77% of Australian AI users report at least one form of botshitting: shipping AI-assisted work they haven’t adequately checked, don’t fully understand, or couldn’t confidently defend.
That is the highest rate across the three countries surveyed for the Work AI Index 2026, compared with 70% in the UK and 64% in the US.
It’s also difficult to reconcile with Australia’s adoption numbers.
90% of Australian digital workers use AI at work, on par with the UK and ahead of the US. 72% say it makes them more productive, and workers report saving roughly 10 hours a week through AI automation.
They are also bringing AI further into the working day:
- 54% reach for AI before trying to solve a problem themselves.
- 74% have used AI as a meeting notetaker.
- 66% have had AI facilitate a meeting.
- 58% have sent a digital twin to a meeting on their behalf.
Australia has moved faster than the US and matched the UK on overall adoption. On organisational impact, it ranks last.
Only 10% of Australian workers say AI has significantly improved their organisation’s performance, compared with 12% in the US and 18% in the UK.
The Work AI Index 2026: Australia, from Glean’s Work AI Institute, surveyed 1,500 Australian digital workers as part of a global study of 6,000 full-time workers. We also spoke with dozens of AI leaders and analysed anonymised, aggregated workplace AI interactions from Glean’s enterprise AI platform.
What we found is a workforce that has embraced AI faster than its organisations have changed the way work happens around it.
AI has moved into the work faster than the work has changed
Adopting an AI tool can happen quickly. An organisation enables a licence, approves a platform, or gives employees access to a new assistant.
Changing the work around it takes longer.
That means deciding which tasks AI should handle, what context it needs, how its output should be reviewed, who remains accountable, and what employees should do when the tool fails.
Across Australian organisations, those systems are still catching up:
- 57% of workers say their organisation provides enough AI training and support.
- 59% say workflows have been redesigned to take advantage of AI.
- 62% say their organisation’s AI policy is regularly reviewed.
- 9% say no one is primarily responsible for overseeing AI use.
- 58% say important information they need isn’t accessible through their AI tools.
In many cases, AI has been added to work that still operates much as it did before. The same handoffs, approvals, systems, and information gaps remain. AI may produce the first answer faster, but the employee still has to fit that answer into a process that wasn’t designed for it.
The result is a high failure rate. Australian workers say 40% of their AI sessions require substantial rework or a full reset.
When AI fails that often, the repair work moves to the employee. They provide missing context, rewrite prompts, check sources, correct mistakes, compare outputs, and rebuild work that looked finished before it was dependable.
Australian workers spend an average of 6.5 hours a week on this labour, which we call botsitting. It accounts for 38% of their AI-related time, more than the 34% they spend using AI to produce work.
When that labour is left unsupported, workers eventually start taking shortcuts. The output moves forward without another check. A weak assumption survives because the deadline is close. A polished answer feels complete enough to send.
That is how fast adoption becomes botshitting.
Australia’s largest organisations have more to unwind
The adoption numbers also conceal a structural challenge.
Many of Australia’s largest organisations were established generations ago. The average ASX 20 company is approximately 104 years old, based on the analysis conducted for the Australian Work AI Index.
Age alone isn’t the problem. These organisations have endured because they built strong institutions, deep expertise, and processes capable of operating through economic, regulatory, and technological change.
But those strengths come with complexity.
Older organisations are more likely to carry layers of core systems, accumulated policies,
established approval paths, and workflows that connect several functions. A change in one part of the organisation can affect systems and responsibilities elsewhere.
As Dominic Price, Partner at Be Luminous and a contributor to the Australian report, put it:
“Australia is not short on AI ambition. But many of its biggest companies are carrying far more legacy than their global peers.”
Adding an AI tool to that environment is much easier than rewiring the organisation around it.
A new assistant may be able to summarise a customer record without understanding why one field is routinely ignored. It may retrieve the formal process without knowing which exception keeps the work moving. It may draft an analysis without access to the system where the most current figures live.
Employees then bridge the gap between the new technology and the existing organisation. They carry information between systems, explain unwritten rules, and decide which part of the output can be trusted.
That work can make an AI rollout look successful at the surface while employees absorb the complexity underneath.
Australians still hold the human responsible
One finding makes Australia’s botshitting rate particularly interesting.
When AI-generated work fails, 49% of Australian workers blame the person who used the tool. Only 22% blame the machine, while 29% are unsure who is responsible.
At first, this sounds like a sign of strong accountability. If the person remains responsible for the result, you might expect the work to receive more careful review.
