CodeA11y: Making AI Coding Assistants Useful for Accessible Web Development
Peya Mowar, Yi-Hao Peng, Jason Wu, Aaron Steinfeld, Jeffrey P. Bigham · 2025
Read CodeA11y: Making AI Coding Assistants Useful for Accessible Web Development by Peya Mowar, Yi-Hao Peng, Jason Wu, Aaron Steinfeld, Jeffrey P. Bigham (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-10884.txt.
Introduction
Peya Mowar and four colleagues teach an AI coding assistant to write web code that disabled users can actually use, in a 2025 paper. Nearly all web pages carry accessibility violations, and novice developers rarely know the specialist tools. AI assistants could bake in good practice, but whether they do was untested.
A study with sixteen untrained developers found three failure modes: they never asked the AI for accessible code, skipped manual fixes like replacing placeholder text, and could not verify compliance. The team's answer, CodeA11y, is an extension for GitHub Copilot that suggests accessible code and pops up reminders for the manual steps. A controlled study with twenty more novices showed it steering developers toward sound practice throughout the session. The authors call it a step toward assistants that treat accessibility as default, not decoration.
Catalog record
| CREATOR | Peya Mowar, Yi-Hao Peng, Jason Wu, Aaron Steinfeld, Jeffrey P. Bigham |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 66 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-10884
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
CodeA11y: Making AI Coding Assistants Useful for Accessible Web Development Peya Mowar, Yi-Hao Peng, Jason Wu, Aaron Steinfeld, Jeffrey P. Bigham Abstract . A persistent challenge in accessible computing is ensuring developers produce web UI code that supports assistive technologies. Despite numerous specialized accessibility tools, novice developers often remain unaware of them, leading to ~96% of web pages that contain accessibility violations. AI coding assistants, such as GitHub Copilot, could offer potential by generating accessibility-compliant code, but their impact remains uncertain ( Mowar et al., 2024 ) . Our formative study with 16 developers without accessibility training revealed three key issues in AI-assisted coding: failure to prompt AI for accessibility, omitting crucial manual steps like replacing placeholder attributes, and the inability to verify compliance. To address these issues, we developed CodeA11y, a GitHub Copilot Extension, that suggests accessibility-compliant code and displays manual validation reminders. We evaluated it through a controlled study with another 20 novice developers. Our findings demonstrate its effectiveness in guiding novice developers by reinforcing accessibility practices throughout interactions, representing a significant step towards integrating accessibility into AI coding assistants. Conference: CHI Conference on Human Factors in Computing Systems; April 26-May 1, 2025; Yokohama, Japan CHI Conference on Human Factors in Computing Systems (CHI ’25), April 26-May 1, 2025, Yokohama, Japan DOI: 10.1145/3706598.3713335 ISBN: 979-8-4007-1394-1/25/04 CCS: Human-centered computing Accessibility design and evaluation methods CCS: Software and its engineering Development frameworks and environments CCS: Human-centered computing Interactive systems and tools Figure 1. CodeA11y is a GitHub Copilot Extension for Accessible Web Development. CodeA11y addresses accessibility limitations of Copilot observed in our study with developers through three features: (1) accessibility-by-default code suggestions, (2) automatic identification of relevant accessibility errors, and (3) reminders to replace placeholders in generated code. Integrating these features directly into AI coding assistants would improve the accessibility of the user interfaces (UIs) developers create. The interface of the extension is displayed along with its features: (1) Accessibility-by-default Code Suggestions, (2) Relevant Error Identification and (3) Code Completion Reminders. CodeA11y: Making AI Coding Assistants Useful for Accessible Web Development 1. Introduction Most websites contain extensive accessibility errors ( WebAIM, 2024 ) , despite decades of investment in standards and guidelines ( Chisholm et al., 2001 ; Caldwell et al., 2008 ) , tools ( Takagi et al., 2003a ; Takagi et al., 2003b ; Bigham et al., 2010 ) , advocacy ( Sloan et al., 2006 ; Martin et al., 2022 ; Pandey and Dong, 2023 ) , and policy. According to a recent analysis by WebAim ( WebAIM, 2024 ) , the homepages of the top million websites each contain 57 accessibility errors on average, including (but not limited to) missing alt-text for images ( Bigham et al., 2006 ; Gleason et al., 