At collegial courts, cases are decided by panels of judges, and all judges on the panel can potentially influence the text of the final judgment. At the Court of Justice of the European Union (CJEU), published judgments are unsigned, but there is a Judge-Rapporteur who is responsible for writing a preliminary draft of the judgment. How much influence does the Judge-Rapporteur have over the content of the final judgment versus non-Rapporteur judges? The answer to this question is important to our understanding of judicial bargaining, which happens behind closed doors. In this paper, we develop a strategy for estimating the influence of individual judges on the content of published judgments. Using a new text corpus that includes the universe of CJEU judgments (1952-2021), we train machine learning algorithms to learn the style of individual Judge-Rapporteurs. We then identify specific sentences in judgments that are more likely to have been written by a non-Rapporteur judge on the panel than by the assigned Judge-Rapporteur. We use our models’ predictions assess the relative influence of Judge-Rapporteurs and non-Rapporteur judges on the text of judgments. We also identify the observable correlates of Judge-Rapporteurs and non-Rapporteur judges who have systematically more influence than their colleagues.