Catch Barriers Before Users Do: Automated Accessibility Testing

AI-Powered Accessibility Testing and WCAG Compliance

Part 1: Understanding What AI Can (and Can't) Catch

What made the Domino's situation particularly painful was that many of the failures on their site were the kind that automated tools could have flagged in minutes. Missing image descriptions. Form fields without labels. Buttons with no readable text. These aren't complex design problems, they're checklist failures. The Web Content Accessibility Guidelines, known as WCAG, exist precisely to prevent them. WCAG is a set of internationally recognized standards, currently at version 2.2, that define what 'accessible' means for digital content. Meeting WCAG Level AA compliance is the benchmark most organizations aim for, and it's the standard referenced in most legal frameworks globally.

Here's the tension at the heart of this lesson: accessibility testing has historically required either expensive specializt consultants, months of manual auditing, or both. Most teams, marketing departments, HR teams building internal portals, small business owners updating their websites, educators creating digital course materials, simply don't have that budget or timeline. AI-powered accessibility tools have changed that equation significantly. They don't replace human judgment. But they've made it possible for a non-technical professional to run a credible first-pass accessibility audit on a Monday morning, without writing a single line of code.

The Problem With Manual-Only Testing

Sarah Chen is a communications manager at a mid-sized healthcare nonprofit in Toronto. In 2022, her team launched a redesigned patient resource website, 80 pages of health information, downloadable guides, and appointment booking tools. A disability advocacy group contacted them three months after launch. Fourteen percent of their target audience had some form of visual impairment. The new site, which had gone through two rounds of visual design review and a full copyediting pass, had 47 distinct WCAG failures. None of them had been caught because no one on the team had thought to test for them. Sarah's team didn't know what they didn't know.

Sarah's situation is the norm, not the exception. A 2024 WebAIM Million report, an annual study that scans the home pages of one million websites, found that 95.9% of home pages had detectable WCAG 2 failures. The average page had 56.8 distinct errors. The most common failures were low-contrast text, missing image alt text, missing form input labels, empty links, and missing document language. Every single one of those failure types can be detected automatically. Yet they persist at massive scale because most teams never run a test. The problem isn't ignorance of WCAG, it's that testing felt like a specializt task.

When Sarah's team finally did commission an accessibility audit, it cost $8,400 and took six weeks. The consultant used a combination of automated scanning tools and manual screen reader testing. The automated portion, which flagged the structural, code-level issues, took about four hours. The manual portion, where a human tester actually navigated the site using assistive technology and judgment, took the remaining time. This ratio matters enormously for understanding where AI tools fit. Automated tools, including modern AI-enhanced ones, can catch roughly 30-40% of WCAG issues reliably. The rest require human experience and context. But that 30-40% is the low-hanging fruit, and it's exactly where most teams currently have zero coverage.

Click to enlarge

A Different Industry, The Same Gap: Education

Marcus Webb is an instructional designer at a large community college in Atlanta. His team produces about 120 online course modules per year, lecture slides converted to PDFs, recorded video lessons, downloadable reading packs, and interactive quizzes built in Canvas. In 2023, the college received a formal complaint from a student with dyslexia and low vision who found the course materials nearly unusable. The fonts were too small, the PDFs had no reading order structure, the videos had no captions, and the color contrast on quiz questions failed minimum standards. Marcus's team was producing inaccessible content at industrial scale, completely unintentionally.

Marcus started using Microsoft Copilot (integrated into the Microsoft 365 tools his college already licensed) alongside a dedicated tool called Grackle for Google Workspace to check accessibility in his documents before publishing. Within three weeks, he'd built a simple pre-publication checklist that caught the most common failures in his team's workflow: missing heading structure in Word documents, missing alt text in PowerPoint slides, insufficient color contrast in designed graphics, and PDFs exported without accessibility tags. His team reduced their detectable WCAG failures by roughly 60% before a single student saw the content. He didn't learn to code. He learned to run a tool before hitting publish.