But Australia also has the highest botshitting rate of the three countries studied. 44% of Australian workers sometimes deliver AI-assisted work they haven’t fully checked, and 45% report shipping work they couldn’t explain if asked.
The contradiction points to a distinction between accountability in principle and accountability in practice.
Workers can believe that the human should own the outcome while operating in a system that makes careful review difficult to sustain. The deadline still rewards speed. The AI output still arrives polished. The employee still has several other tasks waiting. Checking the work may be understood as their responsibility without being given enough time, context, or support to do it well.
The human is held responsible after the failure, but the workflow does little to help them prevent it.
That’s an operating problem, not simply an individual lapse in judgement.
When AI fails, workers are improvising
How workers respond when AI goes wrong provides another clue.
33% of Australian workers say they humanise AI when it fails. They say please, soften their tone, encourage it, or tell it to try harder. That compares with 24% in the US.
Australian workers are also less likely than their US counterparts to use several practical repair strategies, including rephrasing the request, adding context, redoing the work themselves, or switching to another tool.
Humanising AI isn’t inherently a problem. A polite tone may have no effect on whether the next answer is better or worse.
But the wider pattern suggests that many employees are approaching AI failure without a reliable method for diagnosing it. They know the answer isn’t right, but not necessarily why.
The problem could be the prompt. It could be a missing source, stale information, an unclear instruction, or a task the tool isn’t suited to complete. Without training in how to distinguish between those failures, workers fall back on a familiar habit from human collaboration: encourage the other side and try again.
That response becomes even more understandable when the right information isn’t available to the tool. 58% of Australian workers say critical information they need isn’t accessible through their AI systems, the highest share of the three countries surveyed.
An employee can ask more politely, simplify the instruction, or rerun the prompt. None of those changes will help if AI can’t access the current project plan, understand which forecast is final, or apply the internal definitions that shape the task.
Workers end up improvising around an information problem they can’t solve from the prompt box.
Australia needs to design for failure, not only adoption
The next phase of AI adoption requires more than teaching workers how to use a tool when everything goes well.
Employees also need to know how to recognise and recover from failure. That means building four capabilities around the technology.
Redesign workflows around what AI changes
AI shouldn’t be added to every existing step without reconsidering whether those steps still make sense.
Organisations need to map where AI enters the work, which handoffs it changes, what new review is required, and where accountability sits when both a person and a system contribute to the result.
The aim isn’t to preserve every pre-AI process. It’s to remove steps that no longer add value while protecting the judgement, expertise, and checks the work still requires.
Give AI company context, not only access to data
Connecting AI to more systems isn’t enough. It needs to distinguish current information from outdated versions, authoritative sources from working drafts, and relevant context from everything it could retrieve.
That distinction has measurable consequences. In Australia, workers in context-rich AI organisations are:
- 53% less likely to feel worn out by AI.
- 45% less likely to ship work they can’t explain.
- 47% less likely to use unapproved tools.
Better context doesn’t remove human review. It reduces the repetitive reconstruction that consumes employees’ time before meaningful review can begin.
Train people to diagnose what went wrong
Training often focuses on access, prompting, and approved use. Those skills matter, but they don’t prepare employees for the 40% of sessions that fail.
Workers need practical ways to identify whether the problem is missing context, poor source quality, an unsuitable model, a badly scoped task, or a workflow that still needs human ownership.
They also need to know when another prompt is unlikely to help.
AI fluency includes knowing how to repair a failed interaction and when to stop relying on the tool.
Make ownership visible
AI governance has to answer questions employees encounter during the work:
- Who can approve an AI-assisted output?
- Which tasks require a source check?
- When can an agent act without another review?
- Who owns the result when several people and tools contributed?
- Where should an employee go when the approved system can’t complete the task?
A policy that is reviewed regularly matters. So does having someone clearly responsible for AI use. But those structures have to translate into decisions employees can make while the work is happening.
Impact has to be designed
Australia’s rapid adoption is still an advantage. Its workers are willing to experiment, bring AI into consequential workflows, and rethink how much of their work the technology could handle.
But speed becomes an advantage only when the organisation can absorb it.
Right now, workers are doing much of that absorbing themselves. They reconstruct missing context, repair failed sessions, reconcile AI with legacy processes, and decide on their own when an answer is good enough to move forward.
The next phase is to move that burden out of individual improvisation and into the way work is designed.
Australia doesn’t need to slow down. It needs the training, context, governance, and workflows that let fast adoption produce work people can trust.
Adoption can spread through enthusiasm. Impact has to be designed.
Read the full Work AI Index 2026: Australia to understand where Australia’s AI gains are going and what organisations can build around the technology to make them last
