The present study investigates the use of generative strategies and their effect on second language (L2) vocabulary retention and learner confidence in vocabulary knowledge over time. Generative strategies (or generation) are one in a group of learning methods known as desirable difficulties -- strategies that improve long-term learning by adding challenge to the learning process (Bjork & Bjork, 2011). Previous research has established the effectiveness of other desirable difficulties, such as spacing, retrieval, and interleaving, in L2 vocabulary acquisition (Kang et al., 2013; Lafleur & Kanazawa, 2024; Nakata, 2015). However, the effect generative strategies have on L2 vocabulary learning over shorter timeframes (e.g., a few days to a week) has been found to lack efficacy (Barcroft, 2009). Nevertheless, long-term vocabulary outcomes implementing this strategy in L2 learning remain unclear. To address this gap, this study examined the effectiveness of generation tasks used as the sole desirable difficulty for vocabulary recall over two intervals: three weeks and 15 weeks. Using quantitative methods in a classroom action research project, this study investigated 38 first-year Japanese university English language learners at the B1 to B1+ CEFR English proficiency level. The treatments included generation tasks in the form of generating novel sentences for half of the target vocabulary and a control treatment utilizing other non-generative methods for the other half through two separate lessons. Additionally, pre and post-tests were implemented to assess the perceived confidence and actual knowledge of the target vocabulary. The study`s findings on longer-term timeframes were not markedly better than those of shorter timeframes and only showed slight improvement. While the study failed to demonstrate a clear benefit for the additional effort required by generation tasks, it did not identify significant costs from using generation on the specific population studied, as found in other studies. Implications for the findings and future research are included.