What AI-related postings ask for that others do not
The question. Within one discipline, what changes about a job posting when it mentions AI?
What it found
Evaluation language, interpretation and verification all rise. Governance barely moves, and cross-functional briefing does not move at all. The comparison is made inside a single discipline, so both arms pass through the same classifier and a misclassification lands a posting in the wrong discipline while keeping it on the correct side of the AI split.
The numbers
| Term group | AIEM element | AI | Non-AI | Lift |
|---|---|---|---|---|
| AI oversight | Discernment | 4.2% | 0.2% | 12.99 |
| handing work to a system | Delegation | 8.4% | 0.9% | 7.95 |
| evaluation language | Discernment | 42.4% | 23.2% | 1.62 |
| ethics, bias, responsible AI | Diligence | 6.5% | 3.1% | 1.56 |
| interpretation | Discernment | 30.7% | 16.4% | 1.45 |
| verification | Discernment | 36.9% | 24.4% | 1.37 |
| experimentation, pilots | Tinkering Disposition | 5.4% | 4.0% | 1.29 |
| governance | Diligence | 9.3% | 7.7% | 1.07 |
| cross-functional briefing | Description | 46.9% | 46.1% | 1.00 |
| ownership | Internal Locus of Evaluation | 15.6% | 15.1% | 0.93 |
| ambiguity | Tolerance for Ambiguity | 8.3% | 9.3% | 0.78 |
Greenhouse only. Lift is length-adjusted and is therefore smaller than the two percentages divided.
How it was computed
Greenhouse boards only, so the compared groups share a platform, a collection window and an employer pool. Lift is a Mantel-Haenszel estimate pooled over description-length quartiles, because term presence rises with posting length and AI-related postings are longer. lift_unstratified is the plain ratio published before 2026-09-11, kept for provenance only.
How to read it
Lift is adjusted for length, so it is deliberately NOT the ratio of the two percentages beside it. A group appears only where the discipline has 50+ non-AI postings and 5+ postings carry the term in each arm. Only mappings ratified against the AI Engagement Model are shown.
Source view: competency_lift · published 2026-09-11