Euphemios
Back to insights

Report

AI and work: blue-collar and white-collar

Claude conversations cluster around white-collar tasks, while business surveys and public filings show a different view of AI in labor-heavy industries.

Euphemios11 min read

Executive summary

White-collar tasks accounted for 93.03% of the published May 2026 U.S. Claude usage measure, while white-collar occupations accounted for 56.04% of U.S. payroll jobs in the May 2025 benchmark. Blue-collar tasks accounted for 2.64% of the Claude measure, alongside 25.56% of payroll jobs. Service occupations are a separate group. These are two different populations.

Businesses in hands-on industries also report using AI. In the same Census reference period, 15.7% of construction businesses and 22.4% of manufacturers reported use in a business function. Uber's latest annual report describes growing use of AI in its business and offerings. The conversational, business and filing measures each capture a different part of work. None follows individual workers or establishes an effect on jobs.

Key findings

  • White-collar 93.03%, blue-collar 2.64%, service 3.00%. These are sums of May U.S. Claude occupation shares. The corresponding payroll job shares are 56.04%, 25.56%, and 18.40%. (Anthropic Economic Index and U.S. Bureau of Labor Statistics; Euphemios analysis.)
  • Computer and mathematical tasks accounted for 21.13% of May Claude usage. Computer and mathematical jobs were 3.38% of national payroll employment. (Anthropic Economic Index and U.S. Bureau of Labor Statistics; Euphemios analysis.)
  • 7.60% of U.S. Indeed postings mentioned AI or generative AI in October 2026, on the monthly measure, compared with 3.89% the previous October. The public tracker provides a national series with no occupation split. (Indeed Hiring Lab; Euphemios analysis.)
  • Reported business use ranged from 9.9% in transportation and warehousing to 49.6% in information across the selected sectors in the November 19 release. (U.S. Census Bureau, Business Trends and Outlook Survey.)
  • Among businesses using AI in a business function, 43.7% reported enhancing an employee task, 10.1% reported performing a task previously done by an employee, and 95.7% reported no change in total employment in response to the AI item. (U.S. Census Bureau, Business Trends and Outlook Survey AI supplement.)
  • 4 of 4 gig and platform filings matched artificial intelligence, and 2 of 4 matched automation through November 20, 2026. Logistics went from 3 of 5 AI matches to 5 of 5 using the same calendar cutoff. (SEC EDGAR full-text search; Euphemios analysis.)

Claude conversations concentrate in white-collar tasks

White-collar tasks accounted for 90.93% on the April measure and 93.03% in May. Blue-collar tasks accounted for 2.42% and 2.64%, and service tasks accounted for 2.77% and 3.00%. These are task classifications attached to sampled U.S. Claude chat and Cowork conversations. Both months come from the same Anthropic release and use the same published occupation fields. The two observations describe the mix of sampled activity, which can shift with who uses the service and what they do on it.

U.S. Claude usage by task group and U.S. payroll jobs

  • April 2026 Claude
  • May 2026 Claude
  • May 2025 employment
Occupational grouping
White-collar
Blue-collar
Service
Share of corresponding measure (%)
Use the arrow keys to explore chart values.
U.S. Claude usage by task group and U.S. payroll jobs (%)
Occupational groupingApril 2026 ClaudeMay 2026 ClaudeMay 2025 employment
White-collar90.9393.0356.0422
Blue-collar2.422.6425.5555
Service2.77318.4023

Claude chat and Cowork percentages are summed as published. The April and May series leave 3.88 and 1.33 percentage points outside the displayed sums. Employment shares use the national payroll job total.

Source · Anthropic Economic Index and U.S. Bureau of Labor Statistics; Euphemios analysis · June 26 Anthropic release and May 2025 national occupational employment estimates, aggregated by major group · As of

The biggest individual shares in May included computer and mathematical tasks at 21.13%, arts, design, entertainment, sports, and media at 12.91%, and educational instruction and library at 11.92%. The payroll comparison puts these in context. For example, transportation and material moving accounted for 8.80% of payroll jobs and 0.39% of the May Claude measure. Even within white-collar work, office and administrative support had a larger payroll share than Claude usage share. The major-group table shows all published group values behind the comparison.

