Test, Evaluate, and Advocate for Accessible AI

Learning Objectives

After completing this unit, you’ll be able to:

Apply accessibility testing methods to AI products to ensure compliance with Web Content Accessibility Guidelines 2.2.
Identify strategies to prevent AI from creating new accessibility barriers.
Promote an inclusive culture within the AI development cycle.

Test AI Products for Accessibility

Now that you’ve learned how to design your AI experiences with accessibility (a11y) in mind, make sure to set aside sufficient time to test how your products work with assistive technologies. You want them to meet Web Content Accessibility Guidelines 2.2 (WCAG) standards, so you ensure they’re genuinely accessible and usable for everyone. Testing ensures compliance and helps you know whether a product works well for real people in real-world scenarios. While WCAG might not cover all emerging AI interaction patterns, its principles serve as a foundational basis for testing the accessible experience of any technology.

Use these top methods for testing any digital experience to ensure it’s accessible for everyone.

Keyboard and Visual Focus

Ensure all interactive components (such as buttons, links, and form fields) are fully operable and accessible using only the keyboard. Ensure users can navigate to these components using the Tab key, navigate backward with Shift + Tab, and activate them using the Enter or Space keys.
Confirm the keyboard tab order is logical and predictable.
Ensure a visible focus indicator that meets 3:1 contrast is present on all interactive elements as the user tabs through the interface.

Basic Screen Reader and Semantic Structure

Test the interface and the AI’s output using a screen reader such as Job Access With Speech (JAWS), Non-Visual Desktop Access (NVDA), and VoiceOver.
When using a screen reader, ensure it announces the right semantic HTML elements (such as headings, lists, paragraphs, buttons, and links). This helps people who use assistive technology easily find and understand the content.
Validate the implementation of aria-live regions. Observe how screen readers announce dynamic updates, such as streaming or real-time AI agent responses, to ensure prompt and accurate user notification.

Images and AI-Generated Visuals

For all images, ensure a meaningful and accurate alternative (alt) text is present. Use tools such as AxeDev Tools or search manually through the code to locate alt text.
If the tool automatically generates alt text, manually review it for accuracy. Ensure there’s a mechanism for human overrides of alt text in case an admin wants to edit them later.

Forms (Input and Output)

Ensure all form fields, especially those used for prompting AI, include a visible and persistent label that clearly describes their purpose. Using placeholder text alone is insufficient.

Resize and Reflow

Verify that the interface is fully responsive and reflows at smaller viewport sizes, such as a mobile device.
Test the layout with an up to 200% text zoom to ensure no content is obscured or cut off.

Be an Advocate

In a previous unit, you examined the key challenges and limitations of AI as they relate to a11y. AI offers great potential for people with disabilities. To ensure its success, pair adoption with a commitment to responsible design and development. This section recaps the four core challenge areas and offers proactive strategies that you can implement to ensure AI serves as a powerful tool, not a new barrier.

Challenge	What You Can Do
Bias and fairness issues: Because the training data for AI models often lacks representation from diverse populations, these models can reflect and amplify existing biases.	Use diverse datasets for AI training, such as incorporating a variety of speech samples, visual data, and usage patterns. This inclusion should reflect a wide variety of users, including those with disabilities and those who utilize different assistive technologies. Include people with disabilities in user testing. Implement continuous monitoring of AI outputs for bias.
Accuracy and reliability problems: AI is prone to hallucinations, making errors difficult to detect and potentially harmful, a particular a11y concern when it comes to generating alt text.	Verify AI-generated content (such as alt text or code) to ensure accuracy and compliance with a11y standards. Incorporate routine audits and validation testing into team workflows and emphasize meeting WCAG 2.2 criteria.
Human oversight gaps: There is a significant risk of overreliance on AI without adequate validation or clear guidelines for reviewing content.	Ensure there’s always humans in the loop—implement required human review checkpoints for all AI-generated content that could affect a11y. Define an escalation path where AI output might negatively impact a product’s a11y.
Organizational and adoption barriers: Uncertainty and uneven adoption across teams slows down the integration of accessible AI principles and tools.	Mandate shift-left accessibility and make a11y training a requirement for all teams developing products, not just AI-driven features. Celebrate and share case studies demonstrating successful implementations of accessible products and AI features across different teams.

You now have a foundational understanding of digital a11y and AI. You’ve explored core concepts, the challenges of AI biases, and best practices for designing and testing accessible AI experiences. Most importantly, you help Salesforce uphold its commitment to fairness and inclusion for all Trailblazers.

Tempo estimado

Tópicos

Procurando ajuda?

Agentforce Recursos