[en] Non-target screening (NTS) methods are rapidly gaining in popularity, empowering researchers to search for an ever-increasing number of chemicals. Given this possibility, communicating the confidence of identification in an automated, concise and unambiguous manner is becoming increasingly important. In this study, we compiled several pieces of evidence necessary for communicating NTS identification confidence and developed a machine learning approach for classification of the identifications as reliable and unreliable. The machine learning approach was trained using data generated by four laboratories equipped with different instrumentation. The model discarded substances with insufficient identification evidence efficiently, while revealing the relevance of different parameters for identification. Based on these results, a harmonized IP-based system is proposed. This new NTS-oriented system is compatible with the currently widely used five level system. It increases the precision in reporting and the reproducibility of current approaches via the inclusion of evidence scores, while being suitable for automation.