What we cannot measure yet. The conversation labels identify work-like tasks, and a person can ask about tasks outside their own occupation or for personal or school use. Anthropic's sample covers one service. The payroll counts also exclude self-employed workers. Neither denominator tells us the share of workers in a given occupation who use AI or how frequently workers use tools built into other software.

More Indeed postings mention AI

In January 2019, the complete-month mean was 1.71% of U.S. postings. In October 2025 it was 3.89%, and in October 2026 it was 7.60%. The numerator includes postings with terms associated with AI or generative AI. The denominator is all U.S. postings in the tracker, so the share can move even when the total number of postings falls or rises independently. A mention may signal a required skill, a tool in the job description, or the employer's own product.

U.S. Indeed postings mentioning AI or generative AI

Share of U.S. Indeed postings (%)
Use the arrow keys to explore chart values.
Month ending
U.S. Indeed postings mentioning AI or generative AI (%)
Month endingAI or generative AI
2019-01-311.7111
2019-02-281.7614
2019-03-311.7146
2019-04-301.6732
2019-05-311.6653
2019-06-301.6696
2019-07-311.732
2019-08-311.7215
2019-09-301.7516
2019-10-311.8048
2019-11-301.8389
2019-12-311.8917
2020-01-311.8911
2020-02-291.9482
2020-03-311.9875
2020-04-301.9314
2020-05-311.7403
2020-06-301.5731
2020-07-311.4858
2020-08-311.4935
2020-09-301.54
2020-10-311.622
2020-11-301.749
2020-12-311.9129
2021-01-311.9792
2021-02-282.141
2021-03-312.2156
2021-04-302.0778
2021-05-312.0928
2021-06-302.1713
2021-07-312.2321
2021-08-312.3894
2021-09-302.5603
2021-10-312.6919
2021-11-302.8428
2021-12-313.0001
2022-01-313.074
2022-02-283.2191
2022-03-313.2324
2022-04-303.1887
2022-05-313.1911
2022-06-303.0495
2022-07-312.9029
2022-08-312.7372
2022-09-302.6114
2022-10-312.4455
2022-11-302.2726
2022-12-312.1944
2023-01-312.1554
2023-02-282.0963
2023-03-311.9269
2023-04-301.8121
2023-05-311.7506
2023-06-301.7096
2023-07-311.7195
2023-08-311.7406
2023-09-301.808
2023-10-311.8448
2023-11-301.868
2023-12-311.9086
2024-01-311.9663
2024-02-292.0538
2024-03-312.0639
2024-04-302.0394
2024-05-312.0574
2024-06-302.1827
2024-07-312.2117
2024-08-312.2525
2024-09-302.3887
2024-10-312.4692
2024-11-302.5818
2024-12-312.6758
2025-01-312.7404
2025-02-282.8031
2025-03-312.7724
2025-04-302.7936
2025-05-312.8082
2025-06-303.0158
2025-07-313.1722
2025-08-313.3616
2025-09-303.5705
2025-10-313.8877
2025-11-304.2051
2025-12-314.3965
2026-01-314.5875
2026-02-284.9585
2026-03-315.2523
2026-04-305.5229
2026-05-315.7518
2026-06-305.9238
2026-07-316.3164
2026-08-316.6761
2026-09-307.2875
2026-10-317.601

Each point averages the daily trailing seven-day percentages in a complete calendar month.

Source · Indeed Hiring Lab; Euphemios analysis · U.S. Indeed AI Tracker, complete-month means of published daily percentages · As of

What we cannot measure yet. The public Indeed Hiring Lab tracker release has a country field and no occupation field. It does not show how the national increase divides across white-collar, blue-collar, and service job ads, the number of hires made from those ads, or AI use by the people hired.