Comparing AI Accessibility Tools: What They Actually Do

Click to enlarge

What a Sales Director Learned About Accessible Proposals

Priya Nair is a senior sales director at a B2B software company in London. Her team sends roughly 40 client proposals and pitch decks per month, all built in PowerPoint and exported as PDFs. In late 2023, a procurement manager at a large NHS trust told Priya's team that their proposal had been deprioritized in the evaluation process, not because of price or features, but because the PDF wasn't accessible to a member of the evaluation committee who used assistive technology. The trust had a policy of requiring accessible documents from suppliers. Priya's team hadn't known that policy existed. They also didn't know their PDFs were inaccessible.

Priya introduced two changes. First, every proposal now runs through the Microsoft Accessibility Checker before export, a step that takes about 90 seconds and flags issues the sender can fix immediately. Second, she started using ChatGPT to generate descriptive alt text for the charts and diagrams her team included in proposals, because those visuals were consistently the hardest part to describe well. She'd paste a description of the chart, 'a bar chart showing our customer retention rates by industry, with SaaS at 94%, healthcare at 89%, and retail at 76%', and ask ChatGPT to write professional alt text. The result went directly into the PowerPoint alt text field. No technical skill required. Procurement barriers, reduced.

Why AI Makes This Accessible to Non-Technical Professionals

The shift that AI tools have enabled isn't primarily about detection, manual checkers have existed for years. The shift is about interpretation and action. When the WAVE tool tells you 'Empty button, 3 instances,' a non-technical professional used to have to either ignore it or escalate it to a developer who would spend time investigating. Now, you can paste that error message directly into ChatGPT and ask for a plain-English explanation, the impact on real users, and a practical fix you can either do yourself or hand to a colleague with clear instructions. The gap between 'I found a problem' and 'I fixed a problem' has shrunk dramatically.

There's also a content generation dimension that's genuinely new. Writing good alt text is harder than it sounds. It requires you to think about what information the image conveys, not just what it shows, and to write it in a way that's useful to someone who can't see it at all. The same applies to writing descriptive link text (replacing 'click here' with something meaningful), creating captions for videos, or structuring a document with proper headings. These are writing tasks. AI writing tools. ChatGPT, Claude, Microsoft Copilot, are exceptionally good at them when given the right inputs. A marketer who would have spent 20 minutes struggling to write alt text for a complex infographic can now produce something better in 3 minutes.

The honest caveat, and this matters, is that AI tools cannot replace a human accessibility audit. They catch structural and detectable failures well. They cannot reliably assess whether a page is genuinely navigable by keyboard alone, whether a complex interactive form works with a real screen reader, or whether the cognitive load of a document is appropriate for users with cognitive disabilities. The WebAIM research consistently shows that automated tools catch somewhere between 25-40% of all real accessibility barriers. That means 60-75% still requires human evaluation, ideally including people with disabilities. AI tools are your starting point, not your finish line, but for most teams, they haven't had a starting point at all.

Key Lessons From Part 1

1. WCAG compliance isn't optional for most organizations, it's referenced in legal frameworks across the US, UK, EU, and Canada, and affects any digital content you publish externally or internally.
2. The majority of websites and documents have detectable accessibility failures, the 2024 WebAIM Million study found errors on 95.9% of home pages, with an average of 56.8 errors per page.
3. Automated AI tools reliably catch 30-40% of WCAG issues, specifically the structural, detectable failures like missing alt text, low contrast, empty links, and missing form labels.
4. The remaining 60-70% of accessibility barriers require human evaluation, ideally involving people with disabilities. AI is your starting point, not a complete solution.
5. Non-technical professionals already have powerful accessibility tools available to them at no extra cost, the Microsoft Accessibility Checker is built into every Microsoft 365 subscription.
6. AI chatbots like ChatGPT and Claude serve a critical interpretation role, they translate technical error reports into plain-English explanations and actionable fixes any professional can implement.
7. Real professional workflows across industries, healthcare communications, higher education, B2B sales, are already being affected by accessibility gaps, with legal, reputational, and commercial consequences.
8. Writing accessibility content (alt text, descriptive links, video captions) is a task AI writing tools handle well when given adequate context about the content being described.

