And it’s not alone

Nearly one-third of employed Canadians surveyed by TD and Ipsos say they have exaggerated their AI abilities. Other studies suggest workplace expectations are rising much faster than training, policy and verification.

AI fluency has become workplace currency before many organizations have established what the term means, how employees should acquire it or how claimed proficiency should be assessed.

Nearly one-third of employed respondents, 32%, in the new 2026 TD AI Insights Report⁠ agreed that they had exaggerated their ability to use AI to workplace colleagues.

Confidence remains low across the wider population. Three-quarters of the 2,501 Canadians surveyed rated their ability to use AI efficiently and effectively at a “C” level or lower. Only 4% awarded themselves an “A.” The pressure to appear proficient is evident in the rest of the findings. Among employed respondents, 78% believe workplace AI adoption is inevitable, 59% say AI makes them more productive and 58% believe it gives them a competitive advantage over people in similar jobs.

Employer support has not kept pace. Only 37% agreed that they had received adequate workplace AI training, while 53% said their organization was falling behind others in AI adoption. A skill perceived as inevitable, productive and professionally advantageous carries considerable signalling value. Employees have an incentive to use and develop it even when they have received little help developing it.

The generational results also challenge assumptions about which workers are most convinced that AI will transform employment. Boomers were the most likely to call workplace adoption inevitable, at 89%, followed by Gen X at 81%, Millennials at 79% and Gen Z at 65%.

The survey results suggest employers also need to create the conditions in which that confidence can develop. “The future of work will require organizations to help people develop new skills as technology evolves,” says Melanie Burns, Senior Executive Vice-President and Chief Human Resources Officer at TD Bank. “Those that prioritize learning and equip employees with the right tools and support will ensure their people thrive.”

Separate studies find the same gap

A May 2026 U.S. study produced a remarkably close result.

In GCheck’s Automation Anxiety Report⁠, 31% of 1,500 full-time workers said they had exaggerated their experience with AI tools, just one percentage point below the TD figure.

GCheck’s widely reported headline number was much higher: 63% had engaged in at least one form of what the company called “AI fluency theatre.” That broader measure included speaking confidently about AI to avoid appearing behind, allowing colleagues to assume greater expertise and taking credit for AI-assisted work. Sixteen per cent explicitly admitted lying about possessing AI skills.

The structural findings were also similar. Among GCheck respondents who reported some form of AI skills inflation, 52% said they had not received proper training or support. Across the full sample, 64% said an employer had never attempted to verify their AI abilities.

The Canadian results from Employment Hero’s 2026 AI Paradox Report⁠ add another dimension. Sixty per cent of Canadian workers rated their AI competence as low or average, while 51% said their employer was doing little or nothing to develop their skills.

Workers were also uncertain about whether AI use was culturally acceptable. Forty-three per cent felt guilty using AI to produce work, 39% said it felt like cheating and 34% admitted hiding their AI use from their employer. Some employees are exaggerating their fluency. Others are concealing the fact that they use AI at all. Both behaviours indicate unclear expectations.

The international picture is also similar. BCG’s 2026 AI at Work survey⁠, covering nearly 12,000 employees and leaders in more than a dozen markets, found that 72% believed the skills expected of them had changed because of AI. Only 36% felt they had received adequate upskilling, almost identical to TD’s Canadian result of 37%.

“AI skills” still lacks a common definition

Part of the credibility problem is the breadth of the category itself. Using an assistant to summarize a meeting, verifying model output, protecting confidential data, designing an automated workflow and developing a machine-learning system can all be described as AI skills. They represent very different levels and types of capability.

An A-to-F self-assessment made without a shared standard may measure confidence as much as competence. The same problem affects resumés and job descriptions that request “AI proficiency” without specifying the tools, tasks or performance level involved.

Employers can improve the signal by defining AI capability at the role level. Employees should know which tools are approved, which tasks can be delegated, what information must remain protected, when human review is required and what successful performance looks like.

“AI is becoming more prevalent across the world of work but for many people it still feels new and unfamiliar,” says Luke Gee, Senior Vice President, Chief Analytics and AI Officer at TD Bank Group. “The pace of change can feel fast, and it’s natural that confidence takes time to catch up. We’re in a period where people are building new skills and figuring out how these tools can support the work they already do. What matters most is helping people feel supported as they learn, build confidence over time, and use AI in ways that are practical and meaningful to them.”

Training should use real workplace processes rather than generic demonstrations. Assessments should rely on practical work samples instead of self-declared fluency. Organizations should measure improvements in completion time, quality, error rates and customer outcomes instead of counting tool logins or course completions.

Productivity claims also require direct measurement. TD found that 59% of employed respondents believed AI made them more productive. A separate June 2026 Angus Reid Institute survey⁠ found that 38% of Canadian workplace AI users reported improved productivity, while 49% reported no impact. The surveys used different questions and samples, but the variation shows why self-reported productivity should not substitute for operational evidence.

The 32% figure is ultimately a governance signal. AI has entered professional identity, hiring and career competition while organizational training and verification remain immature. Companies that define and develop the required capabilities will gain a clearer picture of what their workforce can actually do. Those that leave AI proficiency undefined will continue managing a mix of inflated credentials, concealed usage and uncertain returns, which can lead to disastrous outcomes with such a powerful yet why the misunderstood technology.

Four ways companies can make employees feel more secure about AI

1. Establish a clear, practical AI-use policy.

Employees need to know which tools are approved, what company information may be entered, when AI use must be disclosed, which outputs require human review and who remains accountable for the final work. The policy should also reassure employees that they will not be penalized for using approved tools appropriately or admitting when they need help. According to the Angus Reid Institute⁠, only 31% of Canadian workplace AI users report having an official policy in place; 30% say one is being developed and 39% report having none.

2. Teach employees how LLMs actually work.

Training should begin with foundational AI literacy rather than jumping directly to prompting techniques or product demonstrations. Employees should understand that large language models generate probable responses from learned patterns; they do not retrieve truth from a conventional database or “know” facts in the human sense. Training should explain context windows, training data, retrieval, reasoning, model limitations and the difference between a model and the tools or data sources connected to it. Practical skills become much easier to develop once employees understand the underlying system.

3. Explain AI’s failure modes and provide verification procedures.

Organizations should clearly articulate that AI systems can hallucinate, produce inconsistent results, rely on outdated information, invent citations, mishandle unfamiliar proper nouns and names, confuse similarly named entities and present incorrect information with complete confidence. Employees should be taught which claims require checking, which sources are acceptable and when an output must be escalated to a subject-matter expert. This allows workers to treat verification as part of the process instead of viewing an AI failure as evidence that they personally used the tool incorrectly.

4. Provide regular briefings as the industry changes.

A major part of the challenge of being AI confident is the pace at which technology evolves. Education cannot be handled as one-time courses. Companies should provide short, recurring briefings covering important model releases, newly approved tools, discontinued features, emerging risks, regulatory changes and lessons from internal use. A central, continuously updated guide can give employees one reliable place to check current policies and capabilities. Without that support, workers are left trying to follow an industry that changes daily and weekly through often unreliable social media commentary, headlines, and informal advice.

Methodology: Ipsos surveyed 2,501 Canadian adults online from February 17 to 23, 2026, and weighted the full sample to Census demographics. The reported credibility interval for the full poll is ±2.4 percentage points, 19 times out of 20. According to the detailed tables⁠, the workplace statements used an employed subgroup of 983 unweighted respondents, with a weighted base of 910. All studies cited above rely substantially on self-reported behaviour and use different samples and questions, so comparisons should be treated as directional.