AI Exposure Map · International

Same methodology. Five labor markets. Different outcomes.

I scored over 500 occupations across five countries using the same JPE methodology. What I found: AI exposure is remarkably consistent. How countries absorb it is not.

The core finding: A bookkeeper scores 8/10 in all five countries. The technology does not care about borders. But the speed at which that score translates into actual job change varies dramatically. Germany has Kurzarbeit and Betriebsräte. The Netherlands has collective agreements and a 48% part-time rate. The US has at-will employment and minimal buffers. The UK sits between the US and EU models — flexible employment with moderate protections. France has the 35-hour week, CDI/CDD contracts and CNIL oversight. Same score. Different trajectories.

The numbers side by side

Metric🇳🇱 Netherlands🇺🇸 United States🇩🇪 Germany🇬🇧 United Kingdom🇫🇷 France
Occupations scored1031069810686
Jobs covered5.8M90.7M29.1M17.7M15.9M
Weighted avg. JPE4.94.854.94.8
Score ≥ 7 (high exposure)3425273022
Score ≤ 3 (low exposure)2628262524
ClassificationCBS BRC 2014BLS SOC 2018KldB 2010SOC 2020PCS 2020
Employment sourceCBS StatlineBLS OEWS 2024Destatis 2024ONS LFS 2024INSEE 2024
Salary sourceCBS / CAOBLS OEWSEntgeltatlas BAONS ASHE 2024INSEE DADS
Forecast dampening0.70.70.60.680.6
AI pressure (score 9+)−8−5*−6−8−6

* US uses a different projection method (direct BLS percentages, factor 1.0) — the lower pressure value compensates for this. Effective impact is comparable.

Same score, different speed

AI exposure is a property of the work, not the country. An accountant processes the same type of structured data in Rotterdam, Houston, Munich, London and Paris. The JPE score captures that. What differs is how fast it translates into actual change on the ground.

In the US, an employer can restructure a team next quarter. In Germany, that same decision passes through the Betriebsrat, gets mitigated by Kurzarbeit, and takes two years. Neither approach is inherently better. The American model is faster. The German model is more predictable. The Dutch model sits in between — collective agreements provide structure, but the labor market is more flexible than Germany and less volatile than the US. The UK post-Brexit model resembles the US in speed but retains more worker protections. Notice periods and unfair dismissal law slow things down compared to at-will employment, but far less than German co-determination.

Why Germany absorbs slower

The model uses a dampening factor of 0.6 for Germany versus 0.7 for NL and the US. Lower dampening means trends continue with less momentum. Why?

Kurzarbeit

Companies reduce hours instead of headcount. The government pays the difference. This keeps people employed through transitions that would cause layoffs in the US.

Betriebsräte

Co-determination is mandatory from 5 employees. Any significant technology change must be negotiated. This slows AI adoption in a way that has no equivalent in the Anglo-Saxon world.

Kündigungsschutz

Dismissal protection is substantially stronger than NL or US. Role changes happen gradually, through attrition and retraining, not sudden restructuring.

Dual vocational training

Germany has 1.3 million apprentices. The Ausbildung system creates deep occupational identity and high switching costs. A Mechatroniker does not easily become a data analyst.

The net effect: an occupation with AI score 8 will change faster in the US, but more predictably in Germany. The Netherlands sits in between. Dutch CAO agreements provide some structure, but the labor market is considerably more flexible than Germany's.

AI pressure: calibrated per market

The AI pressure table — the downward force applied to high-scoring occupations — is different per country. Not because AI is stronger in one place, but because the transmission mechanism is different.

AI score🇳🇱 NL🇺🇸 US🇩🇪 DE🇬🇧 UK🇫🇷 FRRationale
≥ 9−8−5−6−8−6Near-complete task automation. NL and UK have highest pressure; US uses different projection method.
7–8−5−3−4−5−4Core tasks exposed. Strong pressure, moderated by country buffers.
5–6−2−1−2−2−2Partial exposure. AI assists but does not replace.
3–400000Limited exposure. Neutral effect on employment.
0–2+1+1+1+1+1AI creates demand (e.g. more construction via AI-designed buildings).

