Methodology & Statistics

How we built the dataset, how AI exposure scores are calculated, and what the numbers mean for Canada's 20 million workers.

Key numbers

516

NOC 2021 unit-group occupations covered

20.1M

Canadian jobs represented (2023 data)

4.6

Job-weighted average AI exposure (0–10)

37.6%

Jobs in Low AI-exposure occupations (score 1–3)

38.9%

Jobs in High AI-exposure occupations (score 6+)

$590B

Annual payroll in high AI-exposure occupations (6+)

Data sources

All underlying data is open-access, published by the Government of Canada under the Open Government Licence.

Source	Publisher	What we use
COPS 2024–2033 Summary CSV	ESDC	Employment 2023, projected job openings 2024–2033, labour market outlook per occupation
3-Year Employment Outlooks 2025–2027	ESDC / Job Bank	Regional outlook ratings (Shortage / Balance / Surplus) per occupation × province
Job Bank wage reports	ESDC / Statistics Canada	Median hourly wage by NOC 2021 unit group, LFS reference period 2023–2024 (454/516 occupations); converted to annual at 2,080 hrs
Table 98-10-0586-01 (2021 Census)	Statistics Canada	Median annual employment income by NOC 2021 unit group, 2020 reference year — fallback for occupations not covered by Job Bank wage reports
NOC 2021	Statistics Canada	Occupation titles, 5-digit NOC codes, TEER levels, category structure
Job Bank	ESDC	Occupation profile URLs and canonical titles (via internal concordance ID lookup)

Occupations

The 516 occupations come from the COPS 2024–2033 summary CSV, which lists every NOC 2021 unit group with labour market projections. We exclude the "All Occupations" aggregate and all TEER / major-group subtotals, keeping only individual 5-digit unit groups.

485 of 516 occupations have employment figures. The remaining 31 are suppressed by Statistics Canada due to small sample sizes. Those 485 represent 20.1 million jobs — the full Canadian employed labour force.

By major occupational category

Category	Occupations	Jobs (2023)	Avg AI Exposure
Sales and service	61	4.95M	2.9
Trades, transport and equipment operators	94	3.37M	2.2
Business, finance and administration	51	3.26M	5.5
Education, law and social services	52	2.48M	3.8
Natural and applied sciences	66	1.93M	6.0
Health occupations	43	1.62M	3.2
Manufacturing and utilities	68	956K	2.5
Art, culture, recreation and sport	37	647K	4.1
Management occupations	15	428K	5.4
Natural resources and agriculture	24	425K	2.0
Other occupations	5	71K	1.4

Wage data

Pay figures come primarily from Job Bank wage reports, which publish median hourly wages derived from the Labour Force Survey (LFS), reference period 2023–2024, updated November 2025. These are actual paid wages reported by workers — not advertised or offered rates. Hourly medians are converted to an annual equivalent using 2,080 hours (52 weeks × 40 hours), the standard Canadian full-time year.

Coverage: Job Bank wage data is available for 454 of 516 occupations (those with a direct Job Bank profile). The remaining 62 fall back to Statistics Canada Table 98-10-0586-01 (2021 Census, 2020 income reference year) or, where neither source is available, a TEER-level estimate.

Annual conversion note: Multiplying a median hourly wage by 2,080 hours gives a full-time-equivalent annual figure. Workers in high part-time occupations (retail, recreation, food service) may earn less annually than this implies — the hourly median itself is still accurate, but actual annual take-home depends on hours worked.

For NOC 2021 codes starting with 6–9, the TEER level is the second digit of the code. For codes 0–5, the TEER is the first digit. This corrects a common misreading of the code structure.

Labour market outlook

The 3-Year Employment Outlooks 2025–2027 XLSX contains ~44,000 rows — one per occupation × province/territory. There is no pre-computed national row. We derive a national outlook by taking the modal (most common) rating across all 13 regions for each occupation.

Outlook ratings are converted to a numeric proxy for the scatter view axis:

Rating	Numeric proxy	Occupations
Strong risk of Shortage	+12	38
Moderate risk of Shortage	+6	65
Balance	+2	365
Moderate risk of Surplus	−4	11
Strong risk of Surplus	−8	6
Undetermined	—	31

AI exposure scoring

Each occupation receives a score from 0 to 10 measuring how much cognitive AI is expected to reshape that occupation — both through direct automation (AI doing the knowledge work) and through productivity amplification (fewer workers needed per unit of output).

Scope: The AI Exposure score measures cognitive/digital automation only — the impact of language models, AI agents, and digital automation on the information-processing parts of a job. It does not measure industrial robots, autonomous vehicles, or physical automation machinery. Those are captured separately in the Robotics section below. A welder scores low on AI Exposure even though welding robots exist; a truck driver scores low even though autonomous vehicles are coming.

