GUIDE · AIunipedia Founding Lab
An AI feature can produce a useful answer and still be difficult to use. Evaluate the whole path: entering a task, understanding a response, checking evidence, recovering from an error and requesting help.
Begin with observable tasks
Can a keyboard user reach the main actions in a sensible order? Are focus indicators visible? Are headings meaningful? Can a person enlarge the text without losing controls? Does an error explain how to recover without relying only on colour?
These questions complement the AI tool scorecard. They are not a claim that a short checklist establishes conformance with an accessibility standard.
Test AI-specific friction
A long streaming response may interrupt reading or hide the useful next step. An unexplained confidence score may confuse people. A polished answer with no supporting evidence can make checking harder. Offer readable structure, explicit uncertainty and a route to a human where the task needs one.
Learn from users responsibly
Ask people about the task they want to complete and the tools they use. Do not assume a diagnosis determines a person’s needs or request sensitive health information in a public thread. Obtain consent for research and recordings, and report the limits of a small sample.
Discussion: What single change would make your current AI workflow easier to use? Describe the interaction rather than another person’s private circumstances. The project case-study template can help record a concrete before-and-after result.
Explore next
Related learning guide · Continue learning
Browse: AI Tools & Workflows · Learning directory
Written by Michael Oracle. AIunipedia Community is owned and run by Dr Fun and Michael Oracle.