An Accessibility-First Generative AI Chatbot to Support Inclusive MOOC Discovery and Early Onboarding

Massive Open Online Courses (MOOCs) have expanded access to education globally but remain largely inaccessible to learners with disabilities and those facing multilingual barriers. Persistent issues such as insufficient captioning, poor screen reader compatibility, and limited adaptive support constrain equity in participation. This study introduces RECMOOC4All Chatbot, the user-facing interaction interface of the RECMOOC4All MOOC aggregator, designed to reduce friction during course discovery and early onboarding by embedding accessibility at the interaction layer through multimodal (text/voice) support, multilingual mediation, and accessibility-aware prompting based on available course signals. The prototype integrates speech-to-text (STT), optional text-to-speech (TTS), ensemble language identification, neural machine translation, and intent-aware routing across discovery, assistance, and feedback workflows. An exploratory live pilot with 15 MOOC learners (including six reporting disability-related needs) assessed feasibility and early utility for search/refinement, clarification, first-step guidance, and feedback. Results indicate reliable runtime behavior for the targeted tasks: intent routing agreement reached 92.1%, STT achieved a 9.2%-word error rate on a controlled speech set, and language identification precision was 95.82% on a multilingual benchmark; translation fidelity for domain queries was supported by semantic-similarity evaluation (F1 = 0.8962). An offline retrieval comparison showed a 38% precision gain for accessibility-tagged course queries relative to a TF-IDF baseline. Participants reported high perceived usefulness and clarity, and interaction telemetry suggested reduced friction during early discovery across diverse profiles. Findings should be interpreted as feasibility evidence from a small, heterogeneous pilot without a control condition; broader effectiveness and long-term outcomes require larger controlled studies and depend on the completeness of underlying course metadata.