Abstract Background Nursing home (NH) residents with severe dementia use many medications, sometimes inappropriately within a comfort care approach. Medications should be regularly reviewed and eventually deprescribed. This pragmatic, controlled trial assessed the effect of an interprofessional knowledge exchange (KE) intervention to decrease medication load and the use of medications of questionable benefit among these residents. Methods A 6-month intervention was performed in 4 NHs in the Quebec City area, while 3 NHs, with comparable admissions criteria, served as controls. Published lists of “mostly”, “sometimes” or “exceptionally” appropriate medications, tailored for NH residents with severe dementia, were used. The intervention included 1) information for participants’ families about medication use in severe dementia; 2) a 90-min KE session for NH nurses, pharmacists, and physicians; 3) medication reviews by NH pharmacists using the lists; 4) discussions on recommended changes with nurses and physicians. Participants’ levels of agitation and pain were evaluated using validated scales at baseline and the end of follow-up. Results Seven (7) NHs and 123 participants were included for study. The mean number of regular medications per participant decreased from 7.1 to 6.6 in the intervention, and from 7.7 to 5.9 in the control NHs (p-value for the difference in differences test: < 0.05). Levels of agitation decreased by 8.3% in the intervention, and by 1.4% in the control NHs (p = 0.026); pain levels decreased by 12.6% in the intervention and increased by 7% in the control NHs (p = 0.049). Proportions of participants receiving regular medications deemed only exceptionally appropriate decreased from 19 to 17% (p = 0.43) in the intervention and from 28 to 21% (p = 0.007) in the control NHs (p = 0.22). The mean numbers of regular daily antipsychotics per participant fell from 0.64 to 0.58 in the intervention and from 0.39 to 0.30 in the control NHs (p = 0.27). Conclusions This interprofessional intervention to reduce inappropriate medication use in NH residents with severe dementia decreased medication load in both intervention and control NHs, without important concomitant increase in agitation, but mixed effects on pain levels. Practice changes and heterogeneity within these 7 NHs, and a ceiling effect in medication optimization likely interfered with the intervention. Trial registration The study is registered at ClinicalTrials.gov: # NCT05155748 (first registration 03–10-2017).