Accessible AI output still needs an accessibility check
AI-generated alt text and headings can miss context. This article provides a five-point accessibility checklist for workers and managers using AI content tools, covering headings, link text, image alternatives, contrast, tables, and reading order.
Why AI-generated content needs a human accessibility check
When you ask an AI tool to draft a report, generate image descriptions, or structure a webpage, the output can look polished. But accessibility is not about appearances. For the one in five Hong Kong workers who rely on screen readers, keyboard navigation, or high-contrast displays, a missing alternative text or a poorly structured heading can make content unusable. The Web Content Accessibility Guidelines (WCAG) 2.2, the international standard for digital accessibility, sets clear requirements. AI tools do not automatically meet them. This article gives you a practical protocol to check five critical areas before you publish or share any AI-generated content.
Check headings for meaning and structure
Headings are the backbone of navigation for assistive technology users. WCAG success criteria 2.4.6 (Headings and Labels) and 1.3.1 (Info and Relationships) require that headings convey the page’s structure. AI tools often generate headings that look correct visually but lack semantic hierarchy. For example, an AI might produce a heading that is styled as a level-2 heading but is coded as a level-3, or it might skip levels entirely. To check, use a heading inspection tool (many free browser extensions exist) or view the HTML source. Ensure that heading levels follow a logical order: one level-1 heading per page, then level-2 for main sections, level-3 for subsections, and so on. Also verify that each heading is descriptive. A heading like “Section 1” is not helpful; “Preparing your tax documents” is. If you are using AI to draft content, always review the heading outline as a screen reader would see it.
Make link text meaningful
WCAG 2.2 success criterion 2.4.4 (Link Purpose – In Context) requires that the purpose of each link can be determined from the link text alone, or from the link text together with its programmatically determined context. AI tools sometimes generate generic link text such as “click here” or “read more.” This is a barrier for screen reader users who often navigate by jumping from link to link. A better practice is to use descriptive text like “Download the 2025 tax guide (PDF, 2 MB).” When you receive AI-generated content, scan every link. If the link text does not clearly state where it leads or what the user will get, rewrite it. Also include file type and size for documents. This simple step improves usability for everyone, especially on mobile devices where data roaming is a concern.
Assess image alternatives for context
Images must have text alternatives that describe the information or function they represent, as per WCAG success criterion 1.1.1 (Non-text Content). AI tools can generate alt text, but they often miss context. The W3C’s Images Tutorial categorises images into informative, decorative, functional, images of text, complex (like graphs), groups, and image maps. Each requires a different approach. For example, a photo of a bird on a parks website might need only a brief description, but on a birdwatching site the alt text should include species, colour, and behaviour. AI-generated alt text may default to a generic label like “bird.” A human must decide whether the image is purely decorative (in which case a null alt attribute, alt="", is correct) or whether it conveys essential information. For complex images such as charts, the alt text should provide a complete text equivalent of the data. Always review AI-generated alt text with the image’s purpose and audience in mind.
Verify colour contrast and text legibility
WCAG 2.2 success criterion 1.4.3 (Contrast – Minimum) requires a contrast ratio of at least 4.5:1 for normal text and 3:1 for large text. AI-generated content may use colour combinations that look stylish but fail this threshold. For example, light grey text on a white background is common in AI-designed templates but is illegible for many users with low vision. Use a free contrast checker tool to measure the ratio between foreground and background colours. Also check that information is not conveyed by colour alone (criterion 1.4.1 – Use of Color). If a chart uses red and green to indicate “stop” and “go,” add text labels or patterns. Hong Kong’s ageing workforce and high smartphone usage make these checks especially relevant for internal communications and customer-facing materials.
Review tables and reading order
Tables must have proper header cells and a logical reading order. WCAG success criterion 1.3.1 (Info and Relationships) applies here. AI tools sometimes generate tables that look correct visually but lack semantic markup. For instance, a table might use for all cells, including headers, meaning a screen reader cannot distinguish between data and labels. To check, view the HTML or use a table inspector. Ensure that header cells use and that the scope attribute (scope="col" or scope="row") is set. Also test the reading order by tabbing through the content with a keyboard. AI-generated content may place elements in a visual order that differs from the logical order, confusing users who navigate sequentially. A simple protocol: open the content in a text-only browser or use a screen reader emulator to listen to the flow. If the order does not make sense, restructure the content.
Practical checklist for managers and workers
Before you publish any AI-generated content, run through this checklist:
- Are headings in a logical hierarchy (h1, h2, h3) and descriptive?
- Does every link have meaningful text (no “click here”)?
- Is every image’s alt text appropriate for its context (informative, decorative, functional, etc.)?
- Does the colour contrast meet at least 4.5:1 for normal text?
- Are table headers properly marked with and scope?
- Does the reading order make sense when navigated by keyboard or screen reader?
Accessibility is not a one-time fix. Make this checklist part of your team’s standard review process. AI can speed up content creation, but only a human can ensure that the output is truly usable by everyone. In Hong Kong’s competitive business environment, accessible content is not just a legal consideration—it is a practical advantage for reaching a wider audience.
Sources and further reading
Guidance and regulations can change; check the current primary sources before applying them to a consequential workplace decision.