Summary The monitoring of induced seismicity is a common operation in many industrial activities, such as conventional and non-conventional hydrocarbon production or mining and geothermal energy exploitation, to cite a few. During such operations, we generally collect very large and strongly noise-contaminated data sets that require robust and automated analysis procedures. Induced seismicity data sets are often characterized by sequences of multiple events with short interevent times or overlapping events; in these cases, pick-based location methods may struggle to correctly assign picks to phases and events, and errors can lead to missed detections and/or reduced location resolution and incorrect magnitudes, which can have significant consequences if real-time seismicity information are used for risk assessment frameworks. To overcome these issues, different waveform-based methods for the detection and location of microseismicity have been proposed. The main advantages of waveform-based methods is that they appear to perform better and can simultaneously detect and locate seismic events providing high-quality locations in a single step, while the main disadvantage is that they are computationally expensive. Although these methods have been applied to different induced seismicity data sets, an extensive comparison with sophisticated pick-based detection methods is still missing. In this work, we introduce our improved waveform-based detector and we compare its performance with two pick-based detectors implemented within the SeiscomP3 software suite. We test the performance of these three approaches with both synthetic and real data sets related to the induced seismicity sequence at the deep geothermal project in the vicinity of the city of St. Gallen, Switzerland.