- Three open datasets: job postings for 12 countries, AI roles on company career pages, and duty labels, for the article Is AI creating entirely new jobs? I read 17,513 postings. It adds one duty.
- 17,513 job postings from the US, the UK and ten more European countries, October 2026: AI roles and ordinary roles from matching job families, with the employer type and flags computed from the text. The posting texts are left out.
- 5,856 AI roles from the career pages of 930 companies, July to October 2026, and the share of AI roles per 1,000 openings at 348 companies.
- 1,273 labelled postings: what each posting asks the hire to do, with the codebook and three agreement checks.
- Open under CC BY 4.0, as CSV. The summary files hold the numbers behind the tables and charts in the article.
What is in here
One section per dataset below, with its files and the line to cite it.
| Dataset | One row is | Files |
|---|---|---|
| AI Job Postings in 12 Countries, October 2026 | a job posting | 8 |
| AI Roles on Company Career Pages, July to October 2026 | an AI role | 4 |
| What AI Jobs Ask People to Do: Duty Labels for 1,273 Job Postings | a labelled posting | 7 |
AI Job Postings in 12 Countries, October 2026
17,513 job postings from the US, the UK and ten more European countries, collected in October 2026. About four in ten carry AI in the title. Most of the others are ordinary roles from matching job families, collected as a comparison. For every posting I keep the title, the employer, the employer type and a set of flags computed from the full text: does it mention AI, does it ask for Python, does it talk about adoption and change management. The posting texts themselves are left out.
postings.csv: 17,513 rows. Job postings, one row per posting.
employers.csv: 7,173 rows. Employers and their group.
summary_ai_mentions_in_comparison_roles.csv: 228 rows. AI mentions in ordinary roles by country and employer group.
summary_employer_mix.csv: 21 rows. Employer mix of AI roles and ordinary roles.
summary_ai_role_vs_comparison.csv: 38 rows. What an AI title adds against an ordinary role.
summary_ai_depth.csv: 4 rows. How many sentences of an ordinary posting mention AI.
summary_about_us.csv: 3 rows. AI mentions in the company description of ordinary postings.
summary_country_language.csv: 12 rows. AI mentions in ordinary roles by country and language.
Also on Kaggle · Hugging Face
Cite this dataset: Atamanova, M. (2026). AI Job Postings in 12 Countries, October 2026 [Data set]. https://mariascales.com/research/ai-jobs/data/
AI Roles on Company Career Pages, July to October 2026
5,856 job postings with AI, ML or LLM in the title from 930 companies, almost all from their own career pages, between July and October 2026. For each posting: title, employer, employer type, job family, level, location, work mode, when it appeared, whether it closed, and flags computed from the text. Two small tables follow 348 companies over time: how many AI roles they post per 1,000 openings, and how many of them post AI roles at all.
roles.csv: 5,856 rows. AI roles, one row per posting.
share_of_hiring.csv: 12 rows. AI roles per 1,000 open roles, fixed cohort.
breadth.csv: 6 rows. How many companies hire for AI roles.
summary_role_map.csv: 108 rows. AI roles by job family or engineering group and employer type.
Also on Kaggle · Hugging Face
Cite this dataset: Atamanova, M. (2026). AI Roles on Company Career Pages, July to October 2026 [Data set]. https://mariascales.com/research/ai-jobs/data/
What AI Jobs Ask People to Do: Duty Labels for 1,273 Job Postings
1,273 job postings, each labelled by what the posting says the hire will do: classic work of the job family, building AI, rolling it out, training people, setting AI strategy, measuring impact, picking vendors, AI governance, working on an AI product, working with clients, managing a team. 64% of the sample are roles with AI in the title, the rest are ordinary roles from matching job families. Word counts tell you a posting mentions AI. These labels tell you what the AI part of the job is.
labels.csv: 1,273 rows. Duty labels, one row per posting.
summary_duties_by_employer.csv: 24 rows. Duty shares by employer group.
label_check.csv: 150 rows. Blind second labelling of 150 postings.
label_audit.csv: 120 rows. Independent blind audit of 120 postings.
label_verify_v11.csv: 100 rows. Independent blind check of version 1.1 on 100 postings.
summary_duty_parts.csv: 8 rows. Parts of the duty to bring AI into a company, AI roles against ordinary roles.
codebook.md. Codebook used for labelling.
Also on Kaggle · Hugging Face
Cite this dataset: Atamanova, M. (2026). What AI Jobs Ask People to Do: Duty Labels for 1,273 Job Postings [Data set]. https://mariascales.com/research/ai-jobs/data/
Where these numbers are used
Is AI creating entirely new jobs? I read 17,513 postings. It adds one duty.
Cite this
Atamanova, M. (2026). AI Jobs by the Numbers [Data set]. mariascales.com. https://mariascales.com/research/ai-jobs/data/