Scores are generated by GPT-4o (OpenAI) reading a structured Markdown description of the occupation, including: NOC code, TEER level, major category, employment 2023, projected job openings 2024–2033, and labour market conditions. The model is instructed to respond with a single JSON object: {"exposure": N, "rationale": "..."}. Temperature is set to 0.2 for consistency.

Score anchors

Score	Tier	Canadian examples
0–1	Minimal	Underground miners, roofers, landscapers, oil field workers, commercial divers
2–3	Low	Electricians, plumbers, firefighters, welders, dental hygienists, heavy equipment operators
4–5	Moderate	Registered nurses, police officers, veterinarians, social workers, construction managers
6–7	High	Accountants, engineers, teachers, HR managers, journalists, financial advisors
8–9	Very high	Software developers, graphic designers, translators, paralegals, data analysts, web designers
10	Maximum	Data entry clerks — routine digital processing with no physical component

Score distribution

Score 1

11.3%

Score 2

14.4%

Score 3

11.9%

Score 4

15.3%

Score 5

8.2%

Score 6

11.5%

Score 7

11.2%

Score 8

15.1%

Score 9

1.0%

Score 10

0.2%

Percentages are job-weighted (by 2023 employment). The distribution is concentrated at scores 2–4 because Canada's largest employment sectors — trades, transport, sales and service — are physical-work dominated with low cognitive AI exposure. The job-weighted average across all 20.1M jobs is 4.6 / 10.

AI exposure by sector

Highest-exposure sectors are knowledge-work-heavy; lowest are physical/trades. Note that low AI exposure does not mean low automation risk overall — see Robotics section below.

Natural & applied sci.

6.0

Business & finance

5.7

Management

5.5

Art & culture

4.1

Education & law

3.9

Health

3.5

Sales & service

2.9

Manufacturing

2.4

Natural resources

2.3

Trades & transport

2.3

AI exposure by education level

Higher education generally correlates with higher AI exposure, with some nuance at the college/apprenticeship level:

University+

5.7

College / 2yr

4.1

High school

3.4

Vocational

3.0

No degree req.

2.4

Note on TEER 4–5 (no degree / vocational): Workers in these groups score low on cognitive AI exposure — but many face significant physical automation risk from robotics in manufacturing, agriculture, and logistics. Low AI score ≠ safe from automation overall.

High AI-exposure occupations

Occupations scoring 6 or higher represent 7,840,500 jobs (38.9% of the labour force) and a combined annual payroll of $590 billion CAD (based on 2023–2024 LFS median wages).

Occupation	Score	Jobs (2023)	Median Pay
Data entry clerks	10/10	36,000	$49K
Software developers and programmers	8/10	155,700	$100K
Software engineers and designers	8/10	113,100	$117K
Computer systems developers and programmers	8/10	45,700	$90K
Data scientists	8/10	36,600	$96K
Web developers and programmers	8/10	28,700	$80K
Financial auditors and accountants	6/10	247,300	$84K
Administrative officers	6/10	238,500	$60K
Administrative assistants	6/10	233,800	$55K
Information systems specialists	6/10	229,400	$96K

Lowest AI-exposure occupations

These occupations score 1/10 on cognitive AI exposure — but several (miners, oil field workers) face separate high physical automation risk. See Robotics section below.

Occupation	Score	Jobs (2023)	Robotics risk
Underground production and development miners	1/10	29,400	High
Dancers	1/10	13,600	Low
Fishermen/women	1/10	10,600	Low
Oil and gas drilling, servicing and related labourers	1/10	5,100	High
Underground mine service and support workers	1/10	3,100	High
Mine labourers	1/10	3,100	High
Athletes	1/10	2,700	Low

Robotics and physical automation

Physical automation — industrial robots, autonomous vehicles, precision agriculture equipment, and mining systems — displaces workers in sectors that score low on the AI Exposure axis. These two waves of automation attack different parts of the occupational distribution.

The dual-threat framework: A complete picture of technological displacement requires both dimensions. Cognitive AI threatens knowledge workers (score 6–10). Physical robotics threatens manual and trade workers (score 1–3). Some occupations face both (e.g. cashiers, couriers).