Important: the US numbers look lower because the US model uses a direct projection method (factor 1.0) rather than index-based projection (factor 2.5). When converted to the same scale, effective impacts are comparable. The direction is the same across all five countries.

Structural differences that matter

Factor🇳🇱🇺🇸🇩🇪🇬🇧🇫🇷
Dismissal protectionModerate (UWV permit)Weak (at-will)Strong (Betriebsrat + law)Moderate (unfair dismissal after 2yr)Strong (CDI protection, Code du Travail)
Co-determinationCAO collective agreementsMinimalBetriebsrat (right from 5 staff, ~40% of firms)Moderate (union recognition)CSE (right from 11 staff) + CGT/CFDT
Transition safety netUWV + retrainingLimitedKurzarbeit (cyclical) + Qualifizierungsgeld (structural)Universal Credit + retrainingFrance Travail + CPF training credits
Part-time rate43-48% (highest in EU)~17%~28%~25%~18%
AI regulationEU AI ActLight (AI Act 2026)EU AI Act + KI-StrategiePro-innovation (AI Safety Institute)EU AI Act + CNIL data oversight
Labor shortage sectorsHealthcare, IT, constructionVariesPflege, IT, HandwerkNHS, IT, social careHealthcare, IT, hospitality
Economy structureServices-heavyServices + techIndustrial + MittelstandServices + finance (City)Services + public sector + luxury

Germany's industrial base means a higher share of occupations in production, trades and engineering — sectors that are inherently less exposed to AI. The Netherlands is more services-heavy, which pushes the overall exposure profile up. France combines a large public sector (with near-permanent civil service contracts) and strong CDI protections, making it structurally similar to Germany in absorption speed but for different reasons. The UK and the US have less friction when change does come, but the UK's NHS acts as a massive anchor.

The developer paradox — in all four

Software developers score 9/10 in every country. They also show the strongest employment growth in every country. NL: +80% (2015–2025). US: +17% (BLS 2023–2033). DE: +35% (IAB 2015–2025). UK: +12% (ONS 2019–2024).

This is not a bug in the model. Demand for software grows faster than AI replaces the people who write it. GitHub Copilot makes a developer 2x more productive. But the world needs 3x more software. The net effect is growth, not decline.

The same pattern holds for data scientists (score 9, growing in all four), IT consultants (score 8, growing), and healthcare managers (score 6, strong growth due to demographics). High AI exposure combined with rising demand equals growth, not replacement. The score measures exposure, not destiny.

What this means for you

If you work in the Netherlands, your occupation will change at roughly the average pace of AI adoption. Strong collective agreements provide some buffer, but the services-heavy economy means broad exposure.

If you work in the US, changes will come faster. At-will employment means faster restructuring. But also faster creation of new roles. The American labor market is volatile in both directions.

If you work in Germany, Kurzarbeit and Betriebsräte give you more time to adapt. Use it. The transition will be slower but it will happen. The dual vocational system is an asset — but only if Ausbildung programs integrate AI skills. That is happening, but not fast enough.

If you work in the UK, you have a flexible labour market with a strong services economy — particularly finance (City of London) and the NHS. The pro-innovation AI approach means fewer regulatory buffers than the EU but more than the US. The NHS workforce is a unique factor: healthcare demand is rising while budgets are tight, making AI adoption both an opportunity and a necessity.

Across all four: the occupations that thrive are the ones where people use AI as a tool, not the ones where people compete with it.

Methodology note: All five maps use the same Janssen Practical Exposure (JPE) methodology, making scores directly comparable across countries. The forecast model uses country-specific parameters to account for structural labor market differences. Full methodology: NL · US · DE · UK. All data CC BY 4.0.

Important limitations: JPE scores are LLM-estimated, not reproducible science. Dampening factors and AI pressure values are informed judgments, not empirically measured. Employment data precision varies: NL uses observed CBS data; UK, DE, and FR figures are estimates with synthetic history. The US uses a different projection formula than other countries. See the full limitations and known constraints section for details on coverage, classification crosswalks, salary comparability, and what this research does not cover.

🇳🇱 Netherlands map🇺🇸 US map🇩🇪 Germany map🇬🇧 UK map← Back to hub