Review of 31 studies finds AI tailoring helps language learners remember new words short term

Adaptive learning systems were the most common technique, used in 32.3% of studies. Chatbot or conversational AI and NLP-based feedback each accounted for 16.1%. Performance metrics were the most frequently used data type, at 54.8%, while affective and cognitive load indicators were rarely used.

The evidence base has clear gaps: 54.8% of studies reported no standardized effect sizes, and 67.7% targeted university-level learners only. The authors call for more rigorous, more demographically varied research to guide learner-centred AI design.

The review asked three questions: which AI personalization techniques are used, what learner data drives the adaptivity, and how effective these approaches are across different learner profiles. It followed PRISMA guidelines and applied thematic and quantitative coding to 31 studies.

Across the studies, adaptive learning systems led the field, followed by chatbots and NLP-based feedback. Most systems relied on performance metrics, while indicators of learners' emotional state and cognitive load were seldom used to drive adaptation.

Findings pointed to short-term gains in vocabulary retention and motivation, but the benefit varied with learner proficiency, the type of task and how cognitive load was managed. The review draws on Cognitive Load Theory, Input Processing Theory and the Noticing Hypothesis, and calls for theoretically grounded, methodologically rigorous and demographically diverse research.