AI Resilience Report: Scoring Methodology

Overview

The AI Resilience Report scores approximately 1,600 U.S. occupations on AI resilience — the degree to which a career continues to offer sustained economic opportunity, employer demand, and meaningful human contribution as AI transforms work.

Each career receives a score from 0 to 100, built from three questions:

1. How much of this job's work still depends on humans? (Meaningful Human Contribution)
2. Will employers still be hiring for this role? (Long-term Employer Demand)
3. Will this career continue to pay well? (Sustained Economic Opportunity)

We answer these questions by combining multiple independent data sources — our own task-level AI research, AI exposure datasets from Anthropic and Microsoft, economic projections from academic researchers, and employment data from the Bureau of Labor Statistics. We use a multi-source ensemble because research finds that no single AI exposure measure reliably captures labor disruption on its own; combining multiple measures substantially improves predictive accuracy (Frank et al., 2025).

Our model aims to provide a reasonable picture of how and to what degree AI is transforming occupations, based on the latest research and data.

How Your Score Is Calculated

The Formula

Your AI resilience score is a weighted average of three sub-scores:

If a sub-score is unavailable for a given career, the remaining sub-scores are re-weighted proportionally so they still sum to 100%.

The Three Sub-Scores

Meaningful Human Contribution (MHC) — 40% of your score
How much of this career's work still depends on humans? We measure this using four AI exposure datasets (from Anthropic, Microsoft, Will Robots Take My Job, and our internal model). Each source estimates how much of the job AI can handle; we flip those scores so that higher values mean more human contribution. The four sources are combined as a weighted average, with newer data weighted more heavily (see Data Freshness Weighting below).

Long-term Employer Demand (LTE) — 30% of your score
Will employers keep hiring for this role? We use Bureau of Labor Statistics projections, combining annual job openings (60% weight) with projected employment growth over 2024–2034 (40% weight). Both figures are converted to percentiles.

Sustained Economic Opportunity (SEO) — 30% of your score
Will this career continue to pay well? We combine Althoff & Reichardt's wage bill projections (67% weight) — which estimate how total labor income flowing to each occupation will shift as AI advances — with Manning & Aguirre's adaptive capacity index (33% weight), which measures the ability of workers to navigate displacement.

Resilience Categories

The final score maps to six categories:

What Data Goes In

Our scoring model draws on seven data sources organized into two groups: four that measure AI exposure and three that measure economic demand.

AI Exposure Sources

These measure how much of an occupation's work AI can perform, each from a different angle:

CareerVillage AI Resilience Model v1.0 (our internal model)
Our proprietary task-level assessment. Rather than relying on broad statistical trends, we research individual tasks within each occupation to determine where AI is automating work versus augmenting human capability. Updated quarterly. Coverage: 100% of careers.

Anthropic Observed Exposure
Occupation-level exposure scores (0–1 scale) derived from real Claude AI usage patterns. Sourced from the Anthropic Economic Index. Coverage: 50.7% of careers (809 of 1,597).

Microsoft AI Applicability
AI usefulness scores based on Bing Copilot usage data from January–September 2024, measuring real-world AI application across roles. Coverage: 83.0% of careers (1,326 of 1,597).

Will Robots Take My Job (WRTMJ)
AI risk scores built from a trained regression model and user polling, inspired by Oxford University research on automation. Coverage: 90.8% of careers (1,450 of 1,597).

Economic Demand Sources

These measure whether an occupation will continue to offer jobs and earnings:

Althoff & Reichardt Wage Bill Projections
Forward-looking estimates of how AI will reshape total labor income (average wage × employment) across occupations, accounting for both generative AI and physical AI (robotics). Sourced from NBER Working Paper #33053. Coverage: 97.9% of careers (1,564 of 1,597).

Manning & Aguirre Adaptive Capacity
An occupation-level adaptive capacity index measuring worker characteristics relevant for navigating job transitions if displaced — including net liquid wealth, skill transferability, geographic density, and age. Coverage: 66.2% of careers (1,058 of 1,597). Sourced from Manning & Aguirre (2026).

Bureau of Labor Statistics Employment Projections
Annual job openings and projected employment growth (2024–2034) from the U.S. Bureau of Labor Statistics. Coverage: 98.4% of careers (1,571 of 1,597, excluding military occupations).

Source Update Cadences

How We Process the Data

Percentile Normalization

Our sources use different scoring scales, so direct comparison isn't possible. We convert every raw score to a 0–100 percentile rank based on its position across all occupations in our database.

Within the Meaningful Human Contribution sub-score, the four AI exposure sources are "flipped" (1 − score) so higher values represent human contribution rather than AI risk. The demand-side sources (Althoff & Reichardt, Manning & Aguirre, and BLS) are used as-is, since higher values already mean greater economic opportunity and resilience.

Data Freshness Weighting

AI capabilities evolve rapidly, so we apply a freshness discount to the four AI exposure sources in the Meaningful Human Contribution sub-score. Newer research receives higher weight, and data older than 24 months is excluded entirely:

Hierarchical Roll-Up

Our database covers 1,597 career entries organized into four levels using the Standard Occupational Classification (SOC) system: 23 Major Groups, 98 Minor Groups, 460 Broad Occupations, and 1,016 Detailed Occupations.

Not every source covers every occupation at the most granular level. When data is missing for a specific role, we use hierarchical "roll-up" logic:

• Detailed Occupations receive scores directly from available sources when possible.
• Broad Occupations average scores from their child Detailed Occupations.
• Minor Groups average scores from their child Broad Occupations.
• Major Groups average scores from their child Minor Groups.

