Accessible conversational user interfaces: considerations for design
Kate Lister, Tim Coughlan, Francisco Iniesto, Nick Freear, Peter Devine · 2020 · Proceedings of the 17th International Web for All Conference (W4A) · doi:10.1145/3371300.3383343
Summary
This paper presents the first comprehensive review of accessibility considerations for conversational user interfaces (CUIs) — a category that includes chatbots, voice-activated personal assistants (like Amazon Alexa and Apple Siri), interactive voice response (IVR) systems, and virtual assistants. The authors systematically reviewed over 30 sources of guidance, research, reports, and literature on accessible design across nine disability categories: deaf/hearing impairment, visual impairment, mobility/dexterity impairment, mental health, cognitive and learning disabilities, autism spectrum conditions, long-term health/fatigue/pain conditions, speech impairment, and dyslexia/specific learning difficulties. For each category, they extracted guidance relevant to CUI design and identified where existing recommendations are insufficient or conflicting. The paper maps key CUI features — multimodal communication, multi-channel deployment, dialogue-based interaction, hybrid interfaces combining GUI and conversation, and AI-driven logic — against the accessibility needs of each disability group. It also examines how WCAG 2.1 principles (perceivable, operable, understandable, robust) apply to CUIs, noting that many success criteria require reinterpretation for conversational contexts where the interaction flow differs fundamentally from traditional web pages.
Key findings
The review identified 24 specific design considerations for accessible CUIs, organized as open questions requiring further research. Key findings by disability group include: for deaf/hearing users, CUIs must always offer a full-functionality text-based alternative and carry accessibility preferences forward when linking to external content. For visual impairment, text-based chatbots must be screen-reader accessible and support magnification up to 200%, while hybrid CUIs with embedded GUI widgets create particular accessibility risks. For mobility/dexterity impairments, voice-activated CUIs offer significant opportunity but may conflict with co-occurring speech impairments (e.g., after stroke). For mental health, CUIs should reduce "panic triggers" (excessive prompting, intrusive notifications), allow adequate response time, and carefully consider when friction should be deliberately introduced (e.g., delaying impulsive financial transactions for users with bipolar disorder). For cognitive disabilities, CUIs should keep text short and simple, minimize steps, present limited choices one at a time, and allow users to control the interaction pace. For autism, clean uncluttered design, low-arousal color palettes (cream and pastels, not yellow or white), and clearly labeled buttons are essential. For brain fog and fatigue conditions, screen fatigue and light sensitivity require customizable color schemes and no flashing content. For speech impairments, flexible input modality switching mid-conversation is critical. For dyslexia, consistent navigation, plain language, and avoiding italics, block capitals, underlining, and serif fonts are important.
Relevance
As CUIs become ubiquitous — with 25% of customer service operations expected to use virtual assistants — this paper fills a critical gap in accessibility guidance. Most existing accessibility standards and guidelines were designed for visual, page-based web interfaces and do not straightforwardly apply to conversational interactions. The 24 design questions provide a practical framework for developers building accessible chatbots and voice assistants. The paper also introduces the ADMINS project (Assistants to the Disclosure and Management of Information, Needs and Support), a real-world application of these principles: a CUI designed to help university students with disabilities disclose their conditions and organize support through dialogue rather than inaccessible forms. For accessibility practitioners, the cross-disability review highlights that CUIs present both significant opportunities (reducing physical interaction requirements, offering multimodal alternatives, simplifying complex form-filling processes) and risks (AI bias against underrepresented disability groups in training data, unpredictable responses, inaccessible hybrid widgets). The tension between simplifying interactions for some users and providing sufficient information for others underscores the need for flexible, personalized CUI designs.
Tags: conversational user interfaces · chatbots · voice assistants · cognitive accessibility · mental health · autism · inclusive design · universal design
Standards referenced: WCAG 2.1