Encouraging impaired limb use during routine and rehabilitation activities after stroke is challenging. We evaluated changes in stroke arm activity related to vibration prompts delivered by a wrist-worn accelerometer (“CueS wristband”). A pilot randomised controlled trial was conducted with adults < 3 months after stroke causing any arm impairment. All received instruction for a self-directed four-week programme encouraging bimanual tasks, and wore a wristband on the impaired side 8am–8pm. Intervention wristbands were programmed to deliver vibration prompts if hourly activity targets were unmet. To allow for incremental recovery, patient preference and diurnal variability, the target was selected as 5%, 10% or 20% above median levels recorded hourly over the previous 3 days. Data were downloaded at twice weekly reviews. Activity counts per minute (CPM) were calculated from CueS mean signal vector magnitude (1 s epoch) at the first (baseline) and the last (endpoint) review. Standard accelerometers recorded natural arm activity over 72 h starting at 4 and 8 weeks post-baseline. Mann Whitney U test compared CPM between groups, and for 1 hour before/after prompts amongst intervention participants. Thirty-three participants (14 intervention; 19 control) enrolled over 16 months: mean age 71 years (SD11); 13 male; mean days post-stroke 30 (SD19); median baseline Action Research Arm Test 20/57 [IQR 3.5,41]. Median number of prompts delivered were 7 per participant/day [IQR: 6,8]. Median CPM during 1 hour before/after prompts was 651 versus 759 (+16.6%; P = 0.002). CPM increased immediately after prompting, suggesting a direct behavioural impact. Compared to control, the intervention group continued to increase CPM beyond wristband removal. Whilst these results are encouraging, this was a small study and data may not reflect only arm activity. Personalised prompts delivered by a wrist-worn accelerometer may enhance self-directed arm activity after stroke.