For BLS data specifically, we use three methods: direct SOC-code matching (1,250 careers), aggregation from child occupations (286 careers), and inheritance from the nearest parent category (35 careers).

Confidence Indicator

Every score includes a confidence rating so you know how much data supports it. Here's how to interpret it:

About Our Internal Model

Our proprietary AI Resilience Model v1.0 is designed to provide a high-fidelity, task-level assessment of AI's impact on each occupation. Here's how it works.

The Six-Step Process

We run this process quarterly to ensure findings reflect the latest AI advancements:

1. Task Scoring: All 20,000+ O*NET career tasks are scored for automation likelihood using an LLM.
2. Task Selection: For each career, we select up to six core tasks, prioritizing those at the extremes of the automation spectrum — examples of both high automation potential and high human-centric stability.
3. Deep Research: We use OpenAI's deep_research API to search peer-reviewed journals, industry publications, and research institutions for documented evidence of AI impact on those specific tasks.
4. Dual Scoring: Research findings are synthesized through dual LLM prompts to generate independent scores for automation and augmentation.
5. Dynamic Weighting: We combine scores using dynamic weighting, where the influence of augmentation is reduced in roles with extremely high automation potential.
6. Percentile Mapping: Raw scores are mapped to percentiles (0–100) for consistent comparison with external datasets.

Limitations

While our methodology is designed to provide reliable, research-backed insights, we maintain transparency about its structural constraints:

• Relative, not absolute measurement. Percentile normalization measures how resilient one career is compared to others, not how much AI is affecting it in absolute terms. By design, the same share of careers will always appear in each tier regardless of whether AI is transforming a lot or a little of the labor market. This is useful for ranking careers against each other, but cannot capture economy-wide shifts over time.

• Source correlation. Our ensemble approach assumes independent signals, but the four AI exposure models likely share underlying assumptions about what makes work automatable (e.g., cognitive vs. manual, routine vs. non-routine). The three demand-side sources also share labor market fundamentals. This may reduce the diversification benefit.

• Temporal misalignment. Sources operate on different update cadences and reflect different moments in time. Freshness weighting adjusts for this in AI exposure data, but demand-side sources are inherently forward-looking and updated less frequently. We are blending snapshots from different points during a period of rapid technological change.

• U.S.-specific framing.Our methodology relies on SOC codes and BLS projections, limiting it to the U.S. labor market. AI's impact likely varies across economies with different labor structures, automation adoption rates, and regulatory environments.

• Within-occupation variance. SOC-level analysis provides occupational averages but cannot capture how AI exposure varies by industry, employer type, or geography. A financial analyst at a fintech startup faces different AI dynamics than one at a regional credit union; a truck driver on rural routes encounters different automation timelines than one in dense urban deliveries.

• Occupational mapping error.Our methodology depends on mapping between SOC codes, O*NET task libraries, and third-party occupation definitions. Emerging, hybrid, or highly specialized roles don't always align cleanly, and mapping ambiguity can propagate into scores.

References

1. Acemoglu, D., & Restrepo, P. (2019). Automation and New Tasks: How Technology Displaces and Reinstates Labor. Journal of Economic Perspectives, 33(2), 3-30. aeaweb.org
2. Althoff, L., & Reichardt, H. (2025). Task-Specific Technical Change and Comparative Advantage. hugoreichardt.com
3. Anthropic. (2026). Anthropic Economic Index: Observed Exposure Dataset. huggingface.co
4. Brynjolfsson, E., Li, D., & Raymond, L. (2022). Augmentation, Tasks, and Wages: How AI and Advances in Technology Affect the Labor Market. Stanford Digital Economy Lab. arxiv.org
5. Bureau of Labor Statistics. (2024). Employment Projections: 2024-2034 Occupational Outlook. U.S. Department of Labor. bls.gov
6. Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. arXiv preprint. arxiv.org
7. Frank, M. R., Ahn, Y.-Y., & Moro, E. (2025). AI exposure predicts unemployment risk: A new approach to technology-driven job loss. PNAS Nexus, 4(4), pgaf107. academic.oup.com
8. Frey, C. B., & Osborne, M. A. (2017). The Future of Employment: How Susceptible Are Jobs to Computerisation? Technological Forecasting and Social Change, 114, 254-280. oxfordmartin.ox.ac.uk
9. Manning, S., & Aguirre, T. (2026). How Adaptable Are American Workers to AI-Induced Job Displacement? NBER. nber.org
10. McKinsey Global Institute. (2017). A Future That Works: Automation, Employment, and Productivity. mckinsey.com
11. Nedelkoska, L., & Quintini, G. (2018). Automation, Skills Use and Training. OECD Social, Employment and Migration Working Papers, No. 202. oecd-ilibrary.org
12. Teigland, R., van der Zande, J., Teigland, K., & Siri, S. (2018). The Substitution of Labor: From Technological Feasibility to Other Factors Influencing Job Automation. Stockholm School of Economics Institute for Research. ssrn.com
13. Tomlinson, K., Jaffe, S., Wang, W., Counts, S., & Suri, S. (2025). Working with AI: Measuring the Applicability of Generative AI to Occupations. arxiv.org
14. Will Robots Take My Job. (2026). Occupational Automation Risk Analysis and User Sentiment Index. willrobotstakemyjob.com