Text recognition based on machine learning can not only be used to recognize text properly. Some engines also offer the opportunity to output similar readings, e.g., in the form of a confidence matrix. For this article, we use confidence matrices not (as intended) to construct highly reliable text from images but to determine if the output can be leveraged to identify scribes. For the article, we rely on 87 charters from the monastery of Einsiedeln that have been automatically recognized. Based on confidence matrices outputted by HTR+, we demonstrate that visual and semantical similarities are identifiable in the patterns. Using PCA-based dimension reduction, we identify clusters that indicate similarity in the language and, to a certain degree, also in the handwriting. In short, a more in-depth and elaborate analysis would be required to use the approach for scribal identification but already some behavior of text recognition engines are explainable based on the limited data set.