Think You Can Spot AI? Think Again: 70% of Workers Get It Wrong

76% of Americans say it’s important to know whether content was made by AI or by humans. There’s a growing awareness of AI’s presence in our daily lives—and a desire to stay grounded in what’s real.

But new data from BambooHR and Method Research reveals something striking: while many people think they can detect AI, most can’t.

In our latest study, we asked people how confident they were in spotting AI-generated writing. Then we put their confidence to the test with a mix of AI-generated and human writing samples.

The results expose a critical gap between self-perceived skills and true AI literacy.

This gap between confidence and accuracy isn’t just interesting data. It’s a critical insight into digital fluency, trust, and readiness at work. HR leaders looking to roll out AI tools need to understand not just who can use these tools, but who can use them wisely.

In this post, we’ll explore the key findings from our original research and highlight what HR teams can do to improve AI literacy, empower overlooked talent, and build more effective AI strategies from the ground up.

Confidence isn’t competence

79% of participants believed they could spot AI-generated writing. But when we tested their ability, fewer than half correctly identified both AI-written items. In fact, just 30% of respondents accurately classified all four text samples (two AI, two human).

That leaves nearly half (47%) who were overconfident—they thought they could detect AI, but got it wrong. This is a red flag for the workplace: Overconfidence can lead to false trust in AI-generated communications, including manipulated reports, fake candidate statements, or even manipulated evidence in workplace investigations.

But underconfidence has risks, too. Employees who doubt their skills may avoid AI tools altogether—even when those tools could meaningfully improve their work.

That’s why HR must go beyond software training to build AI literacy. To be AI-ready, workers will need to know how to evaluate what AI produces and reflect honestly on their own blind spots.

Who underestimates—and who overestimates—their skills?

Gender confidence gap: Women were 60% more likely to underestimate skills

Women were significantly better than men at detecting AI-generated writing. But most women didn’t expect to be.

Women had higher detection accuracy overall, and yet they were 60% more likely to underestimate their ability. Men, on the other hand, were 34% more likely to overestimate their skills.

This finding reflects broader research on the AI confidence gap. According to a Harvard Kennedy School meta-analysis, men tend to report higher confidence in using AI tools across sectors, regardless of their actual skill level. And in many workplace cultures, that confidence—not competence—can determine who leads, who gets heard, and who shapes adoption.

The stakes are high. Women using AI at work often face a "competence penalty," a credibility gap where their use of tech is scrutinized more than their male peers'. That means women may not just be underestimating their own skills—they may be getting penalized even when they use AI tools confidently.

Mid-career workers show greatest overconfidence in AI skills

Young adults (18–24) not only scored highest in AI detection accuracy—they also had the best calibration between skill and self-assessment. In other words, they knew when they were right.

But surprisingly, it was mid-career professionals (25–44) who showed the highest levels of overconfidence. More than half of respondents aged 25–34 (51%) believed they could accurately detect AI, but their performance didn’t match up. The 35–44 group followed closely behind, with 48% displaying overconfidence.

This mid-career group often holds leadership roles or manages projects where AI tools are already in use. That makes their overconfidence particularly concerning. If decision-makers misjudge their own skills, they may implement AI strategies without sufficient guardrails, or dismiss valid concerns raised by others with sharper detection instincts.

On the other end of the spectrum, respondents aged 45 and older were most likely to underestimate themselves—a barrier that could slow adoption even among those with untapped aptitude.

Educators take the lead on AI detection skills

Detection ability didn’t just vary by demographic—it also differed significantly by industry.

Education professionals had the highest AI detection accuracy of any sector, outperforming retail, healthcare, finance, and even technology.

Educators also showed the most balanced confidence profiles, with lower overconfidence and higher rates of accurate self-assessment.

This makes sense: educators are trained to read critically, detect nuance, and question sources.

By contrast, healthcare professionals showed the highest percentage of overconfidence, with many believing they could spot AI when their performance said otherwise.

In high-stakes industries where accuracy and trust are essential, this overconfidence could have serious consequences.

What HR can do now

AI isn’t going away—but our blind spots around it can.

Here’s how HR and leadership teams can support smarter, safer AI adoption in the workplace:

Emphasize critical thinking and human oversight. Overreliance on AI can obscure the line between assistance and authorship. HR and learning teams should model transparency, encourage fact-checking, and require "AI-assisted" disclosures.

Build heuristics into workflows:

• How do I know this is true?
• What makes this sound human?

Address confidence gaps with intention. Women and older employees are more likely to underestimate their abilities; men and mid-career employees are more likely to overestimate. Don't default to the loudest voices in the room. Ensure your AI champions reflect your full workforce.

Tailor training to foster both skill and self-awareness. Pair practical tool training with exercises in calibration and evaluation. Mixed-peer review sessions can help normalize uncertainty and promote more accurate self-assessment across teams.

Encourage a culture of healthy skepticism. Employees who are overconfident can fall for misinformation, while underconfident employees may disengage from valuable tools. Make it standard practice to question, verify, and "show your work" when using AI.

The future of work depends not only on knowing how to use AI but also on knowing when to trust it, when to question it, and when to step in with a human voice.

Methodology

This report is based on a US-based survey conducted by BambooHR in partnership with Method Research in 2025. The AI detection study was part of a broader project exploring workforce trends and topics. The sample included 1,500 adults across industries, job levels, and demographic groups. Analysis included signal detection theory and categorical comparisons to examine differences in detection skill and self-assessment. Sample sizes varied across industries. While directional differences in AI detection accuracy were observed, some industry-level comparisons should be interpreted with appropriate caution.