Who is workplace AI leaving behind? PwC finds an access gap—but not its cause
PwC surveyed 49,364 workers across 48 countries and regions. AI use rose, while reported access to learning fell; the cross-sectional responses show an association, not proof that AI caused the divide.
The short version
- What happened
- PwC collected 49,364 responses across 48 countries and regions and 29 sectors in May and June 2026.
- Who it affects
- Workers without suitable tools or learning time may miss opportunities even when AI adoption rises around them.
- What remains unproven
- The analysis is cross-sectional and cannot establish that AI use caused confidence, security or career outcomes.
Keep in mind
The analysis is cross-sectional and cannot establish that AI use caused confidence, security or career outcomes.
Worker groups are constructed from self-reported indexes and should not be treated as fixed occupational categories.
Global weighting does not make every national, sector or occupational subgroup independently representative.
Source analysis · 2 linked sources. Primary analysed source date: 29 September 2026.
Read the full report · 5 min The evidence, methods, implications and limitations
What the survey measured
PwC published its Global Workforce Hopes and Fears Survey 2026 on 29 September. It gathered responses from 49,364 workers across 48 countries and regions and 29 sectors in May and June 2026. PwC says the results were weighted to the age and gender distribution of the working population in each country or region. That is a large multinational survey, but it remains a record of what respondents reported at one point in time rather than an observation of their actual output, pay records or job transitions.
Across the weighted sample, 64% said they had used AI at work in the previous 12 months, ten percentage points more than the previous year. Daily generative-AI use rose from 14% to 22%, while 59% expected their use to increase during the next year. At the same time, reported access to learning and development fell from 59% to 51%. These are separate survey measures. Their movement together does not prove that increased AI use reduced access to training.[1][2]
How PwC created four worker groups
PwC built two indexes. One estimates labour-market advantage from respondents' views of demand for their roles and how difficult their skills would be to replace. The other estimates AI advantage from self-reported improvements in work quality, creativity, skills and value to an employer. Combining high and low groups produced four archetypes: front-runners, AI insurgents, indispensables and the engine room. The largest group, labelled the engine room, represents 56% of respondents; front-runners represent 14%, AI insurgents 18% and indispensables 11%.
The labels are an analytical construction, not occupational diagnoses. A worker can have valuable skills that a survey index does not capture, and perceived employer demand may not match vacancies or wages. The 56% figure should therefore be read as the share assigned to that survey cluster, not as proof that more than half of the global workforce has been objectively left behind. Country, sector and job-level detail would be needed before an employer could use the grouping locally.[1]
The practical question for employers
Daily generative-AI users reported greater job-security confidence than infrequent users: 68% compared with 57%. They were also more likely to trust management, expect to learn new skills and seek promotion. Selection is a major alternative explanation. People in better-resourced roles may receive both earlier access to AI and more training, autonomy and managerial support. Confidence could encourage experimentation, or successful experimentation could build confidence. The survey cannot distinguish those pathways.
A useful workplace response is therefore not to mandate more chatbot use. Employers could measure access before outcomes: who has a suitable tool, paid learning time, relevant examples and a route to challenge errors? A bounded pilot should compare people offered the same support and record uptake, completion, corrected outputs, review time and workload. Results should include staff who do not use the tool, rather than treating them as a problem to be removed from the denominator. That would test whether access changes opportunity in the organisation concerned.
Limits, people and what would change our assessment
PwC sells workforce and transformation services, giving it a commercial interest in organisational change. The weighting described covers age and gender within each country or region, but the published method does not make every subgroup nationally representative, nor does it remove non-response and self-report bias. Cross-country pooling can also hide very different labour protections, prices, technologies and training systems. The findings should not be converted into a forecast of redundancies or a personal probability of promotion.
For workers, the strongest practical signal is the reported disparity in opportunity: fewer than 40% of the engine-room group said they could access learning resources, compared with nearly 80% of front-runners. Our assessment would strengthen if longitudinal data showed that comparable workers given protected learning time and appropriate AI access later achieved better pay, mobility or job quality without higher burnout. It would weaken if administrative outcomes showed no benefit after correcting for occupation, seniority and prior advantage. Until then, the survey is a substantial global snapshot and a prompt for fairer evaluation—not a causal verdict on AI at work.