Novel multivariate pattern classification analyses have enabled the prediction of decision outcomes from brain activity prior to decision-makers' reported awareness. These findings are often discussed in relation to the philosophical concept of "free will". We argue that these studies demonstrate the role of unconscious processes in simple free choices, but they do not inform the philosophical debate. Moreover, these findings are difficult to relate to cognitive decision-making models, due to misleading assumptions about random choices. We review evidence suggesting that sequential-sampling models, which assume accumulation of evidence towards a decision threshold, can also be applied to free decisions. If external evidence is eliminated by the task instructions, decision-makers might use alternative, subtle contextual information as evidence, such as their choice history, that is not consciously monitored and usually concealed by the experimental design. We conclude that the investigation of neural activity patterns associated with free decisions should aim to investigate how decisions are jointly a function of internal and external contexts, rather than to resolve the philosophical "free will" debate.