WalkMe, an SAP company
WalkMe Research · 2026

The AI
Confidence Trap

Based on original research among 2,037 U.S. workers who use AI on the job. Confidence in AI has never been higher. Results have not followed.

90%
feel confident using AI at work
24.6%
say it works on the first try
50.2%
have lost time to AI on a task they could have done faster themselves
Scroll to explore
Executive summary

Comfort with AI is rising. Competence is not.

The third annual AI at Work Pulse Survey finds a workforce that has stopped hiding its AI use and started overestimating its results. Nine in ten workers feel confident. One in four says the tool does what they needed on the first attempt. Half have spent more time fighting AI than the task would have taken by hand.

The encouraging news is that concealment is falling. Every behaviour we have tracked since 2025 moved in the right direction. What has not moved is whether the work gets better, and that is now the question leaders have to answer.

How to read this research

  • Act 1 sizes the distance between how confident workers feel and what actually happens at the keyboard.
  • Act 2 shows what that distance costs, and who is caught in it. It reaches further up the org chart than expected.
  • Act 3 isolates what changes outcomes: support that lives inside the workflow rather than beside it.
The confidence
90.0%

of workers say they are confident or very confident using AI for their work tasks. Among senior leaders it reaches 92.6%.

The reality
24.6%

say AI does what they needed on the first try. Another 44.0% get there only after adjustments.

The signal
30.2%

want guidance built into the tools they already use, not separate training. It is the second most requested improvement of eight.

Bottom line: the gap is no longer about willingness or access. It is about whether support reaches people at the moment they are doing the work.

Source: WalkMe AI at Work Pulse Survey 2026. 2,037 U.S. workers who use AI on the job, fielded by Propeller Insights, June 10 to 20, 2026.

The Confidence Gap

Almost everyone feels capable. Far fewer are getting a usable result the first time they ask.

65.4

percentage points separate the share of workers who feel confident using AI from the share who say it works on the first attempt.

Confidence

Feeling capable and being effective are not the same thing

Share of workers, 2026

Feel confident using AI at work90.0%
Say AI works on the first try24.6%
Have lost time to AI on a task50.2%
Confidence is close to universal. Outcomes are not, and the distance between the two is the finding.

Source: AI at Work Pulse Survey 2026, base n=2,037.

First-try success

Most AI work needs a second pass

How often AI does what was needed on the first attempt

Often, usually right with minor adjustments44.0%
Almost always, right the first time24.6%
Sometimes, I need multiple tries23.5%
Rarely, I usually rework or redo it6.4%
Almost never, I have given up expecting it to work1.5%
Three quarters of workers are editing, retrying, or reworking. That effort is invisible in every adoption dashboard that counts log-ins.

Source: AI at Work Pulse Survey 2026, base n=2,037.

Year over year

Every tracked behaviour improved. The outcome did not.

Measures asked in the same terms in 2025 and 2026

Feel confident using AI for work tasks
91.5%
→
90.0%
Have spent more time using AI than the task would have taken manually
59.0%
→
50.2%
Manager or team expects more output in the same time
58.6%
→
51.3%
Pretended to know how to use AI in a meeting or presentation
45.2%
→
28.3%
Used AI then presented the work as their own to avoid judgment
48.7%
→
32.5%
Received conflicting messages about which AI tools are allowed
51.0%
→
39.3%
Workers are not becoming more deceptive. They are becoming more comfortable, and comfort is not the same thing as capability.

Source: AI at Work Pulse Survey, 2025 base n=1,000 and 2026 base n=2,037. The 2024 wave used different screening criteria and is not comparable at the sample level.

What The Gap Costs

Output expectations rose faster than capability. People compensated, and the compensating runs all the way to the top.

51.3%
say AI led their manager or team to expect more output in the same amount of time
33.3%
have pretended to be more skilled or knowledgeable with AI than they are
32.5%
have used AI and then presented the work as their own to avoid judgment
Generation

The most confident group is the most exposed

Gen Z, ages 18 to 29, against Baby Boomers, 62 and over

Gen Z Boomers
Feel confident using AI94.1% vs 75.5%
Have overstated their AI skill45.0% vs 12.5%
Say it caused a real problem at work31.0% vs 6.8%
The generation most fluent with AI is the one most likely to have overstated its ability, and four times more likely to say it caused a mistake, a missed deadline, or lost trust.

