This paper analyses how the persuasive form of AI-generated editing suggestions for language and style in academic writing can manipulate the usage of AI-tools from supporting learning processes to outsourcing sub-tasks - especially for multilingual writers. In a qualitative analysis comparing an undergraduate student draft introduction with a fully AI-generated and two AI-edited versions (one where AI-editing has a dominator function, i.e. there are no explanations and the user only needs to control the output and one with an operator function, with marked changes and explanations also supporting critical evaluation and learning processes). The results show that AI-editing as dominator produces a larger number of changes compared to AI-editing as operator, and these changes also affect content and meaning. Together with the persuasive phrasing of AI outputs and the writers’ expectation of efficiency, this contributes to support an uncritical way of usage. This can be problematic for undergraduate multilingual writers in particular, as they still lack the language skills and knowledge about academic writing conventions to be able to critically evaluate the AI-suggested output they are responsible for. In addition to that, the comparison also shows that changes through AI-editing stay on the language surface. Argumentation-related effects or errors indicating the learning level like mentioning academic titles of authors or very generalized method descriptions stay unchanged in the AI-editing process. These errors can serve as valuable indicators to discriminate AI-edited from fully AI-generated texts especially at undergraduate level.