Research consensus on Canadian automation risk

Study	Key finding
Brookfield Institute (2016) — The Talented Mr. Robot	42% of Canadian jobs at high risk of automation (probability >70%)
OECD (2016) — The Risk of Automation for Jobs	9% of Canadian jobs at high automation risk (task-level); 30% face significant change
Bank of Canada (Georgieva et al., 2018)	~2M Canadian jobs (~10% of workforce) face high automation risk
Statistics Canada (Lu, 2019)	Manufacturing and transportation most exposed; auto sector employment fell 30% 2000–2018 partly due to automation
Acemoglu & Restrepo (2020) — Robots and Jobs	Each robot per 1,000 workers reduces employment-to-population ratio 0.18–0.34% and wages ~0.4%
OECD (2023) — Employment Outlook	~27% of Canadian jobs face high automation risk under updated task-based measures
WEF (2023) — Future of Jobs Report	83M jobs globally displaced vs. 69M created by 2027; net −14M across automation + AI

Robotics risk by NOC sector

NOC Sector	Robotics risk	Key drivers
Manufacturing and utilities	Very High	Auto assembly robots (ON), food processing lines, packaging automation
Trades, transport and equipment operators	High	Autonomous trucks (AV platforms), automated equipment, delivery drones
Natural resources and agriculture	High	Autonomous mining haul trucks, robotic harvesters, precision ag (John Deere 8R)
Sales and service	Moderate	Self-checkout (40%+ in major grocers), robotic food prep, automated retail
All other sectors	Low	Physical automation has minimal direct impact on core job tasks

Dual-threat occupations

These occupations face displacement from both cognitive AI and physical automation simultaneously:

Occupation	AI Exposure	Robotics risk	Mechanism
Transport truck drivers	3/10	High	AI route optimization + autonomous vehicle platforms
Cashiers	6/10	High	AI customer service + self-checkout / scan-and-go
Agricultural workers	2/10	High	AI crop monitoring + robotic harvesting and planting
Postal workers and couriers	3/10	High	AI route optimization + automated sorting + delivery drones
Food processing labourers	2/10	Very High	Computer vision quality inspection + robotic processing lines
Bank tellers	6/10	Moderate	AI banking apps + ATM / kiosk expansion

Pipeline

Seven scripts, run in order, build everything from raw government files to the website:

Script	Output	Description
`build_occupations.py`	`occupations.json`	Reads COPS CSV, extracts 516 NOC unit groups, assigns categories and Job Bank search URLs
`build_jobbank_urls.py`	Updates `occupations.json`	Queries Job Bank Solr API to look up concordance IDs for direct profile URLs (462/516)
`scrape_jobbank.py`	Updates `occupations.json`	Scrapes Job Bank summary and requirements pages to add canonical titles and employment requirement bullets (~3 min, resumable)
`scrape_jobbank_wages.py`	`data/jobbank_wages.json`	Scrapes Job Bank wage reports for median hourly wages (2023–2024 LFS) for 454 occupations; resumable
`generate_pages.py`	`pages/*.md`	Generates one Markdown file per occupation (used as LLM input)
`make_csv_ca.py`	`occupations.csv`	Compiles wages (Job Bank 2023–2024 primary, Census 2020 fallback), employment, education, and outlook into a flat CSV
`score.py`	`scores.json`	Calls GPT-4o via OpenAI API to score each occupation 0–10
`build_site_data_ca.py`	`site/data.json`	Merges CSV + scores + scraped Job Bank data into the single JSON the frontend loads

Provincial data pipeline

Provincial pages are generated by a ninth script, generate_province_pages.py, which builds one standalone HTML page per province from the same data. The script adds an interactive squarify treemap (identical to the national page), all occupation tables re-weighted to provincial employment, and sector exposure bars.

Script	Output	Description
`generate_province_pages.py`	`site/province/*.html`	Generates 10 province pages (ON, QC, BC, AB, MB, SK, NS, NB, NL, PE) with provincial employment estimates, interactive treemap, and sector breakdowns

What gets scraped from Job Bank

For the 462 occupations with a direct Job Bank profile, scrape_jobbank.py collects:

Canonical title — the most common job-posting title on Job Bank (e.g. "Shop Clerk" instead of the NOC label "Retail salespersons, visual merchandisers and related workers"). This is what appears in treemap boxes and tooltips.
Employment requirements — up to 3 bullet points from the Employment requirements section, sourced from the NOC description (e.g. "A bachelor's degree in computer science… is usually required."). Shown in the hover tooltip alongside the TEER-based education tier.

Provincial employment estimation

The COPS 2024–2033 dataset provides employment at the national level only. Provincial employment per occupation is not published. To generate province-level estimates we use the 3-Year Employment Outlooks 2025–2027 XLSX, which contains sector-level (major occupational category) employment for each province and territory.

Estimation formula

For each occupation and each province, estimated provincial employment is:

prov_jobs = round( national_jobs × prov_cat_emp / national_cat_emp )

Where:
  national_jobs — COPS 2023 employment for this occupation (national)
  prov_cat_emp — provincial employment in this occupation's major category (from 3-year outlook XLSX)
  national_cat_emp — sum of all COPS national employment figures within the same major category

This is a proportional scaling approach: if Saskatchewan accounts for 19% of national "Trades, transport and equipment operators" employment, then each trades occupation in SK is assigned 19% of its national job count. It is not a survey — it is an estimate — but it grounds provincial numbers in actual sector-level data rather than a simple population share.

