Glean's Work AI Institute has published research showing that 90% of Australian digital workers use artificial intelligence at work. But only 10% of those surveyed said the technology had significantly improved organisational performance.
The Australian findings are part of the institute's inaugural Work AI Index, which surveyed 1,500 full-time digital workers in Australia as part of a wider study of 6,000 workers across Australia, the US and the UK.
The figures suggest a gap between widespread AI use and the business results companies say they want. While 72% of Australian AI users said AI makes them more productive, much of that value is being lost to hidden labour, repeated checking of output, switching between tools and failed sessions that need to be restarted.
Australian workers said AI already automates 27% of their work output, and they expect that share to rise to 34% over the next year. Even so, the study found that faster completion of individual tasks has not translated into broader gains for employers.
Hidden labour
One of the clearest findings was how much time workers spend managing AI rather than benefiting from it. Australian digital workers spend 6.5 hours a week "botsitting", which the report described as the work required to make AI usable.
About four in 10 respondents said AI sessions fail outright, forcing them to restart, substantially rework the task or go back to the beginning. Some 43% said they feel worn out by AI tools, a higher share than in the US.
The research also highlighted AI's spread into routine office work. It found that 74% of Australian workers have used AI as a note taker in meetings, 66% have had it facilitate a meeting and 58% have sent a digital twin in their place.
AI is also reaching more sensitive areas of working life. One in three Australian workers said it is already being used in performance evaluations, while one in five said it is already being used in termination decisions. Another 35% said they expect AI to be used in termination decisions within a year.
Quality checks
The study found a growing quality-control burden around AI-assisted work. Some 73% of Australian workers said they had corrected or redone AI-assisted work in the past month, and 30% said they do so at least weekly.
It also found that 77% of Australian AI users admitted to at least one unchecked AI-output behaviour. Among them, 45% said they had delivered AI-generated work they could not fully explain, while 36% said they had blamed AI for a mistake that was actually their own.
The research also pointed to poor access to the information AI systems need to work well. More than half of Australian workers, or 56%, said important information needed to do their jobs is not connected to or accessible through AI tools. The same share said vendor contracts limit how effectively their organisation can use AI.
In these "context-poor" AI environments, workers were 112% more likely to feel worn out by AI, 81% more likely to send work they could not explain and 90% more likely to use unapproved tools.
Workforce risk
The findings also linked AI-related strain to staff retention. Frequent botsitters in Australia were 59% more likely to be actively looking for another job, while workers who admitted sending unchecked AI-generated work were 3.4 times more likely to be job-hunting.
The report contrasted those results with what it described as transformative organisations, where workers were more likely to say their employers provide enough AI training and support, review AI policies regularly, explain governance decisions and reward AI skills.
In a statement accompanying the findings, Dr Rebecca Hinds said the central problem is that many organisations focus on adoption rather than outcomes.
"Too many companies are treating AI adoption like a vanity metric - more seats, more prompts, more usage," said Dr Rebecca Hinds, Head of the Work AI Institute at Glean. "But adoption alone doesn't create transformation. The state of AI at work in Australia points to a bigger opportunity: when organisations pair AI enthusiasm with the right operating discipline, they can turn productivity gains into lasting business impact instead of losing them to rework, cleanup, and unverified output."
The next challenge for companies is not persuading employees to try AI, but building the management systems, training and governance needed to use it consistently, the institute said.
Dom Price, identified as a Work AI Institute expert and Work Futurist, said the issue is organisational change rather than software uptake.
"Australia has always been good at moving quickly and making new technologies work in practical ways," said Dom Price, Work Futurist, Work AI Institute. "But AI is different because it doesn't just ask organisations to adopt another tool - it asks them to change how work gets done. Right now, too many companies are trying to push AI-speed change through legacy-speed systems. The winners will be the ones that build the human infrastructure around AI: clearer decision-making, better context, stronger governance, and teams that know when to trust AI, when to challenge it, and when to keep the work human."