Verellax Index · Edition One · Q3 2026 · roles advertised 1 July to 30 September 2026 · as at 2 October 2026, Edition One freeze
AI was involved in 36.9% of the roles advertised in Q3. When it sits at the centre of the work, only 20.4% of those specs say who is accountable for what it produces.
Verellax read the 57,134 job specifications that 284 employers advertised between 1 July and 30 September 2026, in seven named markets. Each role is read on whether it involves AI, whether AI sits at the centre of the work, and whether the role is accountable for what the model produces.
Each measure takes the one before it as its base.
Source: Verellax workforce intelligence · 57,134 specifications advertised in Q3 · all markets · as at 2 October 2026, Edition One freeze · coverage 39.3% of captured Q3 postings extracted · accountable reads the governed view
Every AI-central role is then read on four tests.
As context beside the tests, not a finding of its own: 70.9% of AI-central roles advertised in Q3 meet none of the four (8,966 of 12,648).
Of 12,648 AI-central roles advertised in Q3, all markets · each test on its own, never added into a score · governed view · as at 2 October 2026, Edition One freeze
US test figures print only if the live and frozen US counts agree at the freeze.
Three findings from Q3
- 01
Almost four in five AI-central roles do not say who is accountable for what the model produces.
79.6% of AI-central roles advertised in Q3 do not say who is accountable for what the model produces (10,063 of 12,648).
Accountable, of AI-central roles · roles advertised in Q3 · as at 2 October 2026, Edition One freeze
Of AI-central roles Do not say who is accountable79.6%Say who is accountable20.4% - 02
AI comes into most roles as a tool, inside work that stays the same.
60.5% of roles advertised in Q3 that involve AI are for people who use AI (12,762 of 21,084).
Relationship, of roles that involve AI · roles advertised in Q3 · as at 2 October 2026, Edition One freeze
The role's relationship to AI · of roles that involve AI Uses AI60.5%Builds it21.7%Governs it4.5% - 03
The redesign reaches entry-level roles least, where people learn the work.
15.6% of entry-level roles advertised in Q3 involve AI (1,186 of 7,591), against 36.9% of all roles.
Involved × seniority · roles advertised in Q3 · as at 2 October 2026, Edition One freeze
Involve AI · each on its own base All roles36.9%Entry level15.6%
01 · Roles
Employers hired people to use AI far more often than to build it or govern it.
The role's relationship to AI, read on the roles advertised in Q3 that involve AI.
Builds AI: models, agents, and the products, systems, and workflows that run on them.
Source: Verellax workforce intelligence · roles advertised in Q3 that involve AI · all markets · as at 2 October 2026, Edition One freeze
02 · Decisions
The decisions about the model are the part of the role most often left unwritten.
The four tests read whether a specification sets out how the person works with the model, who can overrule it, who is accountable for what it produces, and which decisions stay with the person.
9.2% of AI-central roles advertised in Q3 name who can overrule the model, the lowest of the four tests. 22.9% state which decisions the person holds.
Of AI-central roles advertised in Q3, all markets · governed view · as at 2 October 2026, Edition One freeze
Employers ask for AI tool use in 14,432 live specifications and AI governance in 3,856.
Execution100.0%
Coordination99.3%
Judgement94.2%
AI-adjacent32.1%Source: Verellax workforce intelligence · live specs · presence, cross-sectional · mapping coverage per layer, never blended: Execution 32.0%, Coordination 57.8%, Judgement 36.2%, AI-adjacent 26.4% · as at 2 October 2026, Edition One freeze
A stated skill is what the employer asks for, never proficiency. Skills figures are composition, never trends, until coverage is flat.
Markets and sectors
Source: Verellax workforce intelligence · roles advertised in Q3 · each market on its own base · as at 2 October 2026, Edition One freeze
Sorted from highest to lowest. Other markets holds only what the employers we track advertise in other countries. It is never a read of hiring in those countries. Markets are told apart by label, never by colour.
Source: Verellax workforce intelligence · roles advertised in Q3 · each sector on its own base · all markets · as at 2 October 2026, Edition One freeze
Sorted from highest to lowest. A translucent bar is a sector below 10 companies, n disclosed.
| Market | Advertised in Q3 | Employers | Involved | Central | Accountable |
|---|---|---|---|---|---|
| All markets | 57,134 | 284 | 36.9% | 60.0% | 20.4% |
| Ireland | 849 | 44 | 42.8% | 60.6% | 25.0% |
| US | 33,871 | 178 | 42.0% | 61.2% | 20.9% |
| Canada | 1,503 | 86 | 40.7% | 66.6% | 16.0% |
| UK | 7,507 | 165 | 32.2% | 57.7% | 23.1% |
| Singapore | 1,085 | 67 | 31.3% | 57.4% | 24 of 195 |
| Australia | 8,124 | 165 | 21.3% | 53.7% | 20.7% |
| New Zealand | 591 | 43 | 12.4% | 53.4% | 9 of 39 |
Cross-market comparisons are sector-matched. A cell below its floor renders translucent with its count, never suppressed and never headlined.
| Sector | Companies | Involved | Central | Accountable |
|---|---|---|---|---|
| All sectors | 284 | 36.9% | 60.0% | 20.4% |
| AI-native | 16 | 69.5% | 65.1% | 19.4% |
| Technology | 68 | 53.4% | 60.1% | 15.9% |
| Financial services | 66 | 36.3% | 62.8% | 27.4% |
| Professional services | 21 | 32.4% | 55.9% | 15.3% |
| Media and entertainment | 27 | 31.1% | 63.4% | 19.6% |
| Advertising agencies | 33 | 24.0% | 47.7% | 12.3% |
| Consumer brands | 29 | 14.8% | 60.2% | 16.7% |
| Retail | 23 | 11.4% | 52.6% | 17 of 103 |
A sector below 10 companies at the freeze renders translucent with its count.
Method
Verellax does not survey anyone. Every figure is read from what an organisation has already published and carries its date, and every projection names the measures it is built on.
Edition One prints on all collected specs, posting-date basis: roles advertised in Q3: posted 1 July to 30 September 2026, or, where the ad carries no date, first collected in that window, on a complete corpus for each named market. Every all-markets figure carries its coverage beside it, 39.3% of captured Q3 postings extracted. Every test figure reads the governed view. Under 30 roles, counts only; from 30 to 149, the share with its count; from 150, the share. Weekly figures elsewhere on the site stay on live specs.
The method reads roles as written, not what happens after the hire, and absence from a specification is not absence from the organisation.
The full methodHow do I cite this edition?
Cite the edition and the date it was frozen. Exhibits may be reproduced in editorial coverage, in full and unaltered, with attribution, under the Verellax press grant. Corrections append below as dated notes.
Verellax press grant · editorial use · not CC BY 4.0 · the press kit
Send a correctionVerellax Index, Edition One, Q3 2026. Verellax workforce intelligence, 57,134 specifications advertised in Q3, 284 employers, all markets, as at 2 October 2026, Edition One freeze. verellax.com/index/edition-one
Edition One, as a PDF.
Leave a work email and Edition One opens at once, with a copy to your inbox.
- Every exhibit, numbered, each base disclosed
- The distribution page: where do you sit
- Appendix: the four tests, the suppression floor, the corrections route
- No employer is named in Edition One.
Edition Two: January 2027. It may name companies, in the positive, above the floor, after fourteen days' notice.
The edition renders from one pinned pack. Corrections append below as dated notes, with the query shown.