2020b ) , inadequate color contrast ( Reinecke et al., 2016 ) , incorrect or missing labels for forms and links ( Islam et al., 2010 ) , and improper use of heading levels ( Brudvik et al., 2008 ) . As a result, many people with disabilities will find it difficult to use these websites effectively and may not be able to use them at all. Front-end web developers ultimately determine the accessibility (or inaccessibility) of the UI code that they write. Getting front-end developers to write more accessible code has proven exceptionally difficult. As Jonathan Lazar et al. wrote twenty years ago in 2004, “Since tools and guidelines are available to help designers and webmasters make their web sites accessible, it is unclear why so many sites remain inaccessible.” ( Lazar et al., 2004 ) . A survey of webmasters at the time indicated that they generally would like to make their web pages accessible but cited a number of reasons they do not: "lack of time, lack of training, lack of managerial support, lack of client support, inadequate software tools, and confusing accessibility guidelines." Sixteen years later, Patel et al. reported remarkably similar results in their 2020 survey of 77 technology professionals ( Patel et al., 2020 ) . Few developers had received formal accessibility training, implementing accessibility was considered confusing, and advocating for accessible development was in conflict with other business goals. Clearly, what we have done so far is not working. We argue that AI coding assistants ( e.g. , Github Copilot ( Git, 2024 ) ) could offer an opportunity to make UI code more accessible. AI coding assistants are already widely adopted, which means that developers do not need to be convinced to use them or to install a specialized tool for accessibility. They produce a wide variety of UI code and are capable enough to both reflect on the quality of arbitrary code and also prompt developers to fix what they are unable to do. This paper explores how AI coding assistants currently help developers create UI code, what problems remain, and presents a system called CodeA11y that shows that AI coding assistants can be made better at enabling developers to improve the accessibility of their UI code. To explore this potential opportunity, we first conducted a user study (Section 3 ) with 16 developers not trained in accessibility to understand how current tools (GitHub Copilot) impact the production of accessible UI code. Our findings (Section 4 ) shows that while Copilot may potentially improve accessibility of UI code, three barriers prevent realization of those improvements: (1) developers may need to explicitly prompt the assistants for accessible code and thus not benefit if they fail to do so, (2) developers may overlook critical manual steps suggested by Copilot, such as replacing placeholders in alternative text for images, and (3) developers may not be able to verify if they fully implemented more complex accessibility enhancements properly. The formative study showed the potential of AI coding assistants to improve the accessibility of UI code, but revealed several gaps that led us to design goals (Section 5 ) for improving AI coding assistants to support accessibility. We then built CodeA11y (Section 6 , Figure 1 ), a GitHub Copilot Extension that addresses the observed gaps by consistently reinforcing accessible development practices throughout the conversational interaction. We evaluated CodeA11y (Section 7 ) with 20 developers, assessing its effectiveness in supporting accessible UI development and gathering insights for further refinement. We found that developers using CodeA11y are significantly more likely to produce accessible UI code. Finally, we reflect on the broader implications of integrating AI coding assistants into accessibility workflows, including the balance between automation and developer education, and the potential for AI tools to shape long-term developer behavior toward accessibility-conscious practices (Section 8 ). The contributions of our paper are: • We conducted a study with 16 developers that uncovered both benefits and limitations of current AI coding assistants for authoring accessible UI code. • CodeA11y 1 1 1 The source code for CodeA11y is available at https://github.com/peyajm29/codea11y/ . : a GitHub Copilot Extension that generates accessible UI code, identifies existing issues and reminds developers to perform manual validation. 2. Related Work Our research builds upon (i) Assessing Web Accessibility, (ii) End-User Accessibility Repair, and (iii) Developer Tools for Accessibility. 2.1. Assessing Web Accessibility From the earliest attempts to set standards and guidelines, web accessibility has been shaped by a