Inadequate conceptual understanding and declining learning motivation remain major challenges in science education. To address these issues, this study implemented a Natural Language Processing (NLP)-based chatbot as a virtual assistant designed to provide adaptive feedback and personalized guidance in science learning. A mixed-methods approach was employed, integrating quantitative and qualitative phases within a quasi-experimental pretest–posttest control group design involving 240 tenth-grade students in Jakarta over eight weeks. Quantitative data from the Science Achievement Test (SAT) and Science Learning Motivation Scale (SLMS) were analyzed using an independent samples t-test, while qualitative data from interviews and learning analytics were used to explain behavioral and motivational changes. The experimental group showed a substantial improvement in conceptual understanding, increasing from a mean pretest score of 42.5 to 88.4, compared to 44.1 to 62.7 in the control group (t(238) = 11.34, p < 0.001, d = 1.56). Motivation scores also increased significantly across all dimensions (p < 0.001), particularly in self-efficacy (η²p = 0.198). Learning analytics indicated higher interaction frequencies and longer engagement times. Students reported five perceived benefits: 24/7 accessibility, personalized explanations, increased questioning confidence, support for complex concept visualization, and stronger self-driven learning motivation. Overall, the NLP-based chatbot effectively enhanced science learning outcomes and motivation.