An employee tells the boss they are making brilliant use of artificial intelligence. In reality, they may be using it occasionally, struggling to get useful answers or worrying that admitting difficulty will count against them. A survey published by workforce analytics company Visier suggests that this kind of performance has become a genuine workplace issue. It found that nearly half of surveyed American employees had exaggerated their experience or use of AI to colleagues or managers at least some of the time. The finding invites a question organisations rarely ask: when management demands more AI use at work, are employees becoming more productive or just better at looking enthusiastic? Visier's research account describes the poll and the concern behind it.
The study does not establish that employees who exaggerate are lazy, dishonest by nature or incapable of learning new tools. Some may be uncertain what their managers mean by ‘using AI’; others may fear that acknowledging problems will make them look dispensable. Nor is a survey of US employees a census of attitudes in Britain or the whole world. It is evidence about what a sample of workers said, which is valuable when companies are judging a transformation largely through their own usage dashboards.
AI use at work can become a measure of loyalty
Many employers now treat familiarity with AI as a useful skill. That can be reasonable when tools genuinely assist people with their responsibilities. Problems arise when a vague order to ‘use more AI’ becomes a proxy for commitment or performance. An employee who cannot see a useful application in a particular task may feel pressure to claim success rather than explain the limitation.
Visier questioned 1,000 US full-time employees with a third-party research partner. Forty-eight per cent reported exaggerating AI usage or expertise at least sometimes. Fifty-nine per cent said they either did not believe or were uncertain that their employer had a clear plan for what AI meant for jobs. The company also reported that 54 per cent felt their career or role had changed significantly because of AI in the previous two years. These are self-reported views within a survey, not independently observed workplace behaviour.
A related number is telling: Visier’s announcement says 45 per cent felt pressure to use AI even when they lacked confidence in doing it well. Pressure without training may create the worst of both worlds. Managers see rising claims of adoption, while staff do not gain the knowledge to apply the technology safely or judge when a task should remain manual.
What the boss thinks is happening may not be happening
A company can buy licences, announce an AI-first strategy and ask employees to report usage. Those actions are easy to count. They are not the same as reducing errors, improving service or giving staff time to concentrate. When reporting becomes the objective, workers have an incentive to produce the evidence managers expect rather than evidence of real value.
There is a distinction between using a chatbot and benefiting from it. Generating a draft can feel fast, but a badly organised or inaccurate draft may create a long editing task. A model may provide an appealing answer that requires a specialist to confirm every key claim. An employee can truthfully say they used AI and still be less productive than if they had completed the work another way.
The same incentive may exist in interviews and promotion discussions. If candidates expect to be judged on AI fluency, some will describe experience more generously than the evidence supports. The remedy is not to interrogate every worker, but to assess practical skills and ask about specific situations: what tool was used, why, what was checked and what happened when it failed.
Visier’s formal survey announcement places the exaggeration finding alongside workers’ worries about career change and the need for clearer leadership. It is important to recognise the sponsor’s commercial interest as a workforce analytics provider. The public evidence here is the company’s survey, not an independently replicated measurement of actual productivity across the same businesses.
Real productivity is harder to boast about
The gains from AI can be subtle. A worker might find it helpful for exploring a difficult spreadsheet, organising notes or creating a first draft. Another might use it to generate a large volume of unnecessary material that colleagues then have to review. A usage count treats both as adoption. A sensible evaluation asks whether a task was completed to the required standard, what checking was needed and whether the result was better than an alternative method.
The distinction becomes more important for sensitive work. It may be inappropriate to enter confidential customer information into an unapproved AI service. In a regulated workflow, record-keeping or oversight obligations could mean that an apparently quick shortcut is unacceptable. Employees need permission to explain those boundaries, rather than feeling that every refusal to automate will be read as resistance to change.
LiveAIWire’s coverage of the workplace performance gap associated with AI offers a separate view of where tools may help. Another examination of hidden work created by AI projects addresses the effort that is easy to overlook when executives focus on the visible output. The Visier survey adds the social pressure that can distort the picture presented to managers.
Training is a better answer than an adoption contest
Useful training should begin with job-specific problems rather than generic demonstrations of what a model can write. A customer service worker needs different safeguards from a software engineer or a finance administrator. Staff should learn how to judge output quality, protect personal information and recognise tasks where AI is simply the wrong choice.
Managers can help by asking for examples of failures as well as successes. If reporting an error invites blame, organisations may not hear about the weak points in their systems until an expensive incident occurs. People also need to know whether access to training, paid time to practise and suitable tools will be provided. It is unrealistic to treat a changing technical skill as a private responsibility while measuring it as a formal business requirement.
A small experiment can be more informative than a grand announcement. Choose a process, record its current time and error burden, let employees test a tool with appropriate controls, then compare the outcome honestly. The approach produces a business case that can survive scrutiny and gives employees permission to say when it does not work. That is a more useful signal than a competition over whose prompt sounds most impressive.
The danger of mistaking confidence for competence
An employee’s confident presentation and an AI system’s confident answer can reinforce each other. Someone eager to appear capable may accept plausible but incorrect output without checking it carefully. A manager eager to demonstrate progress may reward volume rather than verified accuracy. The issue is therefore organisational as well as technical.
There will be settings where AI makes a meaningful difference. It can support new skills, reduce repetitive work and help people try ideas they could not easily explore alone. But those benefits become harder to measure when employees believe they must constantly perform enthusiasm. The Visier results suggest leaders who want reliable adoption need to make honesty about limitations safe.
This also raises a question for employers who demand evidence of technological enthusiasm. A survey can capture what employees say they do, but it cannot reveal precisely how well any individual performed or whether the claimed use actually improved the work. If an appraisal rewards mentioning AI, staff may focus on presentation rather than results. A better evaluation would start with the quality of the finished work, the accuracy of any information used and whether the process respected confidentiality. It might also ask employees which tasks still need more human attention, even when software is available.
There is a wider workplace divide around confidence and capability with AI. Some people have reliable tools, patient training and time to experiment. Others face security restrictions, confusing instructions or workloads that leave no space to learn. Comparing their reported use as though everyone has the same opportunity would reward circumstances as much as skill. A sensible employer would make the rules visible, give people opportunities to develop useful skills and resist turning every meeting into a contest over who can sound most technologically advanced.
The lasting lesson is not that workers cannot be trusted. It is that a workplace can ask a misleading question and receive misleading answers in return. Instead of ‘how much AI did you use?’, the more useful question is ‘what work improved, and how do we know?’
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.