As someone who is legally blind and uses a screen reader, I judge technology by what it helps me do independently. A useful answer can make planning easier. A confident answer based on unchecked information can create another barrier.

Chris Kerrisk’s recent post connects two issues that deserve attention: people with disabilities are among AI’s early and active adopters, and reliable accessibility information is essential to making that technology useful.

The research makes a strong case for listening to the people already using these tools. It also raises a practical question: who owns the answer when someone relies on it?

What the research shows

Randstad’s 2024 Understanding Talent Scarcity: AI & Equity report found that, among respondents:

• 55% of people with disabilities used AI for problem-solving at work, compared with 39% of people without disabilities.
• 51% used AI to write job applications, compared with 27%.
• 57% sought AI learning opportunities outside work, compared with 41%.

The research surveyed 12,429 people across 15 markets. These are people actively using AI and developing their skills, including beyond the opportunities their employer provides. Randstad report, methodology.

Understood.org’s 2026 Neurodiversity at Work survey found that 78% of neurodivergent employees reported using AI tools, compared with 59% of neurotypical employees. Separately, 66% of neurodivergent adults said AI helped make opportunities more equal for people who learn and think differently.

The Harris Poll conducted that survey among 2,073 US adults, including 614 who identified as neurodivergent. It captures reported use and attitudes in the US; it doesn’t establish that everyone gets the same benefit. Understood findings and methodology.

Disability:IN’s 2025 Creativity Over Convention research adds detail about everyday use. It describes neurodivergent participants using AI to clarify ideas and check their communication before sharing it with colleagues. Reported everyday adoption varied by generation: 85% among millennials surveyed, 57% among Generation X, 50% among baby boomers and 29% among Generation Z.

Those percentages describe a particular study group. This was qualitative research involving participants from partner companies, with 75 of 175 respondents selected for analysis to capture geographical and generational diversity. They aren’t population-wide adoption rates. Disability:IN findings and methodology.

Early adopters have something to teach us

Higher reported use doesn’t establish who adopted first. For that, we need the longer history.

Carnegie Mellon University researchers Jeffrey P. Bigham and Patrick Carrington explore it in Learning from the Front: People with Disabilities as Early Adopters of AI. Their paper discusses technologies including computer vision and speech recognition, drawing lessons about trust, control and exclusion from people with disabilities’ experiences. Read the paper.

That experience should inform how organisations develop AI services. People who have relied on imperfect technology can help explain where it works, where it fails and what users need to make their own decisions.

I’d want those voices involved in requirements, procurement and testing from the start.

An accessibility answer creates an expectation

Imagine asking a venue whether you can attend an event independently. The answer may influence whether you buy a ticket, arrange transport or ask someone to accompany you.

“The venue is accessible” gives you very little to work with.

For me, useful information includes where the entrance is, how I’ll find it and how to arrange assistance. Other people will need different details about the same visit.

Now consider a hypothetical venue where the website describes a lift, customer services knows it is awaiting repair, and an organiser has moved an activity upstairs. Unless those changes reach the assistant, it could describe a visit that is no longer possible.

The organisation needs someone responsible for keeping that information current. Users should be able to establish what was checked, when it was checked and any conditions that affect the answer.

If something is unknown, say so. Then provide an accessible contact route to someone who can resolve it.

Test the whole task

I’d test whether someone can book, arrive, enter, participate and leave. An accurate answer about the entrance won’t resolve an inaccessible booking form.

People with disabilities should help shape those tests and take part early enough to influence the service, with appropriate payment for their expertise. Include temporary changes, missing information and follow-up questions that challenge an earlier answer.

The interface needs testing too. Can I use it with a keyboard? Does a screen reader announce replies sensibly? Can I review an earlier answer without losing my place?

Finally, agree how corrections will work. If someone reports that an entrance was locked, a responsible team needs to investigate, update the information and check whether others could be affected. That includes agreeing responsibilities with any supplier operating the assistant.

The research shows substantial AI engagement among the groups studied. Organisations have an opportunity to learn from that experience and deliver services people can trust.

Before launching an accessibility assistant, I’d ask who owns its answers and how people with disabilities have helped shape them.

Both deserve a concrete answer.

Sources

• Chris Kerrisk: Original LinkedIn post, which inspired this article.
• Randstad, 2024: Understanding Talent Scarcity: AI & Equity and research methodology. The report supports the 39% comparison; the announcement contains an inconsistent 41% figure in its body.
• Understood.org and The Harris Poll, 2026: Neurodiversity at Work survey and methodology, with an accessible findings PDF.
• Disability:IN, 2025: Creativity Over Convention research framework and published findings and methodology.
• Jeffrey P. Bigham and Patrick Carrington, Carnegie Mellon University: Learning from the Front: People with Disabilities as Early Adopters of AI.