Your organisation has rolled out Microsoft Copilot. Your team is using ChatGPT. Someone in finance built an AI-assisted reporting workflow last quarter. By all appearances, AI adoption is happening.
But here is the uncomfortable question every leader needs to ask: do you actually know what your people understand about the AI they are using?
The data says most organisations are flying blind. Skillsoft’s Workforce Readiness Report: AI Edition, released in June 2026, found that 86% of employees now use AI tools at work, yet only 24% feel fully equipped to use them effectively. More troubling: 77% of leaders believe their organisations have set employees up for success – creating a 53-point perception gap between leadership confidence and frontline reality.
That gap is not just a training problem. It is a strategic risk.
The Confidence Crisis Behind the Adoption Numbers
Adoption metrics are seductive. They tell you tools are being used. They do not tell you whether people understand what those tools are doing, where they fail, or how to critically evaluate their outputs.
ManpowerGroup’s 2026 Global Talent Barometer found that regular AI usage jumped 13 percentage points to 45% of the global workforce – yet confidence in using technology fell sharply by 18%. It marked the first decline in overall worker confidence in three years.
DataCamp’s 2026 State of Data and AI Literacy Report adds further weight: 72% of enterprise leaders now expect AI literacy as a baseline workplace skill, yet nearly 60% still report a meaningful skills gap – even though 82% already offer AI training.
Training is not the same as literacy. Attendance is not the same as capability.
Why Role-Differentiated Assessment Is Essential
A common and costly mistake is treating AI literacy as a single, uniform capability. A cyber security analyst, a business development manager, a CFO, and a service desk technician do not need the same AI knowledge – and assessing them against the same benchmark will mislead you every time.
The U.S. Department of Labor formalised this thinking in February 2026 with its Artificial Intelligence Literacy Framework, identifying five foundational content areas: understanding AI principles, exploring AI uses, directing AI effectively, evaluating AI outputs, and using AI responsibly. The framework is explicitly designed to scale to role, responsibility, and risk exposure – not applied as a universal standard.
This is where practical enterprise assessment must begin.
A Four-Layer Assessment Framework
Rather than a single survey or e-learning module, structure your assessment around four layers that reflect how AI capability actually shows up at work.
Layer 1 – Foundational Awareness (All Roles)
Can people describe what AI is and is not? Do they understand that AI outputs are probabilistic, not definitive? This is the baseline – and you may be surprised how many cannot answer confidently.
Layer 2 – Applied Fluency (Operational and Specialist Roles)
Can people prompt AI tools effectively and identify when an output is wrong or biased? Study.com’s 2026 State of AI Jobs and Skills Report found that task decomposition – breaking real work problems into AI-manageable steps – was a skill only 34% of employees felt confident in. This layer separates casual users from genuine practitioners.
Layer 3 – Governance Awareness (Managers, Compliance, Legal)
Does this group understand the data privacy implications of what they feed into AI tools? Only 12% of leaders in Skillsoft’s research reported having comprehensive AI governance in place – meaning the people meant to enforce responsible use often lack the fluency to do so.
Layer 4 – Strategic AI Literacy (CIOs, CTOs, Executives)
Can your executive team critically evaluate AI investments? Do they understand what agentic AI means for enterprise workflows and where governance obligations apply? This is not about becoming a data scientist – it is about asking better questions of the people building your AI strategy.
Running the Assessment in Practice
The output should be a literacy heat map across your teams – revealing where confidence is genuine, where it is assumed, and where people are quietly working things out as they go.
A few principles to make it effective:
- Segment by role first. Group roles by their relationship to AI: builders, operators, consumers, and decision-makers. Each group needs different questions.
- Pair self-assessment with scenario testing. Self-reported confidence is consistently inflated. Scenario-based prompts reveal actual reasoning far more reliably.
- Anonymise to surface honesty. People will not admit uncertainty if they fear it affects performance reviews. Psychological safety is a prerequisite for accurate data.
- Assess managers separately. A team lead’s AI literacy shapes their team’s norms – uncertainty at that level flows downstream.
- Repeat every six months. The landscape shifts too fast for annual snapshots to stay useful.
From Measurement to Action
The point is not to expose gaps for their own sake. It is to design interventions that meet people where they are – targeted, role-relevant, and measurable – rather than broadcasting generic training to a workforce with wildly different starting points.
You cannot close a gap you have not measured. Start by finding out where your people actually stand.
Ready to move from assumption to evidence? Explore how Insentra helps organisations assess and build genuine AI capability at AI Momentum.
