Digital Inclusion Resources

Inclusive AI Development: Why Accessibility Cannot Be an Afterthought

Written by Jeffrey Howard | Aug 21, 2025 3:01:29 PM

Artificial intelligence (AI) is developing at breakneck speed, its urgency reinforcing what’s become an arms race among tech companies. If accessibility isn’t built into the early stages of AI development, however, then the roughly 1 billion people globally living with disabilities will be left behind.

This afterthought approach to accessibility is already experienced by those using assistive technology. Barriers to voice recognition and screen reader optimization are two examples.

“AI-powered tools often struggle with voice recognition accuracy for users with non-standard voice patterns, which are variations that can affect how readily others can understand someone,” explains Shlomit Shteyer, Director of Technical Program Management at Salesforce. “Conditions like speech impediments, neurological disorders, or age-related changes can all affect someone's pattern of speech.”

When tech companies discover accessibility issues later in the AI development process, projects are delayed, while large segments of the population are excluded.

“Most organizations realize they have accessibility issues during the testing phase or even after shipping, largely due to the fast-paced development cycle for AI tools,” she continues. “Identifying these issues late in the cycle can be costly in terms of time and resources, as it often requires significant rework and additional testing to address the problems.”

This not only raises AI development costs, but also harms your relationships with current customers and prevents you from reaching new ones.

“This delay can negatively impact user experience, particularly for individuals relying on assistive technologies, leading to frustration and reduced trust in the product,” she says. “As with any development process, identifying and addressing accessibility issues early is more cost-effective and ensures a more inclusive and user-friendly final product.”

Key Takeaways

  • The AI arms race is leaving people behind. Companies rushing to market are overlooking the needs of the billion people globally living with disabilities, turning revolutionary technology into a barrier rather than a bridge to inclusion.
  • Late-stage accessibility fixes are costly and harmful. Treating accessibility as an afterthought leads to expensive rework, project delays, and frustrated users. As demonstrated by ChatGPT's screen reader failures, even high-profile AI launches can exclude disabled users entirely when accessibility isn't considered from the start. Organizations that discover accessibility issues during testing or after shipping face significant costs in time, resources, and customer trust.
  • Build accessibility into AI development from day one. The “shift left” approach involves engaging people with disabilities (PwD) during research, design, and testing phases to inform better solutions.
  • Accessible AI drives innovation and expands your market. Inclusive design doesn't limit creativity, but enhances it. When development teams include diverse perspectives from PwD, they discover new solutions and approaches that benefit all users while expanding their potential customer base. The business case is clear: Accessible AI reduces costly retrofitting, builds customer loyalty, and opens new market opportunities.
  • The choice is clear: Build AI for everyone or build artificial inclusion. Organizations must decide whether to prioritize speed alone or create technology that shapes an inclusive future.

When Speed Is Prioritized Over User Experience

Companies were expediting the product development process, not giving usability its due attention, long before AI hit the mainstream. Odds are: If you were overlooking the experience of users with disabilities in the past, you’re likely doing so while developing new AI features.

ChatGPT: A Case Study in Rushed AI Accessibility

This rush-to-market mentality has real consequences.

Take ChatGPT, one of the most high-profile AI launches in recent history. Despite attracting millions of users daily when released by OpenAI in 2022, accessibility testing revealed fundamental problems that made the platform nearly unusable for people with disabilities.

When CapTech Consulting tested ChatGPT with screen readers, it discovered multiple serious accessibility failures:

  • The main text input area was inadequately labeled as simply "send a message," its lack of context confusing screen reader users.
  • During ChatGPT's response generation, screen readers get stuck announcing "middle dot, middle dot, comma" repeatedly.
  • Chat history and other clickable elements were missing from keyboard navigation.
  • When responses finished, users weren’t directed to the new content, forcing manual navigation backward through the interface.

Their analysis concluded that ChatGPT "does not meet standard accessibility conformance" and "is not fully capable of working for the masses in its native environment."

OpenAI launched a revolutionary AI tool to massive fanfare, but with basic accessibility issues that should have been identified and resolved during development. Instead, accessibility testing and users with disabilities were treated as an afterthought.

The irony is stark: While ChatGPT has enormous potential to assist PwD, its rushed implementation excluded the very users who could benefit most from AI-powered assistance.

Why Accessibility Cannot Be an Afterthought

In business terms, prioritizing accessibility expands your user base and market reach. It also helps your organization meet its moral responsibility to ensure everybody has access to fully participate in society, including the marketplace and suite of products and services that improve people’s lives.

Salesforce has remained committed to accessibility for more than 25 years, initially driven by compliance with the Americans with Disabilities Act of 1990 (ADA). However, the cloud-based software giant has since made accessibility and digital inclusion central to product development, establishing its product accessibility team in the early 2010s and accessibility support team in 2020.

As a values-forward company, integrating accessibility into product development has become second nature.

