This study examines how the AI tool ChatGPT-4o handles nominalizations in the process of adapting Swedish government texts to plain language (“klarspråk”), in accordance with Section 11 of the Swedish Language Act. The aim is to analyze both the linguistic changes and the consequences for textual content and the visibility of agents. The study is based on systemic functional linguistics (SFL) and uses authentic municipal reports from four Swedish municipalities, processed with standardized prompts. The results show that ChatGPT-4o is cautious in its processing: most nominalizations remain unchanged, while some are replaced with simpler expressions or unpacked into congruent forms. These changes occur primarily in shorter texts and may, in some cases, improve comprehensibility by clarifying processes and making agents visible. At the same time, the handling is inconsistent and sometimes leads to changes in meaning and loss of content, which can weaken the text’s precision and transparency. Overall, the result suggests that ChatGPT-4o may provide some support in plain language work, but it should be used cautiously and only as a complement to professional human text editing.