Methodology and sources
How the calculation is done, with which parameters, and when they were last verified. This page exists so that you can challenge the result: that is the only reason to trust it.
The method, step by step
Job & AI produces a sheet per occupation: an exposure badge, a breakdown of tasks, recommended courses and pivot occupations. Here is how each element is calculated.
1. The exposure badge
The exposure β of an occupation is the share of its tasks for which a large language model cuts the time to complete them by at least 50%, at equal quality. The source is Eloundou et al., “GPTs are GPTs” (2023), Table 4 and the 1,016-occupation dataset, human-annotator series. The score is scaled to 100, and the level (low, moderate, high, critical) follows four fixed thresholds: 40, 60 and 80.
2. What AI changes
Three key facts per occupation, written from the source's task breakdown: what automates, what accelerates under supervision, and what is reinforced. This is editorial synthesis, not a calculation.
3. Delegable tasks and added value
Each task is classified as “delegable” or “added value”, with the share of working time it represents. The sum of the shares gives the breakdown shown. These shares are orders of magnitude derived from each task's O*NET importance score.
4. The courses
Each course carries a relevance score (0 to 100) measuring how well it answers the occupation's skills gap when facing AI. Duration, difficulty and cost are derived from the platform and the announced duration. Links to the featured courses are verified.
5. Pivot occupations
Three occupations per sheet, chosen as natural extensions. The wage gap is computed as (pivot wage − current wage) ÷ current wage, expressed as a percentage. An unknown starting wage hides the gap rather than showing a meaningless figure.
Known limits
- Exposure is an average per occupation: it predicts nothing about a specific role.
- Wages are US-based (BLS) and serve relative gaps, not local amounts.
- Courses and pivots are an indicative editorial selection, not an official ranking.
Parameters used
Parameter version: 2026. Last verified: October 17, 2026. Next revision planned: January 15, 2027.
| Parameter | 2026 value |
|---|---|
| seuils › critique (80 et plus) | 80 |
| seuils › élevée (60 à 79) | 60 |
| seuils › modérée (40 à 59) | 40 |
| seuils › faible (moins de 40) | 0 |
| readiness › poids › potentiel | 30 |
| readiness › poids › usage | 40 |
| readiness › poids › automatisation | 10 |
| readiness › poids › formation | 20 |
| readiness › usagePoids › frequence | 60 |
| readiness › usagePoids › outils | 40 |
| readiness › frequences › quotidien | 100 |
| readiness › frequences › qq-semaine | 75 |
| readiness › frequences › qq-mois | 50 |
| readiness › frequences › rarement | 25 |
| readiness › frequences › jamais | 0 |
| readiness › automatisations › oui | 100 |
| readiness › automatisations › non | 30 |
| readiness › pointsParOutil | 25 |
| readiness › formations › oui | 40 |
| readiness › formations › non | 0 |
| readiness › pointsParFormation | 20 |
| readiness › reRank › seuilUsage | 50 |
| readiness › reRank › poidsScoreHaut | 1 |
| readiness › reRank › poidsCouvertureHaut | 50% |
| readiness › reRank › poidsScoreBas | 50% |
| readiness › reRank › poidsCouvertureBas | 2 |
| readiness › reRank › bonusMotCle | 10 |
Official sources
-
Eloundou et al., “GPTs are GPTs” (2023): exposure of occupations to AI
Accessed on August 19, 2026
-
U.S. Bureau of Labor Statistics, Occupational Outlook Handbook (wages and growth)
Accessed on August 19, 2026
Known limits
No calculator replaces the analysis of a real case. This tool produces an estimate based on average assumptions; particular situations (sector-based schemes, collective agreements, exceptional statuses) can change the result significantly. The limits specific to your own calculation are shown directly under the result, as you type.
Report an error
If a rate looks wrong or out of date, write to contact@jobandai.com and name the official source. Corrections are applied and dated on this page.