“This approach aligns with our core values of trust and equality, fostering a positive brand image and customer loyalty,” Shteyer explains. “Inclusive AI also drives innovation by considering diverse perspectives, leading to better, more versatile products. Ultimately, it enhances user satisfaction and engagement, which are critical for long-term success and competitiveness in the market.”

She emphasizes the strong link between accessibility and usability, viewing them as integral parts of a single, overall user experience.

“We keep the conversation open with our customer all the time,” she says. “Customers will shout back and say, ‘Hey, I cannot do my day job because this agent speaks to me in a language I don't understand.’ It could be a blind person who cannot use the AI Agent properly because it hasn't been designed to be compatible with the screen reader’s functionality, preventing them from understanding or getting value from the tool."

Shifting Left: Integrate Accessibility Early in AI Development

Salesforce refers to its accessibility-first design approach as “shift left,” a memorable term reminding team members to always start with the user experience of people with disabilities in mind.

It’s a proactive approach.

Including PwD is not only a matter of ethical responsibility but also a strategic imperative that drives innovation, expands market reach, and creates a more equitable and accessible world for everyone.

“Diverse perspectives drive greater innovation and better outcomes,” says Bonnie Parisi, Vice President of Product Accessibility at Salesforce. “By ensuring representation within our teams, we avoid a single-lens approach and broaden how we think about design and problem-solving. This means intentionally seeking talent in a wide range of places and ensuring we bring in candidates with different lived experiences.

“Equally important is engaging people with disabilities in user research—whether through testing concepts, designs, or overall experiences—so their insights directly shape the products we build,” she continues. “When we consistently integrate these perspectives, we strengthen our decision-making and create solutions that truly meet a wider set of needs. Within Salesforce, we also amplify this work through initiatives like ‘customer zero’ evaluations, product testing, and business resource groups that support people with disabilities and their allies. Together, these efforts not only make our products more inclusive but also fuel innovation and impact across the business.”

Salesforce uses these diverse focus groups to put designs in front of people with lived experience, showing them mockups and gathering information in interview fashion, learning what works and what doesn’t. This feedback is then baked into the design.

Accessibility-forward organizations like Salesforce will also conduct thorough usability testing later in product development to validate how well their solution meets the needs of people with disabilities.

If you’re serious about digital inclusion, however, you won’t only integrate accessibility insights at set points in AI development, but layer them throughout your entire organization. Salesforce has a full-time team specifically dedicated to product accessibility.

"Our product accessibility team, composed of leading accessibility professionals with decades of expertise, including a dedicated AI and Accessibility Architect, advises and educates our engineering and design teams throughout the AI development process," Shteyer says. "This diverse team, spanning various genders, races, abilities, cultures, and ages, provides invaluable insights that our teams might have missed otherwise, such as identifying specific usability challenges and accessibility barriers that only become apparent through lived experience."

Building AI That Is Truly for Everyone From the Start

The AI revolution is here, and it's moving faster than most organizations can keep pace with. Speed without inclusion isn't progress, however. It's exclusion at scale.

As AI becomes embedded in everything from customer service to creative tools, we have a choice: Build AI that works for everyone, or build AI that leaves a billion people behind.

The path forward is clear, but requires intentional action. Organizations serious about inclusive AI development must fundamentally shift how they approach product development. This means integrating accessibility considerations from the earliest conceptual stages, not retrofitting them after launch.

Build diverse development teams. Include accessibility experts and professionals with disabilities throughout your development process, not just as consultants brought in during testing phases. Their lived experiences reveal usability challenges that even the most well-intentioned teams might miss.

Start with inclusive research. Before writing a single line of code, engage people with disabilities in your research process. Understand how they currently navigate the problems your AI aims to solve. What barriers already exist? What creative workarounds have they developed? These insights don't just prevent accessibility issues, but spark innovation that benefits all users.

Test early and often. Conduct accessibility testing at multiple stages of development, from early prototypes to final products. Use both automated tools and human testing with assistive technologies. What seems functional to a sighted, hearing user might be completely unusable for someone relying on a screen reader or voice recognition.

Establish accessibility standards before development begins. Create clear guidelines for AI accessibility that your teams understand and follow. Make accessibility requirements part of your definition of "done," not optional enhancements to consider later.

The business case is compelling: Accessible AI expands your market reach, reduces costly retrofitting, and builds customer loyalty—but the moral case is even stronger. We're building the technologies that will shape how people work, learn, create, and connect for decades to come. Every AI system launched without accessibility consideration is a missed opportunity to create a more inclusive world.

The companies that win the AI race won't just be the fastest. They'll be the ones who build AI that truly works for everyone. The question isn't whether you can afford to prioritize accessibility in your AI development; it’s whether you can afford not to.

In the end, AI that isn't accessible isn't artificial intelligence, at all—it's artificial inclusion, and that's not the future any of us should be building.

A founding partner of InclusionHub, Salesforce is helping bring greater accessibility and digital inclusion to the professional world. Visit its a11y website to learn more.