most worth designing for.