---
id: "lib-2026-06-15-ai-jobs-barometer-methodology"
type: "library"
title: "AI Jobs Barometer methodology"
lang: "EN"
org: "PwC"
source_type: "Article"
date: "2026-06-15"
week: "2026-W25"
url: "https://nosov.ai/library/lib-2026-06-15-ai-jobs-barometer-methodology/"
original_url: "https://www.pwc.com/gx/en/issues/artificial-intelligence/job-barometer/2026/2026-global-ai-jobs-barometer-methodology.pdf"
tier: "A"
tier_name: "Strong evidence"
tier_reason: "Technical appendix to PwC's flagship AI Jobs Barometer that documents the exposure indices, skill-change metric, firm filters and per-country sample sizes needed to judge its claims."
summary_basis: "full-text"
original_minutes: "20"
tags: ["labor market methods"]
---

# AI Jobs Barometer methodology

Tier A (Strong evidence): Technical appendix to PwC's flagship AI Jobs Barometer that documents the exposure indices, skill-change metric, firm filters and per-country sample sizes needed to judge its claims.

Summary based on the full text.

## In short

This 20-page appendix explains how PwC built the 2026 Global AI Jobs Barometer. It covers four parts: a refreshed AI Occupational Exposure index, a sector-level AI Industry Exposure index, the "net skill change" metric, and the filters applied to company accounts in the productivity analysis. It also lists posting counts by country. It is the document to read before quoting any Barometer figure.

## Key numbers

- 449.1 m: US job postings in the sample, out of about 1,089 m across 27 markets
- 144,524 to 52,274: ORBIS firms before and after filtering for the productivity and headcount analysis
- 10 x 52: AI applications scored against O*NET abilities to build the occupation exposure index
- 0.974: scaled AI exposure score for lawyers, the worked example

## What it reports

- The occupation exposure index follows Felten et al. in five steps: take each occupation's O*NET ability profile, score 10 AI applications against 52 abilities, combine these into ability exposure, weight by each ability's importance, then scale to 0-1. Lawyers score a raw 6.85, or 0.974 after scaling.
- The AI-ability matrix was refreshed using judgments from PwC's own Data and AI specialists, to reflect LLMs and multimodal systems.
- Sector exposure weights occupation scores by UK employment mix (SIC07). PwC states that higher exposure "does not imply job loss or automation".
- Net skill change, from Deming and Noray (2020), sums the absolute changes in each skill's share of postings. Skills gained and skills dropped both count as change.
- Firm sample: 144,524 ORBIS firms with at least $50m revenue. 41,096 were dropped for lacking a matching sector code and 47,780 for missing data, leaving 52,274 firms (38,961 in the wage table). Productivity was winsorized at the 1st and 99th percentiles.
- "Superstars" are the top 20% by revenue-per-employee growth within the most AI-exposed quartile.
- Postings by market include the US 449.1 m, UK 121.8 m, Chinese Mainland 86.5 m, Germany 82.6 m, France 79.4 m and India 30.2 m, out of about 1,089 m in total across 27 markets.

## How they got it

PwC UK's economics team built the indices from O*NET, UK employment data and expert scoring. It matched them to Lightcast postings and to ORBIS firm accounts for 2018 and 2024/25, using 2024 data where 2025 was missing. The firm metrics use pooled quartile ratios (total revenue divided by total employees), not averages of each firm's own growth.

## Limits

- Exposure scores rest partly on PwC specialists' judgment of what AI can do. A firm's exposure comes from its sector code, so actual AI use is never measured.
- Filters remove about two-thirds of firms and every firm that exited, which pushes the sample toward larger, surviving companies.
- US postings are about 41% of the total. Global figures lean American and English-speaking.

## Why it matters for companies

Read this before reusing Barometer numbers in a business case. It shows that "AI-exposed company" means "company in an AI-exposed sector", and that the wage and productivity comparisons are descriptive. The net-skill-change method is simple enough to run on your own job descriptions, as a way to see which roles' requirements are moving fastest.

Original: https://www.pwc.com/gx/en/issues/artificial-intelligence/job-barometer/2026/2026-global-ai-jobs-barometer-methodology.pdf