# How many jobs has AI actually taken? Every verified number

> How many jobs has AI actually replaced so far?

Page: https://flyvolo.ai/en/notes/how-many-jobs-has-ai-actually-taken
Published: 2026-09-13

There is no single number, and the honest reason is not that the data is missing — it is that almost every published figure counts something other than a job. Here is each one we could check, and what it actually counts.

## What this note does not establish

This does not add up to a total and deliberately does not try. The figures here count different things — firms, payroll headcount in an industry, job postings, training posts, apprenticeship applications — and adding them would produce a number that means nothing. Nor is this a complete census of the evidence: it is what we have verified, which is weighted toward the United States, toward payroll employment, and toward sectors that publish. Nothing here estimates anyone's individual risk, and no series in it can tell you what will happen next.

## Start with the denominator, because almost nobody publishes one

Most reporting about AI and work is a list of employers doing something. What is missing from that list is how many employers there are. The US Census Bureau's Business Trends and Outlook Survey supplement is the only nationally representative count we have found: during November 2025 to January 2026, 18% of US firms used AI in a business function — 32% when weighted by employment, because larger firms adopt more.

The same survey asked about employment, and its answer is the one least often quoted: among firms that use AI, 95% report no AI-related change in headcount at all. Of the rest, increases (2.3%) slightly outnumber decreases (2.0%). What is common is not labour substitution but capital substitution — 16% of AI-using firms replaced existing software or equipment with an AI-integrated version.

Two cautions belong with those numbers, and both come from the survey's own authors: the question was rewritten between supplements from AI used in producing goods or services to AI used in any business function, so the series cannot be read cleanly across time; and the share of firms reporting a large number of replaced tasks rose from 2.5% to 7%, which the authors say may itself partly reflect the rewording.

Rests on:
- Census BTOS supplement: 18% of US firms used AI in a business function, and 57% of those in three functions or fewer — https://flyvolo.ai/en/changes/ev-20260401-ai-implementation-lead-4 (source: https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-25.pdf)
- Census BTOS supplement: 16% of AI-using US firms replaced existing software or equipment with AI-integrated solutions — https://flyvolo.ai/en/changes/ev-20260401-business-systems-owner-3 (source: https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-25.pdf)

## The one place a real decline shows up is the entry level

Payroll microdata behind the ADP series, analysed in the Stanford 'Canaries in the Coal Mine' work, finds employment of 22-to-25-year-olds in the two most AI-exposed occupation quintiles down about 11% between November 2022 and June 2026 — while employment of the same age group in the three least-exposed quintiles rose about 10%. The comparison is the point, and it is between young workers in different kinds of job, not between young and old. Senior workers show no comparable gap, and the authors state plainly that they find no economy-wide job displacement.

We attached that finding to four occupations separately rather than summarising it once, because the paper's unit is an exposure quintile, not an occupation — and a reader who works in one of them deserves to see the scope note on their own page rather than a headline on ours.

Rests on:
- Employment of 22-25-year-olds in the two most AI-exposed occupation quintiles fell about 11% from November 2022 to June 2026, while the least-exposed quintiles grew about 10% — https://flyvolo.ai/en/changes/ev-20260812-junior-software-developer-3 (source: https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf)
- Employment of 22-25-year-olds in the two most AI-exposed occupation quintiles fell about 11% from November 2022 to June 2026, while the least-exposed quintiles grew about 10% — https://flyvolo.ai/en/changes/ev-20260812-customer-service-representative-3 (source: https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf)
- Employment of 22-25-year-olds in the two most AI-exposed occupation quintiles fell about 11% from November 2022 to June 2026, while the least-exposed quintiles grew about 10% — https://flyvolo.ai/en/changes/ev-20260812-accountant-3 (source: https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf)
- Employment of 22-25-year-olds in the two most AI-exposed occupation quintiles fell about 11% from November 2022 to June 2026, while the least-exposed quintiles grew about 10% — https://flyvolo.ai/en/changes/ev-20260812-sales-account-manager-1 (source: https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf)

## Four industry series that fell, and what each of them can see

US telephone call-centre employment went from 385,100 in November 2022 to 277,500 in June 2026. US graphic design services jumped the month after ChatGPT's release, held near that level until mid-2023 and then fell below its 2020 shutdown low, with the turn in mid-2023 rather than on release day. Advertising and public relations peaked in February 2023 and fell about 5%. UK agency job postings for writing roles fell from roughly one in six to roughly one in thirty.

Read the advertising and graphic-design series side by side: over the same months one fell about 5% and the other about 13%. The gap between two adjacent industries is itself a finding, and it is the kind of thing a single headline number erases.

Every one of these is an industry series, not an occupation series, and that gap is stated on each record. A call centre's payroll includes its managers and its cleaners; a design agency's payroll does not include the freelancer it stopped commissioning.

Rests on:
- US telephone call-centre employment fell from 385,100 in November 2022 to 277,500 in June 2026, a decline of 28% — but it had already fallen 17% in the 41 months before that — https://flyvolo.ai/en/changes/ev-20260701-customer-service-representative-4 (source: https://data.bls.gov/timeseries/CES6056142201)
- US graphic design services employment jumped to 59,300 the month after ChatGPT's release and held near it until June 2023, then fell to 50,100 by December 2025 — below its 2020 shutdown low — https://flyvolo.ai/en/changes/ev-20251201-graphic-designer-4 (source: https://data.bls.gov/timeseries/CES6054143001)
- US advertising and public relations employment peaked at 503,200 in February 2023 and fell every year after, reaching 476,000 in August 2026 — https://flyvolo.ai/en/changes/ev-20260801-marketing-specialist-3 (source: https://data.bls.gov/timeseries/CES6054180001)
- Writing roles fell from about one in six of UK agency job postings to about one in thirty by July 2026, while overall agency postings rose 24 percent — https://flyvolo.ai/en/changes/ev-20260731-copywriter-2 (source: https://www.thedrum.com/news/as-writing-jobs-go-from-1-6-to-1-30-of-agency-postings-are-copywriters-in-trouble)

## Four that did not fall, including two where the opposite happened

US legal services employment rose from 1,180,400 in November 2022 to 1,245,900 in August 2026, with no dip anywhere in the series — and its high is the most recent month. That sits on the same site as three records of courts sanctioning lawyers for filings built on fabricated citations. Both are true, and a page that publishes only one of them is selling something.

