Methodology

Measure AI displacement at the task level

displacement.ai treats job displacement as a transition problem, not a headline number. The model starts with tasks, then separates exposure, automation, augmentation, labor-market pressure, and transition fit.

Task exposure

How much of the work can be assisted or performed by current AI systems, regardless of whether a job disappears.

Automation potential

How much exposed work is repeatable, low-context, low-stakes, and likely to be reassigned to software.

Augmentation potential

How much AI increases worker productivity while humans stay accountable for goals, quality, or relationships.

Labor-market pressure

Wage context, hiring demand, role growth, offshoring pressure, and employer adoption speed.

Transition adjacency

Nearby roles that preserve skills, reduce wage loss, and can be reached with realistic training time.

Source confidence

Evidence strength, source recency, agreement across sources, and whether a score is reviewed or draft.

Draft rollup formula

The public score is a transparent index, not a layoff forecast. The current public model is Seed model v0.4 (seed-v0.4-2026-05), published as a directional seed model while the dataset is expanded and calibrated.

displacement pressure =
  0.35 * automation potential
  + 0.20 * near-term adoption
  + 0.15 * employment scale
  + 0.15 * wage vulnerability
  + 0.10 * routine work share
  - 0.10 * regulatory or safety constraint
  - 0.05 * transition feasibility

Evidence coverage

The current data pipeline imports Task Statements 31.0 (August 2026) from O*NET Resource Center. It currently covers 620 public occupations and 13085 matched task statements. The wage layer uses the official May 2025 national OEWS workbook for 602 of 620 public roles. Roles without a publishable current median are labeled as fallbacks rather than filled with invented values.

O*NET task import

Parsed and matched by SOC code to ground occupation pages in official task statements.

BLS wage import

May 2025 national estimates integrated and versioned separately from model scores.

Score model

Versioned separately from editorial content so score changes can be reviewed and tested.

How new research enters the model

The research library was refreshed through 9 September 2026 with Anthropic's conditional 2030 economic scenarios, OECD capability-gap methodology, Stanford payroll evidence, Anthropic labor-market and retraining studies, ILO and World Bank cross-country evidence, Microsoft Copilot usage research, and current OpenAI task-use studies. These sources improve interpretation, but none is represented as a numeric input to seed-v0.4-2026-05.

Published seed evidence

Only source IDs declared by the versioned model and attached to role records are labeled as used in current scores.

Calibration queue

New capability, usage, employment, and training evidence is tracked separately until mapping and validation rules are published.

Version gate

A score change requires a new model version, dated change log, regression checks, and refreshed role-level provenance.

See each source's authors, method, geography, limitation, and evidence role.

Explore the new 2030 scenario crosswalk for all 620 occupations.

Important caveat

Exposure is not displacement. A role can be highly exposed and still grow if AI removes bottlenecks, expands demand, or shifts workers into higher-value tasks. A role can also be moderately exposed and still experience layoffs if employers use AI mainly as a cost-reduction tool.

Scores are planning signals, not employment predictions.
Occupation-level scores should be localized with employer task inventories and regional wage data.
Policy and workforce decisions should use confidence intervals, not single deterministic scores.

How career paths are ranked

The calculator ranks transition paths with a consumer-first bias toward realistic moves. A high-paying target should not automatically outrank a nearby role if the training burden, credential gap, or skill distance is too large.

Near move

Preserves the most existing skills and can usually be tested with a small proof project or internal move.

Stretch move

Offers upside, but the user needs stronger evidence of skills before treating it as the main plan.

Reset move

May be attractive, but it is a larger career change and should not become the default recommendation.

Current model limits

Seed occupation scores are hand-modeled from public research and official occupational data.
Local wage and demand are strongest for US national references and weaker outside the United States.
Job-posting demand, licensing rules, and employer-specific workflows are not yet ingested automatically.
Recommendations should be validated against real postings, local programs, and a user's resume.