Five questions to ask about every AI deployment
Robin Christopherson | 08 Sep 2026For more than 25 years, accessibility professionals have been helping organisations fix digital barriers that could have been avoided if disabled people had been considered earlier.
Missing labels, inaccessible forms, poor keyboard support and unusable documents have all carried a cost. Organisations have paid for remediation, handled complaints, lost customers and asked employees to rely on cumbersome workarounds. Disabled people have carried the greatest cost through lost time, restricted choices and services they simply could not use. 
As organisations deploy AI at speed, I worry that we are creating a new and deeper form of accessibility debt - and this time, the barriers may be harder to see. They may sit inside training data, third-party models, automated workflows, procurement decisions and systems that classify, predict, rank or recommend. By the time the impact becomes clear, the AI may already be embedded across customer journeys and workplace processes.
Accessibility debt has moved beneath the interface
Most traditional accessibility barriers can be observed in the product; a blind customer using a screen reader encounters an unlabelled button. A keyboard user cannot reach a control. Someone with low vision finds that text lacks sufficient contrast. A document has no headings or meaningful reading order.
The causes can still be complicated - particularly when accessibility has been neglected across a large digital estate - but there is usually something specific to test and fix.
The introduction of AI potentially changes where exclusion can happen; a customer service bot may misunderstand someone with atypical speech. A recruitment tool may interpret disability-related characteristics as signs of poor performance. An AI assistant may produce information in a format that is difficult to navigate with a screen reader. A fraud detection system may treat an unusual pattern of interaction as suspicious without recognising that the customer is using assistive technology or receiving support.
The customer experiences the outcome, but may never know what caused it. The organisation may not know either.
This is where accessibility debt begins to become inclusion debt. Decisions made during procurement, configuration and deployment can quietly accumulate risk across millions of interactions.
Your next customer could be excluded by AI
This is an immediate delivery issue for any large organisation introducing AI into customer or employee journeys.
Consider a bank launching an AI-powered assistant inside its mobile app. The assistant might help customers understand transactions, find information or complete routine tasks. Used well, it could make the service significantly easier for people who struggle with complex navigation, dense content or multi-stage forms. 
Several problems could emerge once the assistant is live. Its voice input may work poorly for some disabled customers. Generated responses may be difficult to follow with a screen reader. The assistant may give an answer without explaining how to reach the underlying source. A customer may be unable to escape the automated journey and speak to a person when the system misunderstands them.
The bank then has a choice. It can treat each complaint as an isolated usability issue, or recognise that the AI integration needs a wider accessibility review.
The same challenge applies to insurers using AI in claims, retailers adding conversational search, technology companies building generative interfaces and employers deploying AI tools across recruitment and workplace productivity.
In each case, accessibility can be lost between existing areas of responsibility. The accessibility team may review the visible interface while the AI team focuses on model performance. Procurement may assess the supplier's broad compliance claims without examining the actual customer journeys. Risk teams may consider bias and data protection without looking closely at disability.
No single team sees the whole problem, yet the organisation remains responsible for the outcome.
Why fixing AI later may be much harder
Retrofitting accessibility into a website can require extensive work, particularly when the underlying design system is poor. However, AI potentially introduces additional layers.
The problem may come from a model supplied by another company. It may involve historical data, internal policies or the way a service has been configured. The same system may behave differently from one interaction to the next. Updating the interface may leave the underlying exclusion untouched.
A fix could therefore require changes to vendor relationships, data practices, escalation routes, governance and testing methods. It may also require an organisation to reconsider decisions that have already been built into important operational processes.
This is why senior decision-makers should care about AI accessibility before deployment. Once a system is widely used, remediation has to compete with commercial commitments, supplier contracts and pressure to avoid disrupting established services.
The relevant question for leaders is whether they can currently see where AI is affecting disabled customers and employees, how that impact is being measured and who is accountable when something goes wrong.
Many organisations cannot yet answer those questions with confidence.
Disability is an early warning system
As a blind person, I experience the benefits and shortcomings of AI very directly. AI tools can describe images, summarise inaccessible information and reduce the effort involved in navigating complex digital environments. They can give me greater independence and help me complete tasks that would previously have required sighted assistance.
The experience can also be unpredictable. An AI feature may be technically present but difficult to operate with a screen reader. It may generate a visually polished response with little useful structure. It may simplify one part of a task while directing me into an inaccessible process for the final step. 
These experiences matter beyond my own access needs. Disabled people often encounter the weaknesses in a system earlier because we interact with it in ways that were poorly represented during design and testing. A failure to support screen readers may expose weak structure. Difficulty understanding atypical speech may reveal narrow training or test data. A confusing automated journey may show that users have too little control or too few ways to recover from errors.
These are important signals about product quality, resilience and trust.
Research by Open Inclusion has explored how AI can create technical and social debt when disability is overlooked. The Center for Democracy and Technology has also examined the role of disability data and the harm that automated systems can cause in areas such as employment and access to essential support.
The broad lesson is clear. Disabled people need to be involved in deciding what is built, how it is tested and how success is measured. Their involvement should start before an AI feature reaches customers or colleagues. This might mean including disabled people in discovery workshops to help identify user needs and risks, or involving them in pilot programmes and usability testing while systems are still being developed. Organisations can also work with employee disability networks, accessibility specialists and disabled customers to review proposed features and challenge assumptions before they become embedded in products and processes.
Five questions to ask about every AI deployment
Organisations need a practical way to identify where accessibility debt may already be accumulating. These five questions provide a useful starting point:
- Which customer and employee journeys now depend on AI? Map the points where AI generates content, interprets speech, recommends an action, makes a classification or controls access to human support.
