Affect Recognition has become a relevant research field in Artificial Intelligence development. Nevertheless, its progress is impeded by poor methodological conduct in psychology, computer science, and, consequently, affective computing. We address this issue by providing a rigorous overview of Emotion Elicitation utilising stimuli datasets in Affect Recognition studies. We identified relevant trials by exploring five electronic databases and other sources. Eligible studies were those reviews identified through the title, abstract and full text, which aimed to include subjects who underwent Emotion Elicitation in laboratory conditions with passive stimuli presentation for Automatic Affect Recognition. Two independent reviewers were involved in each step in the process of identification of eligible studies. The discussion resolved any discrepancies. 16 of 1308 references met the inclusion criteria. The 16 papers reviewed 271 primary studies, in which 3515 participants were examined. We found out that datasets containing video, music, and pictures stimuli are most widely explored, while researchers should focus more on these incorporating audio excerpts. Five of the most frequently analysed emotions are: sadness, anger, happiness, fear and joyfulness. The Elicitation Effectiveness and techniques towards emotion assessment, are not reported by the review authors. We also provide conclusions about the lack of studies concerning Deep Learning methods. All of the included studies were of Critically low quality. Much of the critical information is missing in the reviewed papers, and therefore a comprehensive view on this research area is disturbingly hard to claim.