Positive plant-plant interactions (i.e. facilitation) receive increasing attention as a potentially important driver of community assembly. We conducted a systematic literature review indicating broad support for positive effects of potential facilitator species. However, a large majority of the reviewed studies (83% for field studies, 57% for experiment) share a similar risk of misinterpretation as they assess facilitative effects by comparing plots inhabited by a potential facilitator with randomly placed control plots nearby (paired sampling). As the distribution of facilitator species may itself be environmentally driven, species coexistence caused by facilitation cannot exclusively be separated from environmental effects (habitat sharing). Based on simulated plant communities and sampling protocols, we show how non-random coexistence can occur in the absence of facilitation. This is relevant as both the effect of spatial environmental heterogeneity and of facilitation (stress-gradient hypothesis) are expected to increase with environmental harshness. Nevertheless, 58% of facilitation studies neither undertook measures to minimize potential biases in their sampling approaches nor did they acknowledge such limitations in the discussion. Attention to this problem has significantly decreased in recent years. We propose that facilitation studies could be improved (i) by using random sampling for association studies, (ii) by co-analysing environmental factors, or (iii) by experimentally establishing presumed facilitators. Experimental approaches mimicking facilitative plant characteristics can help to identify facilitation mechanisms. Combining approaches and including functional traits of the involved species should further strengthen inference.