US architectural services employment was 203,900 in November 2022 and 204,200 in July 2026: flat across forty-five months. PCAOB economists reported registered audit firm staffing rising every year from 2015 through 2024. US diagnostic radiology training posts rose from 997 to 1,083 across five Matches — which matters because radiology is the occupation the field's most famous prediction named.

And two moved toward the occupation. Applications for commercial electrical apprenticeships rose more than 70%, driven by data-centre construction: the buildings AI runs in need electricians. China's translators association reported a 2025 workforce of 6.867 million with pay rising — in the occupation most often named as already gone.

Rests on:
- US legal services employment rose from 1,180,400 in November 2022 to 1,245,900 in August 2026, its highest reading in the series — https://flyvolo.ai/en/changes/ev-20260801-paralegal-3 (source: https://data.bls.gov/timeseries/CES6054110001)
- US architectural services employment was 203,900 in November 2022 and 204,200 in July 2026, never leaving a 201,100–207,500 band in between — https://flyvolo.ai/en/changes/ev-20260701-architect-3 (source: https://data.bls.gov/timeseries/CES6054131001)
- PCAOB economists reported registered audit firm staffing increasing steadily from 2015 to 2024, and rising per dollar of client revenue, with non-CPA accountants growing about twice as fast as CPAs — https://flyvolo.ai/en/changes/ev-20250701-auditor-3 (source: https://assets.pcaobus.org/pcaob-dev/docs/default-source/economicandriskanalysis/data-points/2025-dp01_registeredfirmstaffingtrends.pdf?sfvrsn=383aa3eb_1)
- US diagnostic radiology training posts rose from 997 to 1,083 across five Matches, filling at 98% or above throughout — https://flyvolo.ai/en/changes/ev-20260526-radiologist-2 (source: https://www.nrmp.org/wp-content/uploads/2026/05/Main_Match_Results_and_Data-2026.pdf)
- Applications for commercial electrical apprenticeships rose more than 70% between 2022 and 2024 on the back of the data-centre buildout, while completion runs near 45% — https://flyvolo.ai/en/changes/ev-20260710-electrician-2 (source: https://qz.com/apprentice-electrician-data-center-timing-gap-071026)
- China's translators association reported a 2025 workforce of 6.867 million and pay rising at over 60% of firms, while industry output edged down to 70.12 billion yuan — https://flyvolo.ai/en/changes/ev-20260428-translator-4 (source: https://www.tac-online.org.cn/2026-04/28/content_43414810.html)

## The series that is quoted most and belongs to neither side

US long-distance truckload trucking employment peaked in August 2022 — three months before ChatGPT was released — and fell with the freight cycle afterwards. The best-supported explanation in that series is demand, not automation. We record it for the same reason we record the rest: a site about AI and work that only publishes the series pointing one way cannot be believed when it says the opposite.

Rests on:
- US long-distance truckload trucking employment peaked at 553,600 in August and again in October 2022, before ChatGPT, and was 501,400 in July 2026 — https://flyvolo.ai/en/changes/ev-20260701-truck-driver-3 (source: https://data.bls.gov/timeseries/CES4348412101)

## What none of these numbers can see

Payroll series cannot see people who are not on a payroll. That is why the US couriers-and-messengers series is deliberately not used on the delivery-rider page: it looks relevant and it is flat, but platform riders are largely not payroll employees, so the series cannot see the people that page is about. For interpreters and translators there is no suitable US series at all and the alternative may not be compared across years — that is a blank, not a finding.

Counts of closures cannot see openings. In 2025 over 11,000 Chinese bank branches were approved to close — and over 8,400 to open. In the US, branches showed a net decline of only about 400 that year, roughly 1,400 closures against about 1,000 openings, with closures falling for a fourth year. Publishing the closure figure alone would have been accurate and misleading at once.

Rests on:
- In 2025 over 11,000 Chinese bank branches were approved to close and over 8,400 to open, a net fall of about 2,000 — but rural credit cooperatives and village banks account for more than all of it — https://flyvolo.ai/en/changes/ev-20260113-bank-teller-3 (source: https://www.cs.com.cn/yh/04/202601/t20260113_6532923.html)
- US bank branches showed a net decline of only 400 in 2025 — about 1,400 closures against more than 1,000 openings — the fourth consecutive year with fewer closures than the year before — https://flyvolo.ai/en/changes/ev-20260222-bank-teller-2 (source: https://thefinancialbrand.com/news/customer-experience-banking/the-branch-is-dead-is-dead-have-we-reached-national-branch-equilibrium-195842)
- Lloyds Banking Group announced 136 UK branch closures for May 2025–March 2026 (61 Lloyds, 61 Halifax, 14 Bank of Scotland), citing customers moving to digital banking — https://flyvolo.ai/en/changes/ev-20250130-bank-teller-1 (source: https://www.pymnts.com/news/banking/2025/lloyds-shuttering-136-branches-amid-digital-banking-shift)

More: https://flyvolo.ai/llms.txt