AI Job Postings Are 75-85% STEM in All 10 Countries Studied — The New Hiring Divide

Between 75% and 85%. That is how much of AI-related hiring landed in STEM occupations — in every one of the ten countries a new study examined, from the United States to Indonesia. [Fact] Not a third. Not "spreading evenly across the economy." Three quarters to four fifths, packed into a single occupational cluster.

If you have heard that AI skills are becoming as universal as spreadsheet skills, the posting data says otherwise. And whether that is good news or a warning depends entirely on which side of the divide you work.

What the researchers measured

Rafiazka Hilman and Júlia Koltai — researchers affiliated with the MTA-TK digital social science research group in Budapest, the University of Amsterdam, and ELTE — analyzed job postings collected from LinkedIn, Indeed, and Glassdoor between December 2025 and May 2026. [Fact] The sample covers ten labor markets on five continents: the United States, the United Kingdom, Germany, the Netherlands, Brazil, India, Indonesia, China, Singapore, and South Africa.

The pipeline matters, because job-posting studies live or die on extraction quality. The team used a two-stage approach: rule-based natural language processing to pull skill mentions out of raw posting text, followed by zero-shot classification with an open-weight LLM (GPT-OSS) to catch what the rules missed. [Fact] They then built bipartite networks linking occupations to skills, projected them into occupation-to-occupation similarity maps, and ran Louvain and Leiden community detection to see which jobs cluster together — with Jaccard similarity measuring how much any two occupations' skill demands overlap.

In plain terms: they mapped which jobs are being asked for the same things, and watched how AI demand redraws that map.

One core, four skills

The concentration finding is stark. Across all ten countries, roughly 75% to 85% of AI-related vacancies sat in STEM occupations, with only modest variation between nations. [Fact] The skills employers asked for form what the authors call a compact core of data competencies: Python, SQL, machine learning, and data analysis dominated the rankings in every sector and every country. [Fact]

The occupations at the center of that core are familiar: data scientists, software developers, statisticians, and computer network architects led AI-intensive hiring, with business roles like market research analysts and operations managers appearing at the edges — positions that increasingly borrow from the technical skill set.

Four skills. A handful of occupations. Ten very different economies, one pattern.

The other 95%: where AI hiring is not reaching

Now the counterintuitive part. Four broad sectors — health and education, services, production and logistics, and public service and care — each accounted for less than 5% of AI postings. [Fact]

Read that carefully, because it cuts against the dominant narrative. If you are a registered nurse or a schoolteacher, employers in your field are, as of mid-2026, mostly not asking you for Python or machine learning. The popular prediction that every job will soon require AI skills finds no support in what employers actually write in postings today. [Claim]

That is not the same as safety. More on that below.

The entry-level twist

The study's most uncomfortable finding is about seniority. When the researchers applied stricter thresholds to their networks, entry-level positions retained the strongest ties to AI skill demand, while director- and executive-level positions detached from the AI skill core. [Fact]

The authors' interpretation: AI competence is functioning as a precondition for occupational access, not as a mid-career advancement tool. [Claim] In other words, the gate is at the entrance. A 22-year-old applying for a first analyst job faces AI skill requirements that the department head interviewing them never had to meet — and, at current posting patterns, still doesn't.

For career starters, "I'll pick up AI skills once I'm in" is becoming a riskier plan. The demands are being loaded onto exactly the rung of the ladder where you have the least leverage.

Convergence and divergence — both at once

So is AI homogenizing the labor market or splitting it? The study's answer: both, in a specific and unequal way. Within the STEM core, occupations are converging — data scientists, developers, and statisticians are increasingly asked for the same compact skill set, making their profiles more interchangeable. [Fact] Between that core and everyone else, the distance is growing. The authors describe a bifurcated pattern: convergence within an already advantaged, AI-exposed core, and divergence between that core and the rest of the occupational structure. [Fact]

And here is the finding that elevates this above another US-centric report: the pattern held across the Global North and the Global South alike. [Fact] Brazil, India, Indonesia, and South Africa showed the same core-versus-rest structure as the US, Germany, and Singapore. This does not look like a development stage that late-industrializing economies will simply pass through — it appears to be a structural feature of how AI enters labor markets. [Claim]

What this study cannot tell you

Honest limits, because they change how you should use these numbers.

First, job postings measure demand for building and deploying AI, not for using it. A teacher who runs lesson plans through ChatGPT every morning never shows up in this data, because no posting mentions it. Low AI penetration in postings does not mean low AI impact on the job — task-level automation arrives without a job ad announcing it. [Claim]

Second, the window is six months (December 2025 to May 2026) and the sources are three English-interface platforms. LinkedIn, Indeed, and Glassdoor over-represent white-collar, urban, formal-sector hiring — especially in Brazil, India, Indonesia, and South Africa, where informal employment is large. The true economy-wide concentration could differ by several percentage points, though the cross-country consistency suggests the core finding is robust. [Estimate]

Third, the sample includes no Japan, no South Korea, no Middle East. Whether the pattern holds in East Asian markets with different hiring-platform ecosystems remains untested.

What to do with this

If you work inside the STEM core: convergence means your skill profile increasingly overlaps with adjacent roles. Mobility between data and software jobs is rising — and so is the pool of people who can plausibly apply for yours. As the four core skills become table stakes, differentiation shifts to domain knowledge: the data scientist who understands hospital operations or supply chains beats the one who only knows the toolkit.

If you work in the under-5% sectors: the absence of AI requirements in today's postings is breathing room, not immunity. Use it. Our occupation pages track task-level automation exposure — which reaches jobs long before hiring requirements do — for registered nurses, operations managers, and a thousand other roles.

If you are early-career: the study's clearest practical message is aimed at you. AI skill demands concentrate at exactly your level. The four-skill core — Python, SQL, machine learning fundamentals, data analysis — is finite, learnable, and currently the price of admission to a widening set of first jobs. Based on these posting patterns, that price is more likely to spread to adjacent entry-level roles than to disappear. [Estimate]

The labor market is not being homogenized by AI. It is being sorted — and the sorting starts at the front door.

---

AI-assisted analysis: This article was researched and written with AI assistance, based on the preprint cited below. All statistics were verified against the source paper; statements marked [Claim] or [Estimate] are analytical judgments, not findings reported by the authors.

Sources

• Hilman, R., & Koltai, J. (2026). Occupational Convergence or Divergence? Mapping Labor Market Structural Shifts Driven by AI Penetration. arXiv:2607.28798. https://arxiv.org/abs/2607.28798