Objective To compare the immediate response of nursing home residents to dog visits with or without an activity, and the impact of cognitive ability. Methods In a randomly controlled trial, 174 nursing home residents were allocated to 12 bi-weekly 10-minute visits: Either ordinary dog visits (D, n = 57, 49 analysed), dog visits with an activity (DA, n = 56, 48 analysed), or visits with activity but no dog (A, n = 61, 54 analysed). We recorded frequency and duration of residents' verbal and physical interactions with the dog and persons. Data were analysed in three periods of four visits (period 1-3) as binomial variables (generalised linear models) or durations (non-parametric statistics). Results Both visit type and impairment level affected the likelihood of interacting with the dog (D and DA). In some periods increased cognitive impairment lowered odds of touching the dog in DA visits (period 1: F1,85 = 5.17, P < 0.05) and talking to it directly (period 1: F1,90 = 4.60, P < 0.05; period 3: F1,87 = 5.34, P < 0.05). Throughout, residents talked less to persons during DA visits compared to D and A (P = 0.01-0.05), and level of cognitive impairment correlated negatively with talk duration (P < 0.001). Generally, high cognitive impairment level lowered odds of interacting with (period 1: F1,89 = 7.89, P < 0.01; period 2: F1,97 = 6.76, P = 0.01; period 3: F1,92 = 13.57, P < 0.001) and talking about the activities (period 1: F1,89 = 13.78, P <0.001; period 2: F1,88 = 3.27, P = 0.07; period 3: F1,86 = 3.88, P = 0.05). Conclusion Visits without specific activities stimulated residents to interact with the dog, whereas increasing the complexity of dog visits by adding activities resulted in less interaction with the dog for severely impaired residents. The optimal dog visit for the less cognitively impaired residents could include activities and thereby a possibility to interact with the dog in different ways, whereas for severely impaired residents, just being with the dog seems more appropriate.