Are the Best-Paid Jobs the Most AI-Exposed? 120 Countries

Key findings

Pay and AI exposure rise together. In 120 countries we ranked occupation groups by pay and by AI exposure. The median rank correlation is 0.70; in 95 countries it is 0.5 or higher and in only 3 is it negative (ILO ILOSTAT and national wage data, WorldJobsData scores).
Office-type jobs hold a quarter of the workers but 39% of the pay. In the median country managers, professionals, technicians and clerks are 25% of workers and about 39% of pay. In 119 of 120 countries their pay share is larger than their worker share.
Managers earn 2.15 times the national average in the typical country and score 5.5. Elementary jobs earn 0.65 times it and score 2.0.
Clerical work is the exception. It has the highest exposure (8.5) but pays only 1.07 times the average, so exposure alone does not follow pay.
Exposure is not job loss. A high score says AI can help with many tasks in the job. It does not say the job disappears or that pay falls.

The question

Many people expect new technology to hit low-paid work first. We checked whether that holds for AI by comparing two things for every occupation group in each country: how well it pays and how exposed it is to AI. AI exposure is our own score from 0 to 10 for how much of a job's tasks AI could do or help with. Higher means more of the job is writing, calculating or handling information. The scores are model-estimated, and are not measurements.

Occupation groups follow the International Standard Classification of Occupations (ISCO-08): managers, professionals, technicians, clerical support workers, service and sales workers, skilled farm workers, craft and trades workers, plant and machine operators, and elementary occupations (basic, often manual jobs). For each country we use pay from the ILO or the national statistics office, and employment from the same sources. We use the 120 countries that report all nine groups and have pay for at least seven of them.

The pay ladder and the exposure ladder line up

Typical pay of each occupation group relative to the national average (1.0 = the country's employment-weighted average pay). Median across countries with data. Bar colour shows AI exposure: red 5 or higher, amber 3.5 to 5, green below 3.5.

Occupation group	Typical pay vs national average	AI exposure	Countries with data
Managers	2.15x	5.48	120
Professionals	1.51x	6.56	120
Technicians	1.27x	5.63	118
Clerical support	1.07x	8.50	120
Plant and machine operators	0.98x	2.99	119
Craft and trades	0.93x	2.51	119
Service and sales	0.82x	3.50	120
Farming	0.72x	3.00	104
Elementary occupations	0.65x	1.99	120

Each dot is an occupation group. The further right, the better paid relative to the national average. The higher, the more exposed to AI.

The ladder is clear. Managers sit at the top: about 2.1 times the average pay and an exposure score of 5.5. Professionals come next, at about 1.5 times and a score of 6.6, the second highest. At the bottom, elementary jobs pay about 0.65 times the average and score 2.0, and farm work pays 0.72 times and scores 3.0.

Clerical support is the exception. It has the highest exposure of all nine groups (8.5) but sits in the middle of the pay ladder at 1.07 times the average. This matches what our other analyses show: clerical work is where AI tools can already do the most, yet clerks are not among the best paid. See why clerical work scores 8.5.

How consistent is it across countries?

Number of countries by the strength of the link between pay and AI exposure across occupation groups. A rank correlation of 1 means pay and exposure rank the groups in exactly the same order; 0 means no relationship.

The typical country has a rank correlation of 0.70. 59 of the 120 countries are above 0.7, and only 3 are below zero. The pattern is not driven by a few rich countries: it shows up in poor, middle-income and rich countries alike.

Where the link is weakest

Country	Pay-exposure link	Groups compared
Sudan	-0.55	9
Tonga	-0.43	9
Liberia	-0.14	9
Iceland	0.07	8
Ghana	0.09	9
Laos	0.10	9

The weakest cases include small economies and countries with thin pay data. With only seven to nine groups per country, a single odd group can change the ranking. We treat them as places to check, not as evidence against the pattern.

Since the best-paid groups are also the most exposed, the share of pay going to exposed work is larger than the share of workers doing it. We approximate each group's pay bill as its number of workers times its typical pay. Then we compare office-type jobs (the four groups from managers to clerical support) with everyone else.