Provincial average exposure

Each province's average AI exposure score is computed as a job-weighted mean across all occupations, using the estimated provincial job counts:

avg_exposure = Σ( prov_jobs_i × exposure_i ) / Σ( prov_jobs_i )

Provinces with a larger share of employment in high-exposure sectors (Business & Finance, Natural & Applied Sciences) score higher; provinces with heavier Trades, Transport, or Natural Resources employment score lower.

Provincial average exposure by province

Province	Avg AI Exposure	Est. Employment	Highest-exposure sector
Ontario (ON)	4.66	8,227,017	Business & Finance (7.67 avg)
British Columbia (BC)	4.58	2,769,384	Business & Finance (7.67 avg)
Quebec (QC)	4.58	4,610,643	Business & Finance (7.67 avg)
New Brunswick (NB)	4.46	419,742	Business & Finance (7.67 avg)
Nova Scotia (NS)	4.46	536,697	Business & Finance (7.67 avg)
Alberta (AB)	4.45	2,438,058	Business & Finance (7.67 avg)
Manitoba (MB)	4.39	700,986	Business & Finance (7.67 avg)
Newfoundland & Labrador (NL)	4.38	254,916	Business & Finance (7.67 avg)
Prince Edward Island (PE)	4.35	88,893	Business & Finance (7.67 avg)
Saskatchewan (SK)	4.29	617,463	Business & Finance (7.67 avg)

The spread across provinces is relatively narrow (4.29–4.66) because occupation mix at the sector level is similar nationally. The main differentiator is the share of workers in Business & Finance vs Trades & Transport — provinces like Ontario and BC have a larger white-collar knowledge workforce while Saskatchewan and Alberta carry a larger trades and natural resources base.

Treemap rendering on province pages

Each province page embeds a self-contained JavaScript squarify treemap that:

Loads ../data.json (national occupation data) and ../province_data.json (provincial sector breakdowns) via fetch()
Recomputes natCatEmp on the client (sum of national jobs per category)
Scales each occupation: scaled_jobs = national_jobs × provCatEmp[category] / natCatEmp[category]
Filters to occupations with at least 1 estimated provincial job
Renders the squarify treemap on a <canvas> element with hover tooltips showing provincial job counts

This means the province pages share no pre-computed occupation-level provincial data — all estimation happens live in the browser using the same formula as the Python generator.

Known limitations

AI Exposure score does not capture physical automation risk. Occupations in manufacturing, transport, agriculture, and mining score low on AI Exposure because their core tasks are physical — but they face significant displacement risk from industrial robots and autonomous systems. The AI Exposure score alone underestimates total automation risk for these workers. Use the Robotics Risk indicator alongside the AI score for a complete picture.
Annual wage is a full-time equivalent estimate. Job Bank wages are hourly medians multiplied by 2,080 hours. Workers in high part-time occupations (retail, recreation, food service) typically work fewer annual hours, so actual annual earnings will be lower than the displayed figure for those occupations.
62 occupations use older wage data. Occupations without a direct Job Bank profile fall back to Statistics Canada's 2021 Census (2020 income reference year). The 2020 figures reflect the COVID-19 pandemic year and may understate typical earnings for hospitality, tourism, and arts occupations.
5 senior manager codes share one wage value. Statistics Canada suppressed individual figures for NOC codes 00011–00015 in the census table. All five use the consolidated "00018 Senior managers" value as a fallback.
National outlook is a modal proxy. The 3-year outlook XLSX contains only provincial/territorial rows. We use the most common rating across all regions, which may not reflect national conditions accurately for occupations with strong regional variation (e.g., oil-patch trades).
AI exposure is an LLM opinion. Scores come from a language model reading a text description — they are not empirically measured automation probabilities. The model calibration anchors reduce variance, but individual scores should be interpreted as rough tiers, not precise values.
31 occupations lack employment data. Statistics Canada suppresses figures for occupations with very small sample sizes. These occupations are shown in the treemap at a minimal size.
Job Bank URLs: 54 occupations use search fallback. The Job Bank's internal concordance database does not include NOC 2021 codes for 54 unit groups (primarily very new or restructured categories). Clicking those occupations opens a pre-filled Job Bank search instead of a direct profile page.
Employment year is 2023. The COPS dataset uses 2023 as the base employment year, not 2024 or 2025.

Credits

Original concept & code: Andrej Karpathy — karpathy/jobs (US version using BLS data)
Canadian adaptation: Data pipeline, NOC 2021 mapping, and this site
Occupation data: Employment and Social Development Canada (ESDC)
Statistical data: Statistics Canada
Job profiles: Job Bank Canada
AI scoring: GPT-4o via OpenAI API