When Compliance Becomes a Crisis: Lessons from Real Teams

In 2022, Domino's Pizza lost a landmark accessibility lawsuit brought by a blind customer who couldn't use their website or mobile app with a screen reader. The case went all the way to the Supreme Court, which declined to hear Domino's appeal, leaving the lower court ruling intact. What made this painful wasn't just the legal cost. It was the timeline. Domino's had years of warning. Accessibility advocates had flagged the issues publicly. Internal teams had likely seen automated scan reports showing problems. The gap wasn't awareness. It was action. That gap, between knowing something is broken and fixing it, is exactly where most organizations live today.

The Domino's case became a catalyst across the restaurant and retail industries. Marketing teams, HR portals, e-commerce managers, and corporate communications directors suddenly found themselves in emergency meetings about WCAG compliance. Many had never heard the acronym before. The common thread in those meetings: organizations had accessibility scan reports sitting in shared drives, generated months earlier by automated tools, never acted upon. The tools had done their job. The humans hadn't connected the output to real decisions. AI-powered testing changes the stakes here, not by replacing human judgment, but by making the reports impossible to ignore and the remediation path much clearer.

The principle the Domino's story reveals is this: accessibility testing is only valuable when it's embedded in the workflow of people who make decisions, not buried in a developer's ticket queue. When a marketing manager can run an accessibility check on a landing page before it goes live, when an HR director can audit a benefits portal before open enrollment, when a teacher can check a course module before students see it, that's when compliance stops being a legal checkbox and starts being a professional habit. AI tools are finally making that possible for non-technical professionals.

How AI Reads a Page the Way a Screen Reader Does

Here's what most people don't realize about automated accessibility testing. When a tool like Deque's Axe, Microsoft's Accessibility Insights, or WAVE scans a web page, it isn't just checking a visual checklist. It's parsing the underlying structure of the page, the code that tells assistive technologies like screen readers what everything is and how it relates. An image without alt text looks fine on screen. A screen reader hits it and says nothing. A form field without a label looks labeled to a sighted user because the designer put text nearby, but a screen reader has no idea what the field is for. AI-powered tools have become dramatically better at catching these invisible gaps.

The shift worth understanding is how AI has improved the quality of automated detection. Traditional scanners flagged what they could prove was broken, missing alt text, wrong color contrast ratios, absent form labels. Those are called deterministic errors. But roughly 30 to 40 percent of WCAG failures are what researchers call subjective or contextual, things a tool can flag as a warning but can't definitively call a violation without human judgment. Does this alt text actually describe the image meaningfully? Is this error message genuinely helpful to someone using a screen reader? Is this reading order logical? AI models trained on large datasets of real accessibility audits are now much better at making educated assessments on these gray-area issues.

Think of it this way: the old generation of tools was like a spell-checker that only caught misspellings. The new AI-assisted tools are closer to a writing coach, they catch the obvious errors but also flag sentences that are technically correct but confusing. For a marketing manager reviewing a campaign microsite, or an HR professional building an employee self-service portal, this distinction matters enormously. You're no longer just getting a list of broken things. You're getting a prioritized interpretation of what's most likely to prevent a real user from completing a real task.

A School District That Caught Problems Before Students Did

In 2023, a mid-sized school district in the Pacific Northwest began rolling out a new learning management system for its 18,000 students. The curriculum director, not a technologist, but a former classroom teacher, had read about accessibility requirements under Section 508 and was concerned. She asked her team to run Microsoft's Accessibility Checker on every course module before student-facing deployment. What they found was startling: across 140 modules built by 30 different teachers, 94 percent had at least one accessibility issue. The most common problem? PDF documents embedded in course pages with no text layer, meaning any student using a screen reader or text-to-speech software hit a blank wall.