Country	Workers in office-type jobs	Share of pay going to them	Manager pay vs elementary pay	Pay-exposure link
United Kingdom	65%	77%	2.7x	0.46
Germany	61%	73%	3.8x	0.70
United States	55%	69%	3.1x	0.44
Australia	53%	62%	2.0x	0.32
South Korea	45%	57%	5.0x	0.72
Brazil	35%	55%	4.9x	0.67
Mexico	28%	40%	3.1x	0.79
India	14%	29%	4.4x	0.76
Indonesia	15%	26%	3.8x	0.70
Nigeria	8%	11%	1.9x	0.48

In the median country office-type jobs are 25% of workers but about 39% of pay. In Germany they are 61% of workers and 73% of pay. In India they are 14% of workers but 29% of pay: a minority of workers earns a large share of pay, and it is the part of the workforce AI tools reach most. The gap between the two shares is widest in Qatar (36% of workers, 76% of pay), Kenya (17% and 57%), Botswana (25% and 64%) and United Arab Emirates (40% and 77%).

What this means

If AI tools mostly change the tasks in the best-paid jobs, the effects will land on the part of the workforce that earns the most. That can mean different things: for some workers AI tools raise productivity and pay, for others they reduce demand for junior roles that used to lead to those jobs. Our data cannot tell which will happen. What it does show is that the most-exposed work is not low-paid work in most countries. Planning that assumes AI hits the bottom of the pay ladder first does not match these numbers.

It also means that a low exposure score is no comfort for the lowest paid. Farm and elementary workers score low because AI tools reach less of their work today, not because they are secure. Their pay is lowest in the ladder, and they have the least cushion if anything else changes.

Limits to keep in mind

Pay is approximate. We use each group's typical (median, or mean in 17 of the 120 countries) annual pay. Pay within a group varies a lot, and the pay bill is workers times typical pay, not a payroll total.
Sources differ. Pay comes from ILO ILOSTAT for most countries and from national sources for some, such as the US BLS data. We use ratios inside each country, so currency differences do not matter.
Correlation across nine groups. Nine points per country is few. We use rank correlations and look at the whole pattern across 120 countries, not any single country.
Scores are model-estimated. They say how much of a job could be done with AI, not how much will be.

See pay and AI exposure for any country

Open a country page to see every occupation group's pay, workers and AI exposure score.

Open the explorer →

Was this analysis useful?

Methodology and sources. Occupation-group pay is each country's median annual pay by ISCO-08 major group (median_annual_wage_usd in the country files, from ILO ILOSTAT or national statistics offices; mean pay in 17 countries), with employment from the same sources. We combine detailed occupations into the nine major groups, weighting by employment, and use the 120 countries that report all nine groups and have pay for at least seven. Pay relative to the national average divides each group's pay by the country's employment-weighted average pay; the table shows the median across countries. Rank correlation is Spearman's rho across a country's groups. The pay bill is workers times typical pay and is an approximation. Armed forces are excluded. Calculations are ours.

AI exposure scores are model-estimated: an AI language model (Claude Opus 4.7) scored each ISCO-08 occupation group on 28 May 2026, guided by published research. They are informed judgements, not measurements, and not forecasts of job losses. Read the full methodology and how the scores were checked.

Frequently asked questions

Are the best-paid jobs the most exposed to AI?

By our scores, usually yes. In 120 countries the median rank correlation between an occupation group's pay and its AI exposure is 0.70. In 95 of 120 it is at least 0.5, and in only 3 is it negative.

Which occupation group is the exception?

Clerical support. It has the highest AI exposure score (8.5 out of 10) but pays about the national average (1.07 times), well below managers (2.15 times) and professionals (1.51 times).

How much of a country's pay goes to office-type jobs?

In the median country office-type jobs (managers, professionals, technicians, clerks) are 25% of workers but receive about 39% of total pay. In 119 of 120 countries their share of pay exceeds their share of workers.

Does high exposure mean high-paid workers will lose their jobs?

No. The exposure score says how much of a job's tasks AI could help with. For many high-paid roles that means changed tasks or higher productivity, not job loss. We do not forecast pay or job losses.

Sources

ILO (2026) - ILOSTAT, earnings and employment by occupation (ISCO-08), latest available year per country (CC BY 4.0)
US Bureau of Labor Statistics (2026) - Occupational Outlook Handbook, May 2025 wages, used for the United States
National statistics offices - earnings by occupation where used in our country files
WorldJobsData (2026) - Occupation-level AI exposure scores, methodology

Spotted an error or know a better source? Report it and we will check it against the source data.

Are the Best-Paid Jobs the Most Exposed to AI? Data From 120 Countries