New UK data suggests that AI adoption is outpacing formal training, leaving companies exposed to poor-quality outputs, shadow AI use and growing distrust around what is real and what is synthetic.
Artificial intelligence has moved from experimentation into everyday work, but many businesses have not built the skills or governance needed to keep up.
New analysis from GROFU, drawing on Office for National Statistics data, suggests that 90% of businesses are not fully training their workforce to use AI, even though 55% of workers are already using the technology. Around one in five of those workers is reportedly using AI tools without employer approval.
That creates a growing gap between adoption and competence.
Businesses may be deploying generative AI across marketing, research, administration and content creation, yet many employees are still learning through trial and error. The result is not only inefficient use of the technology, but also greater exposure to hallucinations, weak outputs, privacy risks and unsanctioned tools.
Read More:
AI Adoption Is Becoming Informal
The data suggests a significant amount of workplace AI use is happening outside formal company structures.
Only 35% of businesses officially use AI, according to the figures cited in the research, yet more than half of workers are already using it. One in five workers is using AI without employer sanction.
This is increasingly being described as shadow AI: employees adopting generative AI tools independently because they are useful, accessible and often faster than waiting for an organisation to approve a formal solution.
The problem is that convenience can come with risk.
Without clear policies, employees may enter confidential information into external tools, rely on inaccurate outputs or use systems that have never been reviewed for data protection, security or compliance.
The Training Gap Is Affecting Output Quality
GROFU argues that poor AI training is already visible in the quality of content being produced online.
Patrick McFadden, Managing Director at GROFU, said: “What is evident is that a large amount of the ‘slop’ we find on our screens is down to people not knowing how to use AI effectively and efficiently.”
The release also cites research suggesting that two-thirds of LinkedIn posts have been flagged as AI-generated, while the platform has introduced mechanisms allowing users to report content they believe resembles “AI slop.”
The wider point is not that AI-generated content is inherently poor.
Generative AI can accelerate research, drafting, design and production. But without people who know how to prompt, verify, edit and contextualise outputs, faster production can simply create more low-value material.
AI Literacy Is Becoming an Operational Requirement
The workplace AI skills gap goes beyond learning how to write better prompts.
Employees increasingly need to understand where models can make mistakes, how to verify information, which data should never be entered into public AI systems and when human review is essential.
Managers need another layer of literacy. They must decide which tools are appropriate, which tasks can be automated and where accountability remains with a person.
McFadden said: “AI should not be about removing human expertise, but about refining, enhancing and improving workflows so knowledge can be better utilised.”
That distinction may become one of the most important factors separating effective AI adoption from superficial adoption.
Trust Is Becoming Part of the AI Problem
As AI-generated content becomes more common, public confidence is also weakening.
The analysis cites figures suggesting that 81% of people struggle to distinguish real information from fake, while 27% do not believe AI will benefit them.
This introduces another problem for businesses.
Employees are being encouraged to use AI at the same time that customers and workers are becoming more suspicious of AI-generated information. Organisations therefore need to improve not only the technology they use, but also the transparency and quality controls around its output.
A company that uses AI badly may save time in production while losing trust in the process.
Job Fears Are Running Ahead of Current Evidence
Concerns about automation remain strong.
Around 49% of people aged 30 to 49 believe AI could put their job at risk, while 47% of those aged 16 to 29 share the same concern. Yet only 7% of businesses report reducing headcount after implementing AI.
That gap between perception and current evidence matters.
Workers who do not understand how AI fits into their role may see it primarily as a threat. Formal training can make its function clearer by showing where AI can remove repetitive work, support decision-making or improve existing workflows.
“If staff are properly trained to use the tools at their disposal, two things happen,” McFadden said. “The fear drops, because they can finally see where they fit. And the work gets better, because they’re using AI to enhance what they do rather than replace it.”
Shadow AI Will Be Hard to Eliminate
The spread of unsanctioned AI tools presents businesses with a difficult choice.
Companies can try to block external systems, but employees often adopt them because they solve genuine productivity problems. A more sustainable approach may be to create approved AI environments, clear usage policies and practical education around responsible use.
That means AI governance cannot remain purely an IT responsibility.
Security teams, HR, legal departments, managers and employees all need to understand how AI is being used across the organisation.
The companies that manage this transition well are likely to treat AI literacy in the same way they treat cybersecurity awareness: as a baseline capability rather than a specialist skill.
The Next AI Advantage May Be Human Expertise
The current phase of AI adoption has been dominated by access to models and tools.
The next phase may depend much more on whether organisations can build the human capabilities needed to use them responsibly.
If almost everyone can access similar AI systems, competitive advantage will not come simply from having a chatbot or generative AI licence. It will come from how effectively employees combine those systems with expertise, judgement and verification.
The training gap identified by GROFU therefore points to a broader reality: businesses may not need more AI tools as urgently as they need people who understand how to use the ones they already have.
About GROFU
GROFU is a brand creative and performance agency based in Stockport. Founded by Patrick McFadden in 2021, the agency works across sectors including technology, finance, property, retail and food and beverage.
Its analysis draws on Office for National Statistics research covering artificial intelligence in UK businesses from 2023 to 2026, alongside ONS public opinion data from June 2026.