The curriculum director didn't panic. She used the Accessibility Checker's built-in guidance to create a two-page remediation guide for teachers, written in plain language, with screenshots. No coding required. Teachers learned to run the checker themselves before submitting modules, use Word's built-in alt text generator for images, and export PDFs with the 'accessible PDF' setting enabled. Within one semester, the district's module accessibility error rate dropped from 94 percent to 23 percent. That remaining 23 percent, mostly complex data tables and scanned historical documents, was flagged for specializt review. The curriculum director had turned an AI tool into a peer review process that ran before any student encountered a barrier.

Comparing AI Accessibility Tools: What Each One Actually Does

How a Recruiter Used AI to Fix a Hiring Process That Was Quietly Excluding Candidates

A talent acquisition manager at a financial services firm noticed something in her applicant data. Drop-off rates on the online application form were unusually high, around 67 percent, at the step where candidates uploaded their resume and filled in work history fields. She assumed the form was too long. A colleague suggested running it through WAVE. The results showed 11 missing form labels, two fields with placeholder text that disappeared when candidates clicked into them (a known accessibility failure for screen reader users), and a file upload button with no accessible name. The form wasn't just annoying. For any candidate using assistive technology, it was functionally broken.

The recruiter took the WAVE report to her HR technology vendor, who patched the issues within two weeks. Drop-off at that step fell to 41 percent, still high, but a 26-point improvement with no changes to the form's content or length. She later estimated the firm had been losing a measurable percentage of qualified candidates with disabilities for over two years, people who had given up on a form that was invisible to the hiring team but impassable for them. The WAVE tool cost nothing. The fix cost two weeks of vendor time. The accessibility barrier had been costing the company far more.

Click to enlarge

Turning Scan Results into Actions Your Team Can Actually Take

The most common failure point after an accessibility scan isn't the scan itself, it's the handoff. A non-technical professional gets a report listing 47 issues with codes like 'WCAG 2.1 AA 1.1.1' and has no idea what to do with it. The smart move is to use an AI assistant. ChatGPT, Claude, or Microsoft Copilot, to translate the technical report into a prioritized action list written for your specific team. Paste the scan output, describe who will be doing the fixing (a vendor, an internal web team, a content manager), and ask for a plain-language remediation brief. You'll get something your team can actually act on in a Monday morning meeting.

Prioritization is where most teams get stuck. A scan might return 60 issues, but not all of them carry equal weight. WCAG violations are tiered as Level A (must fix, basic access), Level AA (should fix, legal standard in most jurisdictions), and Level AAA (best practice, often aspirational). AI tools can help you sort your results by these tiers, then further by user impact: which failures block task completion entirely versus which ones create friction. For a sales team auditing a client-facing proposal portal, a broken checkout button is a Level A catastrophe. A slightly off color contrast on a footer link is Level AA friction, important, but not the first call you make.

There's also a documentation angle that many professionals overlook. If your organization is subject to ADA, Section 508, or the European Accessibility Act, you need records of your accessibility testing, not just the fixes. AI tools can help you generate a Voluntary Product Accessibility Template (VPAT) summary, draft an accessibility statement for your website, or create a testing log that shows dates, tools used, issues found, and remediation status. These documents matter enormously if you ever face a complaint or audit. Building the habit of documenting your scans, even informally, is a form of organizational protection that costs almost nothing when you have AI helping you draft the language.

Click to enlarge

Key Lessons from the Examples in This Section

1. Accessibility failures often hide in plain sight, the Domino's case and the HR enrollment portal both show that scan reports existed but weren't connected to decision-makers who could act on them.
2. You don't need to be technical to run a meaningful accessibility audit. WAVE, Microsoft Accessibility Checker, and Canva's built-in tool are designed for professionals without coding knowledge.
3. AI writing assistants transform accessibility scan outputs from technical jargon into plain-language briefs that any team member can understand and act on.
4. The school district example shows that building accessibility checking into existing workflows, as a step before submission, not after, dramatically reduces error rates without requiring specializt knowledge.
5. Drop-off rates, bounce rates, and form abandonment data can signal accessibility failures that aren't visible to sighted users, the recruiter's story is a reminder to look at user behavior data alongside scan results.
6. Prioritize by WCAG level (A before AA before AAA) and by user impact, does this failure block task completion, or does it create friction? Focus fixes accordingly.
7. Documentation matters legally and operationally. AI tools can help you draft accessibility statements, testing logs, and VPAT summaries that protect your organization and create accountability.

