A Data-Driven AI Co-designer for Inclusive Web Interfaces: From Accessibility Audits to Design Actions

Abstract. We present a text-based AI co-designer that translates automated audits (Lighthouse, WAVE) and an interdisciplinary knowledge base on WCAG 2.2 (Web Content Accessibility Guidelines), Universal Design, and ethics into ready-to-use redesign artefacts: annotated wireframes, contrast-checked palettes, and role-specific micro-tutorials. The architecture normalizes audit evidence, maps it to WCAG criteria, clusters errors, and activates playbooks with auditable rationales. Based on a large-scale crawl of ~1,000 sites (~25,000 violations), dominant issues included missing alt text, low contrast, and non-descriptive links. Findings show that audit data combined with inclusive knowledge can generate actionable out-puts with traceable rationales (RQ1); recommendations mix universal safeguards (keyboard access, visible focus) with domain-specific fixes (forms, multilingual content, media) (RQ2); and artefacts can be integrated into design thinking work-flows and CI/CD pipelines, supported by participatory validation with people with disabilities (RQ3). Aligned with the EIDD – Design for All – Gaia Declaration, the assistant advances accessibility from technical compliance toward inclusive, culturally sensitive, and socially responsible outcomes.