- Have those journeys been tested with disabled people? Standard compliance testing remains important, but the variable behaviour of AI means that inclusive user research is essential.
- What happens when the AI is wrong or inaccessible? Users need clear ways to understand, challenge and correct an outcome, with a reliable route to human assistance where the impact is significant.
- Do procurement and governance cover accessibility explicitly? General assurances from suppliers are insufficient. Requirements should address the AI feature, its outputs, the journeys around it and the way its performance will be monitored.
- Who owns the resulting risk? Accessibility, product, AI, customer experience, compliance and procurement teams need an agreed process for identifying problems, setting priorities and tracking improvements.
These questions will not resolve every issue, but they can expose gaps before they become difficult and expensive to untangle.
Financial services should pay particular attention
Banks, insurers and pension providers have strong reasons to explore AI. It can help explain complex information, support colleagues, identify patterns and make some customer journeys easier to complete.
These services also involve high-stakes decisions. A poor experience may affect someone's ability to manage money, make a claim, understand a product or receive support at a difficult moment.
In this context, an inaccessible AI journey carries more than a usability cost. It can affect customer trust, regulatory confidence and an organisation's ability to demonstrate that it understands the needs of disabled and vulnerable customers.
AI governance in financial services therefore needs to connect with accessibility, customer experience and existing approaches to Consumer Duty. Teams should be able to show how disabled customers were involved, which journeys were tested, what forms of human fallback are available and how emerging problems will be identified. That evidence will also help organisations make better decisions about where AI adds genuine value and where automation creates unacceptable risk.
What good AI accessibility looks like
A mature approach begins with visibility. Organisations need to know where AI is already being used across their products, services and internal tools. They then need to identify the journeys where a failure could cause the greatest exclusion or harm. 
Testing should combine expert accessibility review with disabled user research. This matters because AI features can behave differently according to the user's language, communication style, access technology and previous interactions. An automated scan cannot reveal the full experience.
Findings also need to reach the people who can act on them. A list of defects will have limited value if procurement, product, engineering and governance teams continue working in isolation. Organisations need clear ownership, useful measures and a process for responding when the AI behaves in unexpected ways.
This work also creates opportunities. Testing with disabled people can reveal where AI removes friction, provides greater choice or makes information easier to understand. Those insights can shape a stronger product rather than serving only as a record of problems.
How AbilityNet can reduce the risk before it becomes embedded
At AbilityNet, we are already helping organisations examine how AI is changing the accessibility of customer and employee experiences.
Our work can focus on a specific AI component, such as a chatbot, voice assistant, generative interface or agentic journey. It can also look more broadly at the organisation's strategy, procurement, governance and ability to measure AI's effect on accessibility.
Our consultants combine expert accessibility assessment with inclusive research involving disabled people. This helps organisations understand both technical conformance and the real experience of using an AI-enabled service.
We can review high-risk journeys, identify where existing testing is no longer sufficient, help teams prioritise issues and build accessibility into AI requirements before contracts and designs become difficult to change.
For organisations with established accessibility programmes, we can help connect that work to AI governance and delivery. For those at an earlier stage, we can help identify the most important risks and define a practical starting point.
Explore AbilityNet's accessible AI integration and consultancy services or speak to our experts about an AI feature, customer journey or deployment that you want to assess.
Talk to us about accessible AI integrations
The debt is still avoidable
AI can remove barriers that have persisted for decades. As a blind person, I am genuinely excited by that potential. As an accessibility professional, I also know how easily exclusion becomes embedded when speed, novelty and convenience dominate early decisions.
Organisations still have an opportunity to shape how this technology is introduced. The most valuable step they can take now is to identify where AI is affecting disabled people, test those experiences properly and give someone clear responsibility for improving them.
Once inaccessible systems become trusted, scaled and woven into everyday operations, the cost of change rises quickly. The accessibility debt of the web era took decades to address. We should use what we learned from that experience while there is still time to prevent an even larger debt from building inside AI.
More from Robin Christopherson on AI and accessibility
Accessible AI: Who is responsible when AI acts on our behalf?
- Published 2 Sep Aug 2026. As AI takes on more tasks, accessibility becomes more important than ever.
Why accessibility is one of the most important AI procurement decisions you can make
- Published 24 Aug 2026. Inaccessible AI can create costly remediation, governance risks and poor customer experiences.
The hidden cost of AI bias: Why disability may be the ultimate test of trustworthy AI
- Published 17 Aug 2026. As AI adoption grows, disability is emerging as a critical test of trustworthy AI.
The EU AI Act has a message for accessibility teams: get involved earlier
- Published 10 Aug 2026. The EU AI Act's requirement for human oversight highlights the need to involve accessibility experts and disabled users much earlier in AI development.
Using AI tools to drive an accessible dev process
- Published 05 Aug 2026. How do we move beyond AI-generated code and start using AI to reduce accessibility debt before it reaches production?
AI could make financial services more inclusive. But only if accessibility is part of the infrastructure
- Published 27 July 2026. Why accessibility must be treated as core infrastructure if AI is to improve financial inclusion.
The trust gap: why AI's edge cases matter to customer loyalty
- Published 30 June 2026. Why accessibility and edge cases are critical to building trust in AI-powered services.
The delicious irony at the heart of trusting AI
- Published 17 June 2026. A personal reflection on trust, accessibility and the unexpected ways AI can both help and hinder.
Executive briefings for senior leaders
Robin offers invite-only executive briefings exploring AI, accessibility, governance, customer trust, regulation and digital inclusion.