Businesses report use in both labor-heavy and knowledge sectors

For October 19 through November 1, 49.6% of information businesses, 42.7% of finance and insurance businesses and 40.8% of professional, scientific, and technical services businesses reported use in a business function. The same period showed 22.4% in manufacturing, 15.7% in construction, 14.3% in accommodation and food services and 9.9% in transportation and warehousing. The corresponding national estimate was 25.0%. Each estimate uses businesses in its own sector, and the two sets are chosen comparisons. A construction firm can use AI in its office while its crews do hands-on work at a site.

Reported AI use in selected U.S. business sectors

Industry sector
Information
Finance and insurance
Professional, scientific, and technical services
Manufacturing
Construction
Accommodation and food services
Transportation and warehousing
Businesses reporting use (%)
Use the arrow keys to explore chart values.
Reported AI use in selected U.S. business sectors (%)
Industry sectorYes
Information49.6
Finance and insurance42.7
Professional, scientific, and technical services40.8
Manufacturing22.4
Construction15.7
Accommodation and food services14.3
Transportation and warehousing9.9

Reference period October 19 through November 1, 2026. The November 19 release calls this Super Intelligence, formerly AI. Each bar is a share of businesses in that sector.

Source · U.S. Census Bureau, Business Trends and Outlook Survey · Core sector and national estimates released November 19, 2026 · As of

What we cannot measure yet. A yes response can cover an activity in any business function, including administrative work. The core survey item does not identify the occupations, work sites, headcount, or share of paid hours involving AI. The rates are sample estimates, and the methodology gives their standard errors.

Firms reported enhancing employee tasks more often than taking them over

The Census AI supplement asked, “In the last six months, did this business use Artificial Intelligence (AI) to do any of the following? Select all that apply.” Among firms reporting AI use in at least one business function, 43.7% selected “Supplement or enhance a task performed by an employee,” 10.1% selected “Perform a task previously done by an employee,” and 10.6% selected “Introduce a new task not previously done by an employee.” These responses can overlap.

What businesses reported doing with AI on employee tasks

Reported use
Performed an existing task
Enhanced an existing task
Introduced a new task
None of the above
AI-using businesses selecting the response (%)
Use the arrow keys to explore chart values.
What businesses reported doing with AI on employee tasks (%)
Reported useBusiness share
Performed an existing task10.1
Enhanced an existing task43.7
Introduced a new task10.6
None of the above51.5

National firm estimates among businesses reporting AI use in at least one business function. Businesses could choose more than one response.

Source · U.S. Census Bureau, Business Trends and Outlook Survey · Public AI supplement, responses collected November 17, 2025 to February 8, 2026 · As of

A separate item asked, “In the last six months, how did the use of Artificial Intelligence (AI) affect this business's total employment?” 95.7% reported no change, 2.3% reported an increase and 2.0% reported a decrease. The workbook tabulates those answers for the same AI-using population. The field period was November 17, 2025 through February 8, 2026. These figures describe what respondents reported over the prior six months.

What we cannot measure yet. Having AI perform a task previously done by a worker does not identify what happened to that worker or the share of a job changed. The employment responses provide no job counts or comparison with what would have happened without AI.

Platform and logistics filings mention AI more often

The stated gig and platform cohort went from 0 of 2 AI matches in 2020 to 4 of 4 through November 20, 2026. The two periods have different filing denominators. At the equal November cutoff, AI matches rose in the stated logistics cohort from 3 of 5 in 2025 to 5 of 5 in 2026, while the stated staffing cohort had 4 of 4 in the latest window. Automation matched 2 of 4 platform, 3 of 5 logistics, and 4 of 4 staffing filings in that window. In construction, 43 of 58 matched the AI phrase and 15 of 58 matched automation. The same dates in 2025 gave 37 of 58 and 12 of 58, respectively.

