How the pieces join

The unit is the occupation BLS counts in its Occupational Employment and Wage Statistics (OEWS): a 2018 SOC detailed code, or one of twelve OEWS codes that combine several. Each of O*NET's roughly one thousand occupation profiles (8-digit O*NET-SOC codes) rolls up to one of them by its first six digits, the same way the Predicted exposure app does it. Where several O*NET profiles share a code, their ratings are averaged with equal weight, because O*NET publishes no employment for them. Groups in the treemap are coloured by the figure BLS publishes for the group (median wage, projected change) where there is one, and otherwise by an average of their occupations weighted by May 2025 employment.

BLS state files publish major groups and detailed occupations only, not minor or broad groups, and they leave out cells where the estimate is unreliable or would reveal an employer. Those cells are hatched as "not published"; nothing here fills them in. The state averages of LLM exposure and Claude use weight each published occupation by its state employment; their coverage is shown in the tooltip.

The map by meaning embeds each item's name and short description with Qwen3-Embedding-8B, projects the vectors to two dimensions with UMAP (seeded), cuts the map into topics with k-means, and names each topic with a local language model reading its most central items. Distances on the map are approximate; the nearest-neighbour lists use the full vectors.

Data

1. U.S. Bureau of Labor Statistics, 2018 Standard Occupational Classification System, structure (soc_structure_2018.xlsx), 2017. bls.gov/soc/2018
2. U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025: national and state estimates (oesm25nat.zip, oesm25st.zip), 2026. bls.gov/oes/tables.htm
3. U.S. Bureau of Labor Statistics, Employment Projections program, Occupational projections, 2025–35, and worker characteristics, 2025 (occupation.xlsx, Table 1.2), 2026. bls.gov/emp
4. National Center for O*NET Development, for the U.S. Department of Labor, Employment and Training Administration, O*NET 31.0 Database (occupation data, Job Zones, education, interests, essential and transferable skills, abilities, knowledge, work activities, work context, software skills, task statements, tasks to DWAs, GWA–IWA–DWA reference, emerging tasks, alternate titles, sample of reported titles), August 2026. CC BY 4.0. onetcenter.org/database.html
5. National Center for O*NET Development, Bright Outlook occupations, O*NET OnLine, updated October 6, 2026 (based on the 2025–35 projections). onetonline.org/find/bright
6. European Commission, ESCO (European Skills, Competences, Qualifications and Occupations) v1.2.1, English labels, read from the ESCO web service API, 2026. © European Union, reused under CC BY 4.0 (Commission Decision 2011/833/EU). esco.ec.europa.eu
7. ESCO Secretariat with the U.S. Department of Labor, Employment and Training Administration, ESCO–O*NET crosswalk (occupations), version 1, September 2022 (O*NET-SOC 2019 to ESCO v1.1.0), and Crosswalk between ESCO and O*NET: technical report, European Commission, 2022. esco.ec.europa.eu/en/use-esco/other-crosswalks
8. Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock, "GPTs are GPTs: Labor market impact potential of LLMs", Science 384 (6702), 2024; occupation-level GPT-4 β from github.com/openai/GPTs-are-GPTs, joined in the Predicted exposure app.
9. Anthropic, Anthropic Economic Index, release 2026-06-26, Claude.ai usage by O*NET occupation, via the Predicted exposure app. huggingface.co/datasets/Anthropic/EconomicIndex
10. Mike Bostock, us-atlas 3.0.1 (states-albers-10m.json, from the U.S. Census Bureau's 2017 cartographic boundary files), with d3 7.9.0 and topojson-client 3.1.0. github.com/topojson/us-atlas
11. Qwen team, Alibaba, Qwen3-Embedding-8B, 2025, and Leland McInnes, John Healy and James Melville, "UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction", arXiv:1802.03426, 2018. Topic names by Qwen 3.6 run locally through Ollama. huggingface.co/Qwen/Qwen3-Embedding-8B

Things to keep in mind

• OEWS counts wage and salary jobs, not people, and leaves out the self-employed. Wages are nominal and not adjusted for living costs between states.
• The projections are BLS's 2025–35 set, the current one; they are projections, not forecasts of any one company or place.
• O*NET ratings come from surveys of workers and occupation analysts at different dates; each profile carries its own date. "All other" codes usually have no ratings.
• The ESCO–O*NET crosswalk was made against ESCO v1.1.0; links to ESCO occupations that changed or were retired by v1.2.1 drop out.
• The O*NET 31.0 download list has no Tools Used file, only Software Skills (with hot technology and in-demand flags), so the technology view covers software, not physical tools.