Augmented Reality is being increasingly used in industrial training, helping to improve learning and task performance. In recent years, various approaches have been developed to create training assembly frameworks that are accessible to industry experts. This study aims to identify these frameworks, analyze the most used approaches, their advantages and disadvantages, and understand how they benefit industrial worker training. For this, a systematic review was conducted in the main databases, defining four research questions and applying inclusion and exclusion criteria. In the final stage, an additional Snowballing process was used to expand the results and include relevant articles in the analysis. The findings shows that AR enhances knowledge retention and reduces errors, though challenges like object tracking and hardware limitations remain. The review found a trend in using the Authoring-by-Demonstration approach, where training is created based on demonstrations. This method showed the most promising results, especially with the rapid advancements in artificial intelligence.