UK workers spend almost as long double-checking AI’s outputs as using it, according to new research from UnlikelyAI, a level of mistrust that could be costing enterprise firms £29 billion a year in lost productivity.

Based on a Censuswide survey of 1,000 decision-makers across critical sectors like energy, finance, insurance, and healthcare, the AI Trust Report reveals a gap between confidence in workplace AI and actual behaviour, with 99% of respondents claiming their staff regularly checks outputs when using built-in LLMs like ChatGPT, Gemini, and Claude.

These range from quick “sense checks” and minor verifying (18%) to redoing some or all of the task manually in order to verify it (20%), or even ignoring the output entirely (18%).

Despite AI’s promise to make tasks quicker, according to employers’ estimates, staff spend on average 2 hours 30 minutes every week going back and verifying, checking or redoing what it’s produced, compared to 2 hours 41 minutes using these AI tools.

Scaled to the UK’s large-business workforce, UnlikelyAI estimates this verification time equates to more than £29 billion lost every year for organisations with 250+ employees. 

Beyond productivity, the survey points to a new form of cognitive strain, with more than half (51%) of employers saying their workers report frustrations with verifying AI, leading to issues like “AI burnout”, mental fatigue from checking and rechecking outputs (32%), and “AI blindness”, losing perspective on output quality after repeated rounds of prompts and answers (30%).

Relatively few report clear cognitive upsides, with just 19% saying AI makes them feel energised and empowered, and frees up headspace, while only 17% believe it makes them better at their job, compared to 33% who believe dependence on AI is costing them hard-won skills.

Although the majority of those surveyed (87%) claim to trust the outputs of AI at work, over 65% still say they feel anxious or nervous when using AI tools for work that others will see, while 31% find it difficult to decide whether to trust AI or their gut.

Among the major issues undermining trust is the inability to explain where AI prompts go, how they are used or how outputs are generated, while workers also report constant fears of AI hallucinations or factually incorrect outputs.

Recommended reading

• Financial Services Faces an AI “Reality Check”, Warns Aveni
• KPMG UK Interview: Why Trust is AI’s Most Underrated Asset
• 88% of UK Execs Would Trust AI To Handle a Crisis

“These findings highlight a critical challenge: there has to be a better way to use AI,” said William Tunstall-Pedoe, CEO and founder of UnlikelyAI.

“LLMs have strengths in specific, limited areas, but there’s a huge lack of understanding about when to use them and when to look to other, less fallible models. That’s where this trust gap is coming from.”

To fix the productivity drain of untrustworthy AI, the study suggests firms establish clear ground rules around AI hygiene, educate their staff on the capabilities and limits of AI, and prioritise explainability over novelty, choosing tools that produce consistent, verifiable outputs that leave a transparent audit trail.