Scoring Methodology

We designed this methodology to navigate the uncertainty of the AI era by using a dual-lens approach. We wanted to know - how are jobs being impacted by AI, and is that impact amplified or mitigated by broader market factors?

Scoring Process Overview

Starting from the official US Bureau of Labor Statistics list and labor data, we score each role on 12 AI-resilience attributes and group those into 3 pillars. The pillars are then used to bucket careers into an AI Impact category. Finally, we add labor market data and generate a 0–100 score and a universal leaderboard.

1
Acquire BLS Data
Career list + labor projections (2024–2034)
↓
2
Score Attributes
12 AI-resilience attributes (A1–A12)
↓
3
Group into 3 Pillars
Exposure · Necessity · Elasticity
↓
4
AI Impact Category
Sort the role into 1 of 4 transition archetypes.
↓ Output ↓
Output · Ranking
5
Final Score (0–100)
Rank every role on one 0–100 leaderboard.
Inputs:AI Impact CategoryBLS labor data
Jump to Section
1 · Labor Market Data
2 · Score Attributes
3 · Core Pillars
4 · Categories
5 · Final Score
Sources

1 · Labor Market Data

The career universe — and the labor input for Step 5

We start from the official career list and labor market data from the Bureau of Labor Statistics (BLS) - every role we score comes from this list, and we later combine the scores with this data to identify careers that are both hard to automate and actually growing:

  • Projected Job Openings (2024-2034): Total number of job openings expected from BLS Employment Projections, including new jobs and replacements. This captures both growth and churn, showing where actual hiring demand will be.
  • Projected Growth Rate (2024-2034): We use numeric employment change data from BLS Employment Projections as the primary source — this distinguishes fast-growing fields from those merely replacing retirees. For occupations not listed separately in BLS projections (e.g. specialty subcodes), we fall back to the O*NET qualitative growth category (Decline, Average, Much faster than average, etc.).

A high AI-proof score paired with strong job growth signals a career that's both defensible and in-demand. A resilient job in decline tells a different story: skills are safe, but opportunity may be limited. A low resilience job with strong growth often means the BLS numbers haven't fully accounted for recent technology advancements.

(Note: We recalculate scores as BLS data is updated annually)

2 · Score Attributes

3 · Core Pillars

Our methodology for calculating AI resilience is inspired by OpenAI's AI Jobs Transition Framework. We built upon its rubric-based approach, expanding it to capture the nuance of human necessity and market elasticity alongside raw technical exposure.

The 12 attributes are grouped into three underlying pillars:

  • Exposure Pillar: How technically exposed are the core tasks?
    Formula: (A11 + A9 + (6-A3) + (6-A5) + (6-A7)) / 5
  • Necessity Pillar: How badly does the job require a physical human or legal/trust relationship?
    Formula: (A1×1.5 + A4×1.5 + A2×1 + A8×1 + A6×0.7) / 5.7
  • Elasticity Pillar: Will making the job's core output cheaper drive massive new market demand?
    Formula: (A12 + A10) / 2

4 · AI Impact Categories

We used the 3 core pillars to sort jobs into 4 distinct categories, an approach adapted from OpenAI's AI Jobs Transition Framework. These categories define the nature of how AI will impact the role.

High Exposure, High ElasticityGrowing with AI
AI drives productivity and market expansion. These roles benefit directly from AI handling low-value work, freeing the human to produce vastly more high-value output.
High Exposure, High NecessityChanging with AI
AI automates the core tasks, but a human remains strictly necessary for legal, physical, or trust reasons. The job changes from doing the work to managing the AI doing the work.
Low ExposureLess Changed by AI
AI cannot automate the core physical or relationship tasks. These roles are largely insulated from immediate disruption due to hard physical or social constraints.
High Exposure, Low Elasticity, Low NecessityAt Risk from AI
AI automates the core output, and making it cheaper does not unlock massive new market demand. These roles are highly vulnerable to near-term displacement.

5 · Final Score

To rank every occupation on a single universal leaderboard, we combine the AI impact scores with BLS employment data into a 0.0–1.0 composite score (displayed as 0–100).

Final Score = Necessity(35%) + Elasticity(25%) - Exposure(20%) + Growth(15%) + Openings(5%)

BLS Growth and Openings are log-transformed and min-max scaled to prevent massive job counts from skewing the underlying AI resilience metrics.

Interpret with Context

A "Strong" score does not mean a career is bulletproof, nor does a "Risky" score mean it will disappear overnight. These tiers measure relative resilience and stability based on current technology and labor trends. A role can be shifting quickly due to AI adoption, and see increased hiring due to market demand.

Sources

  • The AI Jobs Transition Framework — OpenAI (2025). Basis for the rubric-based exposure scoring and the four transition categories (Grow with AI, Will Evolve, Less Immediate Change, High Automation Risk).
  • O*NET Online — US Department of Labor. Occupation definitions, wage data, education requirements, job descriptions.
  • BLS Employment Projections 2024–2034 — US Bureau of Labor Statistics. Projected job openings and growth rates by occupation.
  • BLS Occupational Employment and Wage Statistics (OEWS) — US Bureau of Labor Statistics. Median wages and employment by occupation.
  • Anthropic Economic Index — Anthropic. Real-world Claude usage patterns mapped to O*NET tasks; basis for the observed technical exposure score (A11) and the task-level automation/augmentation data shown on each career page.
  • altPath — Nonprofit platform providing supplemental career titles data.

Open Data & Code

The full dataset and scoring scripts are available at our GitHub repository under a CC BY 4.0 license. You are free to use, adapt, and republish with attribution.

↓ Download Dataset (CSV)View on GitHub

Author: Yu Chen · February 2026, updated at least annually