You can’t move these days for mention of Artificial Intelligence tools, and accessibility is no different! AI is increasingly positioned as a solution to accessibility challenges, with tools that offer features such as automatic alt text, automated code remediation and AI powered ‘compliance monitoring’.
In a time of increasing pressures on resources and budgets, it is understandable why this is appealing. Accessibility can seem technical, resource-intensive and complex, particularly for small to mid-sized arts organisations.
But with AI and accessibility there are two growing misconceptions that deserve careful attention.
Misconception 1: AI can fix accessibility automatically
AI tools can generate alternative text. They can scan for colour contrast failures. They can suggest improvements to code. They can even rewrite content for clarity. These are not insignificant capabilities. However, accessibility is rarely about simple pattern recognition. It is about context and meaning.
Consider an image from a theatre production, what matters most?
- The actor’s race?
- The emotional ‘tone’ of the image?
- The presence of a prop relevant to the plot?
- A visual cue that suggests a content warning?
- The aesthetic style of the production?
AI can describe what is visible in a probabilistic way. It cannot reliably interpret organisational intent. When done properly alternative text communicates purpose in context, and that requires judgement.
It’s also important to remember that AI systems are also trained on large datasets that inevitably contain bias. Descriptions can reinforce stereotypes or misidentify people. In accessibility work, where nuance and respect are essential, that risk matters.
Misconception 2: AI can continuously ‘fix’ accessibility in the background
Some vendors suggest AI can run constantly, detecting and correcting accessibility issues automatically. This framing positions accessibility as a stream of errors to be patched.
But accessibility failures are not random glitches. They are usually the result of design decisions, content workflows, coding errors and organisational priorities. If a content team has not been trained to write meaningful alt text, an AI tool generating descriptions is not solving the underlying capability gap. If a booking system is inherently inaccessible, an AI layer cannot rebuild its logic.
Where AI can be helpful
When used responsibility there are ways that AI can support accessibility work, for example:
- Drafting first-pass alt text that is then reviewed by a human
- Identifying likely contrast failures at scale
- Highlighting structural inconsistencies across large sites
- Supporting plain language rewrites
- Assisting with transcript generation for audio content
Used as a support tool, AI can increase efficiency but used as a substitute for human oversight, it becomes risky. At its core, accessibility is about whether someone can navigate independently, undertaking tasks such as booking a ticket or engaging with content (on equal terms as a non disabled user). AI doesn’t experience frustration, it doesn’t navigate with a screen reader or encounter barriers in the way that disabled users do. That’s why testing with real people remains essential. Why training teams matters. Why structural fixes are vital.
A final note for 2026 and beyond
As AI becomes further embedded into website builders, content management systems and marketing tools, we expect more products to position themselves as ‘automatically accessible’ or ‘AI-compliant’.
Before you use these types of tools we recommend asking some careful questions:
- What exactly is being automated?
- What remains my responsibility?
- How is bias mitigated?
- Has this been tested with real users?