What to Carry Forward

• Accessibility testing is now a professional responsibility for anyone who publishes digital content, not just developers.
• The best free tools (WAVE, Microsoft Accessibility Checker, Accessibility Insights) catch the majority of high-impact WCAG violations with no technical skill required.
• The value of AI in this space isn't just detection, it's translation. Turning technical scan output into human action is where AI assistants create the most immediate value for non-technical teams.
• WCAG 2.2 is the current standard. If your organization's compliance documentation references only WCAG 2.0 or 2.1, it needs updating.
• The Tab Key Test is a 30-second manual check that gives you immediate intuition for keyboard navigation failures, use it before and after any digital launch.
• Embedding accessibility checks into existing workflows (before publication, before enrollment opens, before a job posting goes live) costs far less than fixing problems after users report them.

From Audit to Action: Making AI Accessibility Testing Stick

In 2022, Domino's Pizza lost a landmark accessibility lawsuit when a blind customer couldn't order food through their app using a screen reader. The company had run automated tests. They had compliance checklists. But no one had connected the test results to actual workflow changes, the findings sat in reports while the product shipped broken. The gap wasn't technical. It was organizational. Domino's had tools producing data, but no process for turning that data into accountability. The case cost them in court and in reputation, and it became a reference point in accessibility law that legal teams still cite today.

The lesson extracted from Domino's isn't "run more tests." It's that accessibility testing only creates value when someone owns the output and acts on it. AI tools can scan pages, flag contrast failures, and generate remediation suggestions in seconds. But those suggestions need to land on someone's desk, get prioritized, and get fixed before the next release. The principle here is ownership over automation: the tool finds the problem, a human decides what happens next. That decision loop, test, assign, fix, retest, is where most organizations break down, and where AI can help most if used deliberately.

Turning AI Findings Into Team Habits

The BBC's accessibility team faced a version of this challenge when they scaled their digital products across hundreds of pages updated daily. Automated scanners were generating hundreds of alerts per week. The volume was paralyzing, developers didn't know where to start, and the backlog grew faster than it shrank. Their solution wasn't more tools. They used AI to categorize and prioritize alerts by severity and user impact, then routed specific issue types to specific team owners. A contrast failure on a news article went to the content team. A missing ARIA label on a navigation component went to the front-end developer. The alerts stopped being noise and started being assignments.

This triage approach, using AI to sort, not just scan, is something non-technical professionals can replicate even without enterprise software. If you receive an accessibility report from a tool like Axe, WAVE, or Microsoft Accessibility Insights, you can paste its findings into ChatGPT or Claude and ask it to organize the issues by priority, explain each one in plain language, and suggest who on your team should own the fix. You don't need to understand the underlying code. You need to understand the business impact of each gap and who has the authority to close it.

The BBC's approach also revealed something important about sustainability: accessibility compliance isn't a project with an end date, it's an ongoing practice. AI tools make continuous monitoring feasible for teams that previously could only afford quarterly audits. Weekly automated scans, combined with AI-generated summaries sent to team leads, shift accessibility from a reactive scramble before launch to a steady background process, the same way a finance team monitors budget variance rather than discovering overspending at year-end.

A Hospital System That Got It Right

Cedars-Sinai Medical Center in Los Angeles began integrating AI accessibility checks into their patient portal development cycle after receiving complaints from patients using assistive technology to book appointments. Their communications team, not their IT department, took ownership of the content-layer fixes: rewriting image alt text, adding captions to instructional videos, and restructuring PDF forms. They used AI writing tools to generate draft alt text descriptions for medical images at scale, then had clinical staff review for accuracy. The result was faster remediation of content issues without requiring developers to rewrite a single line of code.

