New AI Tool Checks Whether Apps Work for People With Disabilities

Building an app that works for everyone — including people with low vision, shaky hands, or dyslexia — is slow, expert-level work. Designers call it inclusive design. Checking a product against all those different needs can take weeks, so many companies decide it costs too much and skip it. The result: whole groups of people get shut out of banking apps, job portals, and government websites. That's the trade-off a team from Oregon State and other universities set out to shrink.

Their answer is a tool called Apply-<x>Mag. You feed it a product plus a description of who you're designing for, and the AI — a large language model, the same kind of technology behind ChatGPT — scans the product and flags places where specific groups would struggle. The researchers made it flexible: it can run almost any inclusive design method, as long as that method can be written down as simple rules about user traits. They call those rules "Mags."

The team tested it with real researcher and practitioner teams on seven different products, using different combinations of methods. The tool's flags were right 90 to 99 percent of the time, and it caught 82 to 89 percent of the problems that were actually there. Running it used roughly the same energy as two to four ordinary Google searches — a surprisingly small footprint for an AI tool.

Here's the catch. This is a research paper, not a product you can buy or install, and it's not perfect: it missed up to one in five real issues, so a human designer still has to double-check. Any AI judging who gets left out is also only as good as the rules it's given. Still, if the approach holds up, it could make inclusive design cheap enough that companies stop treating it as a luxury — and more apps could simply work for more people.