Source: AI at Work Pulse Survey 2026. Gen Z n=471, Boomers n=192. Both bases are smaller than the full sample.

Leadership

The trap does not thin out near the top

Senior leaders, meaning C-level, President and Vice President, against individual contributors

Approved an AI tool they personally cannot use
Senior leaders39.1%
Individual contributors25.3%
Pretended to understand the company AI strategy when speaking to employees, peers or board members
Senior leaders41.6%
Workers who have felt a senior leader does not fully understand the strategy they champion publicly
All workers53.6%
Confidence is highest where the purchasing decisions are made. Nearly four in ten senior leaders have approved a tool they could not use themselves.

Source: AI at Work Pulse Survey 2026. Senior leaders n=202, individual contributors n=760. Subgroup differences under roughly six points are directional.

Workers stopped hiding their AI use. They started overestimating what it produced. Those are different problems, and only one of them has been solved.

The AI Confidence Trap, WalkMe Research 2026
Escalation

When people get stuck, nobody is the answer

First move when a manager does not know how to do something with AI at work

Ask the AI itself for help22.4%
Search online for tutorials or examples20.0%
Trial and error on their own19.2%
Ask their own manager12.7%
Ask a coworker12.0%
Check internal company documentation11.5%
Give up and do the task without AI2.2%
Managers are not the escalation path, and neither is documentation. More than six in ten managers solve it alone or ask the tool that just failed them. 26.8% have pretended to know how to use AI in a meeting themselves.

Source: AI at Work Pulse Survey 2026, managers only, base n=731. Single selection.

What Actually Works

The workers getting results are not the ones with more tools or more courses. They are the ones with support inside the workflow.

33.1%

of workers say that when AI worked well, the reason was guidance on the screen while they were doing the work. It is the highest-ranked factor an employer can actually provide.

What workers ask for

The request is guidance in the tools, not more courses

What would most improve the experience of using AI at work. Multiple selection, so options do not total 100%.

Better integration between AI and other work tools33.7%
Guidance built into the tools I already use, not separate training30.2%
Real examples of how people in my role use AI well30.1%
AI that understood my specific role and tasks better29.8%
Help knowing when to use AI and when not to26.7%
A culture where using AI is openly accepted, not judged24.9%
Clearer permission about which AI tools are allowed21.2%
Nothing, my current AI experience is great11.6%
The top three answers describe the same thing from three angles: make the help appear where the work happens. Only 11.6% say nothing needs to change.

Source: AI at Work Pulse Survey 2026, base n=2,037.

What made it work

On-screen guidance is the top thing a company controls

What made the difference when AI worked well. Multiple selection.

Conditions, not company actions
I had time to experiment without pressure41.9%
The tool was easy to figure out on my own40.7%
Things an employer provides
Guidance right there on the screen as I used it33.1%
Someone showed me exactly how to use it for my tasks29.4%
Clear company guidance on what was allowed26.3%
Coworkers I could ask for help23.4%
The two highest answers are circumstances, not interventions. Nobody can budget for unhurried time. On-screen guidance is the first item on this list a company can actually deliver.

Source: AI at Work Pulse Survey 2026, base n=2,037. 3.3% said AI has not worked well for them yet.

The training gap

Courses are not reaching people, and the rules are not clear

Preparation to use AI at work based on training received, and clarity of policy

Have had no AI training at all23.3%
Only a few hours of training in total18.8%
Moderate training, but it does not feel like enough18.6%
Have received conflicting messages about which AI tools are allowed39.3%
No training at all is the single most common answer to the training question. Four in ten are also getting mixed signals about what they are permitted to use.

Source: AI at Work Pulse Survey 2026, base n=2,037.

The next measure of AI is not adoption. It is whether the work got better.

Access has been solved. Comfort has been solved. What remains is the harder question: can an organisation show where AI is being used, and what changed as a result? On the evidence here, most cannot, and the workers closest to the problem are asking for the same thing in three different ways.

Read the full release Media enquiries
Sample
2,037 working U.S. adults
Screening
Employed, uses AI on the job
Fielded by
Propeller Insights
Field dates
June 10 to 20, 2026
Prior wave
2025, n = 1,000
Wave
Third annual