As artificial intelligence technology is becoming increasingly advanced, the line between natural speech and computer-generated speech is starting to blur. Machines are starting to comprehend nuances in conversation and read between the lines. One of the more complex abilities of natural language is the possibility of saying one thing, but mean something else entirely. As AI assistants, i. e. personal virtual assistants, are made to help users with increasingly advanced tasks and speak to them more naturally, it is important that they are able to understand possible underlying intentions in that regard to fulfil their duty. This paper offers an insight into the degree of indirectness a personal virtual assistant can comprehend without losing the main focus of the sentence. Using traditional pragmatic concepts like speech acts, implicature, convention and the cooperation principle, a theoretical framework is constructed in order to shed some light on the human side of the interaction. As for the machine side, contemporary literature from computer science with an emphasis on natural language processing is reviewed and a rudimentary overview of the main concepts is given, which should help elucidate some of the processes a virtual assistant goes through in order to correctly interpret a sentence. The analysis consists of five basic and frequently used direct commands, which are turned into multiple indirect requests using a different sentence structure, i. e. in the form of a declarative and interrogative sentence with varying degrees of indirectness. The analysis will try to clarify the ways a virtual assistant processes a specific request and, if interpreted incorrectly, where the interpretation went wrong. The importance of a pragmatic framework will also be tested to see to what extent can it be applied in a conversational analysis, where one of the participants is a machine and how it consequentially correlates to the programming of the virtual assistant.