The use of AI chatbots and apps for therapeutic purposes has multiplied in recent years. While these have allowed multiple users to benefit of mental health help in crisis situations, ethical considerations have arisen regarding the privacy of data shared from users, the responsibility of medical mistreatment and the role of the therapeutic alliance between the patient and the app. This paper will argue that, on top of these ethical shortcomings which are essential elements in the traditional therapeutic setting, this type of approach can also become ineffective if used as more than a tool that needs anyway to be coupled with traditional therapy. The way emotions are a result of cognitive, social and neurological factors will be explored, in particular how our brain perceives emotions. Furthermore, it will examine how emotions are influenced by hormones and chemical signals, and how these, together with the mirror neural system are fundamental to the formation of empathy and comprehension of other people’s emotional state. AI, which lacks this neurophysiological structure, is not able to develop an authentic connection or an empathetic intuition, which are essential in the therapeutic process. For this reason, chatbots cannot substitute human therapists, and if used as sole therapeutic resource can be both inefficient and harmful.