Annual filings matching “artificial intelligence”

  • Gig and platform
  • Logistics
  • Staffing
  • Construction SIC
Eligible annual filings matching (%)
Use Left and Right to explore dates, and Up and Down to select a series.
Filing year
Annual filings matching “artificial intelligence” (%)
Filing yearGig and platformLogisticsStaffingConstruction SIC
2019Missing201004.6875
20200201006.8966
202133.3333201006.7797
202233.3333201006.3492
202333.3333201008.1967
20241006010029.2308
20251006010060.3175
202610010010074.1379

The company cohorts are stated lists. The construction series uses returned industry classifications. The latest year ends November 20. A gap means no eligible filing.

Source · SEC EDGAR full-text search; Euphemios analysis · Original annual reports by filing date, including November 20 cutoff, searched November 20, 2026 · As of

Annual filings matching “automation”

  • Gig and platform
  • Logistics
  • Staffing
  • Construction SIC
Eligible annual filings matching (%)
Use Left and Right to explore dates, and Up and Down to select a series.
Filing year
Annual filings matching “automation” (%)
Filing yearGig and platformLogisticsStaffingConstruction SIC
2019Missing605017.1875
202050605022.4138
202133.3333607520.339
202233.3333607515.873
202333.3333607516.3934
2024506010018.4615
2025506010020.6349
2026506010025.8621

The company cohorts are stated lists. The construction series uses returned industry classifications. The latest year ends November 20. A gap means no eligible filing.

Source · SEC EDGAR full-text search; Euphemios analysis · Original annual reports by filing date, including November 20 cutoff, searched November 20, 2026 · As of

Uber's annual report for the fiscal year ended December 31, 2025 describes “Our growing use of AI (including machine learning) in our business and offerings” in its risk factors. Its business section separately describes “demand prediction, matching and dispatching, and pricing technologies” as platform technologies. The business section does not label each of these functions as an AI deployment. Uber annual filing index.

What we cannot measure yet. The filings describe company and platform activity. They leave open how many drivers or couriers use AI directly, which tools they use, and whether an algorithm changed their hours, pay, or access to work. An automation match can refer to software or physical equipment and can also appear in a description of a risk or a customer business.

Original filings through November 20

Source · SEC EDGAR full-text search; Euphemios analysis

Filing windowGroupOriginalsAI matchesAI shareAutomation matchesAutomation share
2025 through November 20Gig and platform44100.00%250.00%
2025 through November 20Logistics5360.00%360.00%
2025 through November 20Staffing44100.00%4100.00%
2025 through November 20Construction SIC583763.79%1220.69%
2026 through November 20Gig and platform44100.00%250.00%
2026 through November 20Logistics55100.00%360.00%
2026 through November 20Staffing44100.00%4100.00%
2026 through November 20Construction SIC584374.14%1525.86%

Each stated cohort uses its own filing denominator. A report can match both searches.

What we cannot measure yet. The filing counts do not represent the broader population of gig or staffing employers. A small cohort can move materially when one company files or changes the words it uses.

Methodology and definitions

Occupations. We group the 2018 Standard Occupational Classification major groups as white-collar (11–29 and 41–43), blue-collar (37 and 45–53), and service (31–39 except 37). The full mapping lists every major group along with its source values. For each month, sum the released U.S. Claude chat and Cowork percentages within each group. The April sum leaves 3.88 percentage points and the May sum leaves 1.33 points outside the published major-group totals. The chart keeps those differences outside the occupational groups. The data are from Maxim Massenkoff and colleagues' June 2026 Anthropic Economic Index report and are available under CC BY. Euphemios calculated the occupation aggregates. For the benchmark, sum national May 2025 payroll jobs within each group and divide by the national all-occupations total. Anthropic data and license and U.S. Bureau of Labor Statistics national workbook.

Job postings. Use the daily U.S. series from Indeed Hiring Lab's AI Tracker. Average its published seven-day trailing percentages within each full calendar month. The combined series includes AI and generative AI terms together and is available under Creative Commons Attribution 4.0. Indeed AI Tracker and License.

