The short answer
A confidence survey tells you what people believe they can do. A knowledge assessment tests what they understand. Neither, by itself, proves they can apply AI responsibly in a particular task or that the organisation is ready to support them.
Start with the decision, then choose the evidence
If the question is whether employees feel able to use a new tool, ask them. A survey can reveal hesitation, perceived barriers and where people want support. Its answer is a perception, so high confidence can coexist with gaps in knowledge. Low confidence can also reflect caution rather than low capability.
If the question is whether people understand common concepts, risks and human oversight, use a consistent knowledge assessment. The Sova-delivered genAssess Core-6 measures foundational AI knowledge across the same six areas for candidates and employees. It gives a comparable baseline, but its result is not a measure of job performance or an assessment of the organisation's controls.
What each method can and cannot tell you
A self-report survey can surface confidence, access and perceived training needs. Its wording and the respondent's self-awareness affect the result. Treat it as a conversation starter rather than proof of competence.
A knowledge test can reveal understanding of AI foundations, use cases, bias and risk, data governance, output evaluation and human oversight. It does not observe somebody doing a specific job task. A practical exercise or structured interview could explore that task, but it would need to be designed and assessed separately by the employer.
Workflow evidence answers another question: does the person have the approved tools, data boundaries, escalation route and authority to act on what they know? A knowledgeable person can still be blocked by the surrounding process. Measure those conditions through interviews, process reviews and observation rather than assigning them to a person's score.
Use the signals together without collapsing them into one score
For an L&D cohort, collect a short confidence and access survey, establish a Core-6 knowledge baseline where appropriate and select one real workflow to review separately. Revisit the same questions after learning and a period of supported practice. Report changes in each measure on its own terms.
Do not describe a confidence score as proficiency or a Core-6 result as proof that someone can use AI safely in every role. The most useful report tells a leader what to do next: teach a topic, remove a workflow barrier, or gather more evidence before making a consequential decision.
Put it into practice
A simple way to start
- Name the decisionIs this for hiring, learning, rollout support or a different choice? Record who will use the result.
- Match the methodUse survey responses for perceptions, Core-6 for common knowledge and a separately designed method for a defined task.
- Report limitsShow what each measure cannot establish, and avoid turning several distinct signals into an unsupported readiness score.
Sources and further context
- Sova and genAssess webinar Assessment format and why objective evidence matters
- NIST AI Risk Management Framework Separating people, oversight and organisational risk management