World Bank: Under 10% of MENAAP Jobs Face AI Automation, 13-20% Gain

Less than 10% of jobs in the Middle East, North Africa, Afghanistan and Pakistan face near-term automation by generative AI, while 13% to 20% could be made more productive by it. That is the labor-market core of the World Bank's October 2026 economic update for the region, "From Divide to Opportunity: AI, Jobs, and Growth", published on October 7 by the Office of the Chief Economist for MENAAP. Its data cutoff was September 24, 2026.

Three findings

1. Low automation risk comes from manual work. [Fact] Using the ILO and Poland's NASK exposure index, which scores nearly 30,000 tasks, automation risk covers 0% to 10% of jobs, highest in Jordan and the United Arab Emirates and lowest in Afghanistan and Pakistan. Iraq and Lebanon have the largest augmentation shares, driven by sales workers and machine operators. Commerce is the most augmentable sector; agriculture and construction are largely unexposed. The report adds that the low risk reflects a workforce concentrated in lower-productivity occupations.

2. Online freelancers already feel it. [Fact] Egypt and Pakistan are among the top five suppliers on major online work platforms. Between January 2022 and December 2024, in the quarter of skills most substitutable by AI, the value of jobs completed by MENAAP workers fell 9.8% and the number fell 5.6%, relative to the least substitutable quarter. Active workers fell 5.8%, and first-time entrants fell 12.1%. [Estimate] The entry decline is about twice the decline among established workers, a gap the authors read as AI squeezing newcomers first. Workers in these skills completed fewer jobs but earned more per job, and they added new skills 6% faster.

3. The Gulf captures a small share of the gains. [Fact] A labor cost equivalent built from Claude conversation data, the cost of paying people to do the tasks those chats handle, equals about 4.2% of GDP in high-income countries worldwide. In the region's high-income Gulf economies it equals 1.1%, and 0.4% in its middle-income economies. [Estimate] Gulf economies capture roughly a quarter of the share their global income peers do. Within countries, the first beneficiaries are software engineers, data analysts, lawyers and other high-paid professionals.

Why adoption stays shallow

[Fact] In Jordan, 24% of surveyed firms use some AI, and fewer than 5% use it for decision support or automated workflows. Across the region, 24% of firms invest in physical capital and 19% provide formal training, against 37% and 30% in other developing economies. Arabic, spoken by over 500 million people, makes up 0.7% of global URLs; Urdu, 0.03%. Models lose accuracy on Arabic dialects, the language of most call-center and service work. JobMatch.ps, a job-matching platform run by the Palestinian Ministry of Labor, lost accuracy because its first model was trained largely on European labor market data.

[Fact] Exposure is higher for urban and university-educated workers in all seven economies studied, and highest for tertiary graduates in Pakistan. In Egypt, Iran, Morocco, Pakistan and the West Bank and Gaza, exposed clerical and ICT jobs sit heavily in the public sector.

Which jobs this touches

Pakistan's Claude use centers on graphic design, content creation and translation, which the report links to its freelance platform work; Egypt's centers on software. That puts interpreters and translators and software developers closest to the evidence. Dialect gaps bear on customer service representatives. Commerce's augmentation potential applies to retail salespersons.

What to prepare

• Freelancers: widen the skill portfolio. Workers who added skills after ChatGPT were better placed to offset lost volume, and clients kept paying more for complex tasks.
• New entrants: start above the substitutable tier. Entry into the most substitutable skills fell twice as fast as overall participation.
• Bilingual and dialect skills have value. Models handle Modern Standard Arabic better than dialects, so local-language service work is harder to hand over.

Limits

The ILO-NASK scores come from Polish task data and assume the same task content in every country, and the World Bank notes lower-income countries would score lower still after adjustment. The labor cost equivalent uses Claude users, who may be more educated than the workforce, and holds wages and employment fixed. The online-work results cover one platform window ending December 2024.

Sources

• World Bank, "Middle East, North Africa, Afghanistan and Pakistan Economic Update, October 2026: From Divide to Opportunity - AI, Jobs, and Growth", October 7, 2026: https://documents.worldbank.org/curated/en/099116410062615883
• World Bank, charts companion to the same report, October 21, 2026: https://documents.worldbank.org/curated/en/099826110212613588

About this analysis

This article was written with AI assistance, from the full text of the report (Chapter 2 and the Executive Summary). The report is released under CC BY 3.0 IGO. The two figures marked [Estimate] are our own arithmetic from the report's numbers.