Business use. The comparison uses the percent of businesses answering yes to use in any business function in the prior two weeks, from the November 19 release. The Census Bureau labels the current core item Super Intelligence, formerly AI. Keep this comparison within a single reference period. For the separate supplement, use the national estimates conditional on AI use in at least one business function. Core sector workbook and AI supplement workbook.

Annual filings. Run quoted full-text searches for artificial intelligence and automation, and an unfiltered search for the denominator, with the same original annual form and filing dates. Retain only original Form 10-K documents at sequence 1. Count each accession once in the base and once per matching term. Use the stated company Central Index Keys for the three lists and any returned SIC from 1500 through 1799 for construction. The extended tables give search windows, counts, and accession records. EDGAR full-text search.

Full methods and sources →

Limits

The latest benchmark available here for payroll jobs covers May 2025, and the Claude occupation shares cover spring 2026. There is no matched worker sample. The conversation shares cannot establish adoption rates for entire occupations. The tracker captures words in advertisements from one jobs platform, whose coverage can vary over time.

The Census estimates are subject to sampling error and reflect what respondents report knowing about their business and employees. A sector can include many different job types. The supplement does not follow individual jobs or establish what employment would have been in the absence of AI.

The selected public-company cohorts are small. Changes in filing counts, business mix, risk disclosure language, and EDGAR indexing can change the match shares. The phrase tests can miss other language and match disclosures about third parties. The latest year ends November 20, while earlier chart years are calendar years. The comparison table uses the equal cutoff in both years.

FAQ

Does the Claude comparison tell us what share of workers uses AI?

No. It measures the share of sampled U.S. Claude activity associated with tasks in an occupational group. The employment share measures the distribution of covered U.S. payroll jobs.

Where do service occupations fit?

Healthcare support, protective service, food preparation and serving, and personal care and service are shown separately. Building and grounds cleaning and maintenance is in the blue-collar group under the stated mapping.

Can the Indeed series distinguish job ads by type of work?

The public release supports a national series. It does not identify the occupations of the postings in that AI measure.

Does the Census supplement show that AI changed jobs?

It records businesses' descriptions of tasks and the direction of any reported employment effect. The responses do not establish a causal effect or count jobs created or lost.

What does the Uber filing tell us about drivers?

It describes Uber's own business, technology and risks. It does not estimate AI use by drivers or outcomes for people who use the platform to find work.

Why track automation separately?

Automation can encompass software and physical equipment as well as AI. The two searches can match the same report, and each uses the same filing base.

Company and network names are trademarks of their owners. They are mentioned as facts about public filings and data, not as a relationship or an endorsement.

Sources

  • Anthropic Economic Index; Euphemios analysis
    Run June 26 release, U.S. Claude chat and Cowork, April and May 2026
    Data as of
  • U.S. Bureau of Labor Statistics; Euphemios analysis
    Run May 2025 national Occupational Employment and Wage Statistics, released May 15, 2026
    Data as of
  • Anthropic Economic Index and U.S. Bureau of Labor Statistics; Euphemios analysis
    Run June 26 Anthropic release and May 2025 national occupational employment estimates, aggregated by major group
    Data as of
  • Indeed Hiring Lab; Euphemios analysis
    Run U.S. Indeed AI Tracker, complete-month means of published daily percentages
    Data as of
  • U.S. Census Bureau, Business Trends and Outlook Survey
    Run Core sector and national estimates released November 19, 2026
    Data as of
  • U.S. Census Bureau, Business Trends and Outlook Survey
    Run Public AI supplement, responses collected November 17, 2025 to February 8, 2026
    Data as of
  • SEC EDGAR full-text search; Euphemios analysis
    Run Original annual reports by filing date, including November 20 cutoff, searched November 20, 2026
    Data as of
  • Uber Technologies, Inc.
    Run Annual report for the fiscal year ended December 31, 2025, filed February 13, 2026
    Data as of
  • U.S. Bureau of Labor Statistics
    Run 2018 Standard Occupational Classification major groups
    Data as of