AI & Accessibility
How AI tools work for people with disabilities — what helps, what's broken, and what to look for when choosing AI tools that need to be accessible.
AI tools are changing how people with disabilities interact with technology. For some, AI is transformative — voice interfaces, real-time captioning, image descriptions, and text simplification are opening doors that were previously closed. For others, poorly designed AI creates new barriers: inaccessible chat interfaces, missing keyboard navigation, outputs that don't work with screen readers, and voice-only features that exclude deaf users.
This guide covers both sides: where AI is genuinely improving accessibility, and where the industry still has significant gaps.
Where AI Is Helping
Voice Interaction
AI chatbots with voice input and output capabilities can be powerful tools for people with motor disabilities, visual impairments, or conditions that make typing difficult. Voice mode in ChatGPT, Gemini, and Grok allows hands-free interaction with AI. For people who rely on voice as their primary input method, this removes a significant barrier — though quality varies between providers and accents, and Australian English recognition is sometimes less reliable than US English.
Image Description and Alt Text
AI tools can now describe images in detail, generate alt text for web content, and help visually impaired users understand visual information they encounter online. This has practical applications from reading documents and charts to understanding social media content. Most major AI tools (ChatGPT, Gemini, Claude) can analyse uploaded images and provide detailed descriptions.
Text Simplification and Translation
AI can simplify complex text into plain language, which benefits people with cognitive disabilities, learning difficulties, or limited literacy. Government documents, legal text, medical information, and technical content can be made more accessible through AI-powered simplification — though the accuracy of simplified versions should always be verified.
Real-Time Captioning and Transcription
AI-powered captioning in video calls (Microsoft Teams, Google Meet, Zoom) provides real-time text for deaf and hard-of-hearing users. AI transcription services can convert audio content into text. While not perfect, these tools have improved dramatically and are increasingly reliable for Australian accents.
Where AI Falls Short
Chat Interface Accessibility
Many AI chat interfaces are not fully accessible. Common problems include:
- Screen reader compatibility: AI-generated responses often use formatting (code blocks, tables, lists) that screen readers struggle with. The streaming text effect (where responses appear word by word) can be disorienting for screen reader users
- Keyboard navigation: Some AI interfaces lack proper keyboard shortcuts, focus management, or skip links. Users who cannot use a mouse may struggle to navigate conversation history or access settings
- Colour contrast: Interface elements, particularly status indicators and subtle UI distinctions, don't always meet WCAG AA contrast requirements
- CAPTCHA barriers: Some AI tools use visual CAPTCHA or verification methods that are inaccessible to users with visual impairments
Output Accessibility
AI-generated content is not always accessible:
- Images without alt text: AI-generated images (from DALL-E, Midjourney, or built-in image generators) typically don't include alt text descriptions — ironic given AI's ability to generate them
- Complex formatting: AI responses that include tables, code, mathematical notation, or nested lists may not convey their structure properly through assistive technology
- Document generation: AI-generated documents (PDFs, presentations, spreadsheets) often lack the semantic structure and tagging required for accessibility
Bias in AI Accessibility
AI models can exhibit bias against people with disabilities:
- Training data may underrepresent disability perspectives, leading to stereotypical or inaccurate responses about disability
- AI hiring tools have been documented discriminating against candidates with disabilities (see our AI in Hiring & Discrimination guide)
- Voice recognition accuracy drops significantly for users with speech disabilities, non-standard speech patterns, or who use augmentative and alternative communication (AAC) devices
Evaluating AI Tools for Accessibility
If you or your organisation needs AI tools that work for people with disabilities, here's what to assess:
AI as an Accessibility Aid
If you have a disability, AI tools can be actively used to improve accessibility in your daily life:
- Ask AI to simplify text: Paste any complex document and ask the AI to rewrite it in plain language or at a specific reading level
- Describe images: Upload images to an AI tool and ask for a detailed description. This works for everything from social media photos to product images to charts
- Generate alt text: If you create web content, use AI to generate alt text for your images. Review the output for accuracy before publishing
- Transcribe and summarise: Use AI to transcribe audio or video content, then summarise key points
- Format conversion: Ask AI to convert content between formats — for example, turning a table into a descriptive list, or converting bullet points into flowing text
Australian Context
The Disability Discrimination Act 1992 (DDA) requires that services, including digital services, be accessible to people with disabilities. While the DDA doesn't specifically address AI tools, its principles apply: if your organisation deploys AI tools that customers or staff with disabilities need to use, those tools must be accessible.
The Australian Human Rights Commission has published guidance on digital accessibility, and the OAIC's APP 1 requirements mean organisations must consider the accessibility implications of AI-powered services that handle personal information.
Draft AI Safety Guidelines for Health Contexts
Copy this prompt into any AI tool. Fill in the [BRACKETS] with your details.
You are creating AI safety guidelines for [INSERT HEALTH ORGANISATION/PRACTICE]. Draft practical guidance covering: (1) what health information should never be entered into commercial AI tools, (2) de-identification requirements before using AI for clinical support, (3) TGA and APRA requirements for AI in healthcare settings, (4) patient communication about AI use in their care, (5) specific privacy obligations for health data under the Privacy Act. Include scenario-based examples for clinicians. Format as a clinical governance policy.