AI literacy at work is the ability to use, question, and govern AI appropriately within a person's role. It is not the ability to recite model terminology or write one clever prompt.
An AI-literate employee knows when AI may help, what information can be used, how to brief the system, how to review the result, and when a human must remain accountable.
Six workplace capabilities
Understand the tool
Employees should know that generative AI produces plausible outputs rather than verified truth. They need a practical understanding of context, limitations, variability, and why confident language is not evidence.
Choose appropriate work
AI is often useful for drafting, summarizing, classifying, comparing, brainstorming, and structuring. Suitability changes when the task is consequential, highly sensitive, poorly defined, or difficult to review.
Handle data safely
People should recognize data classifications, approved environments, connected sources, retention considerations, and restricted content. Product name alone is not enough; account type and configuration matter.
Give useful instructions
Prompting is a briefing skill. A strong request clarifies objective, context, audience, constraints, source material, desired format, and quality bar. For complex work, employees should break the process into checkpoints.
Review output
Employees need habits for checking facts, calculations, sources, omissions, bias, confidentiality, and alignment with the intended decision. Review depth should increase with impact.
Act accountably
Capability does not equal permission. Employees should know which outputs they may use directly, which require review, and where to escalate uncertainty or incidents.
Literacy should vary by role
Every employee needs foundations, but managers, executives, developers, legal teams, and system owners have additional responsibilities. Managers shape expectations and reinforcement. Executives set risk appetite and investment priorities. Technical teams manage system behavior, access, and monitoring.
Measure performance, not familiarity
A self-reported confidence survey is useful but insufficient. Use realistic scenarios: choose an appropriate tool, identify prohibited data, improve a weak brief, find errors in an output, decide when to escalate, and redesign a simple workflow. Measure transfer into real work through manager observation, output quality, and repeated use.
Build literacy into the operating environment
Training works best when employees also have approved access, clear policy, relevant examples, manager support, office hours, and a path to propose use cases. Otherwise, knowledge remains separate from behavior.
AI literacy is an operating capability. The standard is not whether someone has attended a workshop. It is whether they can make a better, safer decision with AI in the context of their actual work.