How to Verify Your Workforce’s AI Skills

This post was written by a guest contributor.

Over the past couple of years, I have had the opportunity to sit in on quite a number of AI transformation reviews. I’ve noticed that the slides are almost always the same. They show adoption rate, usage frequency, and, usually, a self-reported confidence score. Everyone in the room looks at the chart, agrees the numbers are increasing, and moves on to the next item on the agenda. 

It took a while for me to see the problem. Every number on these presentations measures whether an employee touched an AI tool, but none of them measures whether the output or processes actually got better. 

If you believe the data, “AI fluency” seems universal. But if you listen to conversations happening in hallways afterward, employees describe tools they’ve opened a couple of times before they abandon them. Some rewrite the output manually before managers see it. Unfortunately, dashboards did not capture that part of AI adoption. 

Over twenty-five years of building and scaling technology businesses have taught me that it’s not wise to build decisions on an unverified self-report. Yet that’s exactly what an average AI dashboard runs on. For HR officers and managers, this guide will help you understand what’s going on in your workforce with regard to AI use, and it’ll teach you a seven-step process for measuring actual AI capability in a way employees cannot easily distort. 

Why your AI adoption numbers are unreliable 

The Census Bureau’s Business Trends and Outlook Survey shows that from December 2025 through May 2026, AI use among U.S. businesses sits between 17% and 20%. Firms with 250 or more employees reported use at nearly double the rate of firms with fewer than five. What this means is that two companies can check the same box while having wildly different scales of actual use. 

AI adoption signals get even worse inside an organization. GCheck’s Automation Anxiety Report, a national survey of 1,500 full-time employed US adults, found 63% of workers have lied about their AI skills to appear more knowledgeable. Surprisingly, the same survey found that 81% have engaged in at least one behavior to discourage or limit AI use at work. These findings describe the same population. 

It’s easy to see how this distorts data from both ends. On your dashboard, skills data is probably high because employees are inflating their AI skills. At the same time, usage data is unreliable because some are limiting their AI use even if the metrics say they’re engaged. 

Gallup’s 2026 State of the Global Workplace report depicts the same scenario. It shows that 65% of US workers in AI-adopting organizations report a positive impact on their productivity, but only 12% strongly agree that AI has transformed how work gets done in the entire company. 

The behaviors that never reach your dashboard 

It helps to understand the behaviors you’re actually looking for before trying to fix your measurement. Here’s how AI skills inflation happens according to the Automation Anxiety Report: 

• 40% of workers speak confidently about AI in meetings so they won’t appear behind 
• 33% let colleagues assume stronger AI skills than they have 
• 31% exaggerate hands-on experience with AI tools 
• 25% take credit for AI-assisted work as completely their own 
• 19% appear to use AI when they’re not 
• 18% volunteer for AI tasks they aren’t qualified for 
• 16% explicitly lie about having AI skills 

Meanwhile, avoidance is evident in how: 

• 53% choose manual approaches specifically to avoid deeper reliance on AI 
• 45% raise risk concerns more strongly than they personally hold them 
• 28% claim AI takes more time or effort than it actually does 
• 24% dismiss an AI tool as unhelpful when it actually helped 
• 23% treat acceptable AI output as inaccurate 

Practically none of these behaviors would get flagged in a usage report. An employee doing all of them would still be recorded as an engaged AI adopter, which could make the data truly misleading. 

Why workers do this 

First, it’s important to emphasize that this is not a character problem. In the same survey, among workers who reported inflating their AI skills, 76% said they plan to build those same skills they’re claiming. Seventy percent believed most people in their industry were doing the same thing. Meanwhile, 52% said they didn’t receive proper training. Forty-eight percent said they do not believe their employer could verify their AI skills, anyway. 

Looking at avoidance, 71% of workers who reported limiting AI use said they do not totally trust the accuracy or quality of AI output. Finally, 63% were worried that depending on these tools could erode their own skills over time. 

The picture becomes clearer when you put all of these signals together. These aren’t employees who like to lie. These are merely people who recognize that their organizations reward the appearance of AI fluency but do almost nothing to truly check it. This discrepancy is where HR needs to zero in. 

7 steps to verify AI capability 

1. Separate activity from outcome data. It’s important to distinguish the two. Logins, seat counts, session frequency, and self-rated confidence go in the activity column. Cycle time, error rate, rework volume, and throughput go in the outcome column. Don’t be surprised if the outcome column is almost empty. The imbalance there is the very thing the company needs to fix. 
2. Baseline workflows properly. Choose workflows with a countable output, such as first drafts in marketing or time to resolution of tickets in support. Record where they stand before AI adoption, so you can clearly compare what happens after AI is used. 
3. Tell employees what will be verified before you start verification. The research found 29% of workers are willing to present their skills more honestly if employers clearly disclose what they plan to independently check. Doing this should shift behavior even before the test runs. Thirteen percent said they would overstate their qualifications regardless, but this already gives you a realistic ceiling for what disclosure alone could achieve. 
4. Have people run structured demonstrations. Ask an employee to finish one real task using an AI tool while their colleague reviews the output. It’s best to keep these demos short. This shouldn’t come across as suspicion, but genuine support because nearly half of people surveyed said they want direct testing of AI competencies. 
5. Apply the same standard to everyone. Workers cited consistency as the highest trust builder when evaluating AI skills. Ad hoc testing is not advisable because it just raises suspicion, which makes data unreliable. Consistency means defining the task once and then applying it to everyone in the role. This includes recording results the same way every single time. 
6. Keep a human in the decision process. Workers also widely support establishing human review of hiring and evaluation decisions even when AI tools are used. And it makes sense. While a demo tells you what an employee can do with a tool, context about workload, tenure, or training history should still be provided by a human manager. 
7. Close the gap with training when applicable. It’s extremely important that you follow up on all these steps with a commitment to resolving issues and training employees when the data reveals the gap. More than half of workers with inflated AI claims cited missing training as the reason they overstated. Building a skills development plan should make the next round of claims more accurate and relieve some of the pressure. 

What workers are really asking for 

The research revealed that workers actually don’t resist verification. However, they do want transparency on how AI is being used in hiring decisions, as well as clear communication on what specific details will be verified. Moreover, many of the same workers admit that misrepresenting AI skills creates genuine business risk even while they believe that their coworkers are exaggerating. 

This only means that people aren’t blind to the problem. What they’re clamoring for is the proper infrastructure to fix it. 

Where to start this quarter 

Starting the verification process doesn’t need to be complicated. Simply take one workflow and baseline it correctly. Don’t forget to tell employees what you plan to check and when. Next, run a short, structured demonstration while colleagues check the output. Once gaps are discovered, direct your training budget toward closing them. 

You might think that running this across every department is a good idea, but it’s not. The advantage of choosing a single workflow first is that you can defend it better in front of a leadership team. This, in turn, might win you more budget for an improvement plan than vague top-line numbers across departments. Furthermore, it can limit the damage if your first demonstration misses the mark. This is common on the first try and will be easy to correct on your next attempt. 

Wrapping up 

Companies are used to the process of verifying claims before acting on them; this is, after all, what pre-employment background checks are for. The challenge is how to implement the same standard in an AI adoption plan. You don’t need new software or another engagement survey for this. It only requires focusing on the right metrics to measure, and more importantly, being honest with employees about why you’re measuring them.