Cognitive offloading (CO) refers to the use of external tools to reduce cognitive demands during information processing. While CO typically enhances initial task performance, it often impairs later memory when the external aid is no longer available. Decisions to offload are closely tied to metacognitive judgments about one’s own cognitive abilities as well as the perceived reliability of, or trust towards, the available external tool. Moreover, CO can inflate memory confidence, creating “knowledge illusions” where individuals overestimate how much they actually remember after offloading. As generative artificial intelligence tools are increasingly used for learning and information retrieval, understanding their cognitive consequences becomes critical. In the present study, we aim to investigate how ChatGPT use interacts with the testing effect, the well-established finding that active, but effortful retrieval practice enhances later memory relative to a restudy opportunity. Using a two-day design, participants on day 1 read educational texts and are then asked to practice their contents by means of restudy or retrieval practice (short-answer questions). Retrieval practice is carried out either without any assistance (no offloading), with the option to use ChatGPT (free-choice offloading), or with mandatory ChatGPT support (forced offloading). Additionally, judgements of learning as well as a measure of trust in ChatGPT are collected. On day 2, participants complete memory tests on all four texts without assistance to assess long-term retention. The results will show whether and how cognitive offloading interacts with desirable difficulties for learning, as they typically become evident in testing effects.