Urban bikeability assessment has undergone a profound computational transformation over the past two decades: from manual audits to GIS-derived indices, from GPS-validated demand models to computer vision applied to street-level imagery and its integration with geospatial analysis (GeoAI), and toward three-dimensional urban digital twins. This paper reports a scoping review conducted and reported according to the PRISMA extension for Scoping Reviews (PRISMA-ScR). Three open scholarly sources (Semantic Scholar, Crossref, Consensus) were searched using nine concept-cluster query strings; after deduplication and two-stage screening, 425 studies (2003–2025) were included and assigned to three computational generations, understood as methodological strata rather than exclusive publication periods: Generation I denotes static GIS-based indices, exemplified by the Level of Traffic Stress (LTS) framework; Generation II denotes behaviourally validated, demand-model approaches based on GPS telemetry and crowdsourced data; Generation III denotes GeoAI approaches applying computer vision to Street View Imagery (SVI), including automated LTS classification, GAN-based perspective correction, and biosensor fusion. Four persistent gaps emerge across all generations: a standardisation deficit, dynamic-variable blindness, a scalability–equity trade-off, and an unresolved validation gap between index computation and modal-shift outcomes. The evidence suggests a convergence toward hybrid approaches combining GIS network analysis, behavioural demand data, and SVI-derived perceptual indicators, with the urban digital twin as a plausible — but still under-validated — integration platform.