In the version of logic programming (LP) based on interpretations where variables occur in atoms, a goal reduction via unification can be seen as a transition labelled by the most general unifier. Categorically, it is thus natural to model a logic program as a coalgebra. In the paper we represent: (i) goals as the substitutive monoid freely generated by the predicate symbols; (ii) the LTS as the structured coalgebra defined by the SOS rules implicit in the LP semantics; (iii) the bisimulation semantics of a logic program as its image on the final coalgebra.