The AI confidence gap: what the data shows
79% of tech workers admit to overstating their AI knowledge. 56% of self-proclaimed AI experts have received no formal training. The gap between AI confidence and AI capability is not a perception problem — it is a measurement problem.
Ask your team if they are good at using AI. Most will say yes.
Now test them.
The results will likely tell a different story.
The overstating problem
A 2025 survey of 1,200 technology decision-makers and practitioners in the US and UK by Pluralsight found that 79% of tech workers admitted to overstating their AI knowledge to employers or colleagues. The problem was most acute at the top: 91% of C-suite executives were the most likely group to overstate their capabilities. 65% of organisations had abandoned AI projects specifically due to AI skills gaps.
This is not a niche finding. These are technology professionals — the people organisations already assume are the most AI-capable. If 79% of them are overstating, the gap in non-technical teams is likely larger.
Source: Pluralsight AI Skills Report 2025
The expert problem
A September 2024 survey of 2,000 workers and leaders in the US and UK by Multiverse found that 56% of people who described themselves as AI experts had received no formal AI training. 93% of workers said they were confident they used AI ethically — while only 28% of their organisations had any AI governance practices in place.
The confidence is there. The competence — and the structures to support it — are not.
Source: Multiverse AI Maturity Gap Report, September 2024
The academic evidence
Academic research mirrors these findings. A 2025 study of 1,465 university students across Germany, the UK, and the US — published in Computers in Human Behavior: Artificial Humans — found that US students reported the highest AI self-efficacy but did not achieve the highest scores on the objective AI knowledge test. German students did. Confidence and competence diverged across all three countries.
This pattern is consistent across the literature. What people believe about their AI capabilities and what they can actually demonstrate are measuring two different things.
Hornberger, M., Bewersdorff, A., Schiff, D. S., & Nerdel, C. (2025). A multinational assessment of AI literacy among university students in Germany, the UK, and the US. *Computers in Human Behavior: Artificial Humans*, 4, 100132. DOI: 10.1016/j.chbah.2025.100132
Why this matters for organisations
The confidence gap is not a personality quirk. It is a structural problem created by how organisations currently measure AI readiness.
When you ask people to rate their own AI skills, you get their confidence. When you watch them complete a real AI task, you get their capability. These two numbers do not match — and the gap between them is where your organisation's AI risk lives.
The people most likely to publish unverified AI outputs, accept hallucinated data as fact, or make poor decisions about when to use AI are the same people most likely to rate themselves highly on a self-assessment survey.
The measurement fix
Closing the confidence gap starts with measuring the right thing. Performance-based assessment — where participants complete real workplace tasks rather than rating their own abilities — surfaces the gap rather than concealing it.
In our first cohorts, the difference between self-rated confidence and actual assessed performance has averaged more than one full point on a two-point scale per dimension. HR teams have consistently described seeing the data as the moment the conversation about AI development became concrete rather than abstract.
If you want to close your organisation's AI confidence gap, start by measuring it accurately. Book a demo at probelearning.com.