The forced-choice triad task has become increasinglypopular in use over recent years. While it is seen as beinga categorisation task (Lin & Murphy, 2001) variation intask instructions often leads to different results. Shipp,Vallée-Tourangeau, and Anthony (2014) used the triadtask to show that when participants are asked to choosean option that ‘goes best with the target’, they are morelikely to select the choice that shares an action relationwhen it also shares taxonomic information. Howeverusing the instruction to select the item that “goes best” isvague and might encourage a strategy other than acategorical decision. The present experiment used thesame triads as in Shipp et al. to test whether participantswould match items based on shared actions or sharedtaxonomic relations when given specific categorisationinstructions. The task instructions were manipulated sothat participants either selected the item that “goes best”,“goes best to form a category” or is “most similar” to thetarget. The results found instances where the instructionsof “goes best to form a category” led to a higherprobability that participants would select the actionchoices over the instructions of “goes best”. Howeverwhen participants were encouraged to use similarityoverall action choices were lower. Therefore the triadtask does encourage a natural categorisation strategy anddifferences in task instructions across research are aresult of the stimuli used.