Science's main norms prescribe scientists to use citations as acknowledgements of cognitive content irrespective of geographical location. Previous studies, however, suggested that there is a considerable geographical bias in scientific citations. We argue that this geographical bias does not, in itself, falsify the notion that citations reflect acknowledgement of cognitive content, because cognitively related knowledge may be geographically concentrated as well. We analyse the role of organizational, regional and national co-location on citation likelihood for 5.5 million article pairs, and find that the geographical bias in citations is weak once cognitive relatedness is accounted for. Furthermore, we find that the effect of co-location on citation likelihood is strongest at the organizational level, weaker at the regional level, and weakest at the national level. In addition, we show that geographical co-location particularly increases the citation likelihood between two papers when knowledge relatedness between articles is low, suggesting that interdisciplinary research benefits most from co-location. Finally, we find that, when knowledge relatedness is high, the effect of geographical co-location on citation likelihood is non-existent. We discuss the implications regarding policies aimed to discourage strategic citations and to foster interdisciplinary research.