What Cedars-Sinai demonstrated is that accessibility work is not exclusively a technical job. Content, communications, and operations teams carry a significant share of the compliance burden, and AI tools give those teams real capability they didn't have before. A communications manager who can generate accurate alt text for 200 images in an afternoon, or an HR professional who can audit their job posting PDFs for screen reader compatibility, is doing meaningful accessibility work without writing a single line of code.

The HR Manager Who Became an Accessibility Champion

At a mid-sized financial services firm in Chicago, an HR manager named Priya had no accessibility training and no design background. But after a new hire mentioned they couldn't access the onboarding documents using their screen reader, Priya decided to fix it herself. She uploaded each PDF to Adobe Acrobat's accessibility checker, pasted the error reports into Claude, and asked for plain-language explanations of each issue and how to fix it without technical help. Within two weeks, she had remediated the entire onboarding document library, adding reading order tags, replacing scanned images of text with actual typed text, and adding document titles.

Priya's story isn't exceptional, it's repeatable. She didn't need a developer. She needed a clear problem, a willingness to act, and AI tools that could translate technical output into actionable instructions. The organizations that make the most progress on accessibility aren't necessarily the ones with the biggest budgets or the most sophisticated tools. They're the ones where someone like Priya feels empowered to start, rather than waiting for IT to schedule a project.

Building a Sustainable Practice, Not a One-Time Audit

The organizations that sustain accessibility progress share one habit: they've made it part of their existing review cycles rather than treating it as a separate initiative. A marketing team that adds an AI accessibility check to their content approval checklist will catch more issues than a team that schedules an annual audit. A product manager who includes accessibility scan results in their weekly status report creates visibility that a quarterly compliance review never achieves. The frequency matters more than the sophistication of the tool.

AI tools make this frequency affordable. Running a WAVE scan takes two minutes. Pasting results into Claude for a prioritized action list takes five. Sharing that list in a Slack channel or email takes one more. A ten-minute weekly accessibility check, embedded in an existing workflow, compounds into meaningful compliance progress over a quarter. Teams that try to solve accessibility in a single sprint usually find themselves back at square one six months later when new content has been published and old issues have returned.

The final shift in mindset is from compliance as a destination to accessibility as a quality standard, the same way a professional wouldn't publish a report with spelling errors, they wouldn't publish a page with missing alt text. AI tools lower the effort required to meet that standard so much that the barrier is no longer skill or time. It's simply deciding that accessibility belongs in your definition of done. Once that decision is made, the tools are ready and waiting.

Click to enlarge

Key Takeaways

1. AI accessibility tools find problems fast, but human ownership determines whether those problems actually get fixed, the tool is only as useful as the process it feeds into.
2. WCAG 2.2 AA compliance is the legal standard in most countries; AI tools can automatically check many criteria, but human judgment is still required for comprehension and context.
3. Non-technical professionals, content editors, HR managers, marketers, operations staff, own a significant share of accessibility fixes and can act on them using AI tools without writing code.
4. Pasting accessibility scan results into ChatGPT or Claude and asking for plain-language prioritization is a practical, free technique available to anyone right now.
5. Frequency beats intensity: a ten-minute weekly AI accessibility check embedded in existing workflows produces more sustainable progress than annual compliance audits.
6. The Domino's case and others demonstrate that organizations face real legal and reputational risk when accessibility testing produces reports that don't translate into action.
7. AI tools like WAVE, Microsoft Accessibility Checker, Adobe Acrobat AI, and Grammarly AI cover different layers of accessibility, using more than one catches more issues.
8. Accessibility is a quality standard, not a compliance checkbox. Teams that internalize this ship better products for everyone, including users without disabilities who benefit from cleaner, clearer content.