Before deployment18 questions · ≈ 10 min
Identify anticipated effects.
Describe the task, affected groups and planned conditions before beginning the pilot.
- No names requested
- Data processed locally
- Descriptive profile without a global score
Assess · Instrument in development
Where is your project now?
Choose the path that matches the project stage. The detailed method and its limitations follow.
Key point. These tools structure work analysis and collective discussion. They provide no diagnosis, clinical threshold or automatic evidence of compliance.
Before and afterNine dimensionsNo global score
An equilibrium to follow
The Voyage of Life: Youth · 1842
The luminous palace makes the destination appear secure, while the boat remains exposed throughout the journey. The work reminds us that an expected benefit can quickly become fragile if the real conditions of deployment are not followed: workload, autonomy, skills, collective work and health. View the artwork ↗
01 / Choose
Three paths for three situations.
Start with the project stage. Methodological limits remain visible in every path.
After deploymentReal work
Compare observed effects.
Review workload, autonomy, errors, skills and work groups after the system goes live.
- Same scope as the initial assessment
- Incidents and workarounds included
- Corrective decisions documented
Research and depth27 items · 9 dimensions
Understand the long instrument.
Review the T0/T1 architecture, provisional interpretation rules, limitations and validation programme.
- Three items per dimension
- No decision threshold
- Scientific validation still required
02 / Architecture
The Gollac framework, supplemented by three AI-specific mechanisms.
The instrument does not measure general concern about technology. Its questions describe concrete changes in workload, decisions, relationships and know-how.
18 items · Reference framework
Six psychosocial risk factors.
The six families in the Gollac report are applied to work changes caused or amplified by an AI system.
G01–G03Work intensity and working time
G04–G06Emotional demands
G07–G09Autonomy and room for manoeuvre
G10–G12Social relationships and recognition
G13–G15Value conflicts
G16–G18Job insecurity
9 items · Exploratory extension
Three AI-related mechanisms.
These additional dimensions make visible problems that may otherwise remain dispersed across the six general categories.
AI01–AI03Opacity and ability to challengeUnderstand the output, see uncertainty and obtain an effective human review.
AI04–AI06Cognitive review workloadPrompt, review, correct and take responsibility for an output that is not fully under one’s control.
AI07–AI09Skill erosionPractise and transfer knowledge less, while becoming dependent on a system that may be unavailable or wrong.
Why three items per dimension?
A single question would poorly cover a complex phenomenon. Three formulations examine several expressions of the same dimension and support an initial subscale analysis.
02 / What is assessed
Nine dimensions linked to observable work situations.
Each dimension combines three questions. The examples below summarise their rationale without replacing the questionnaire’s full wording.
| Dimension | Change examined | Example signals |
|---|---|---|
| 01Work intensity and working time Gollac | Volume, pace, fragmentation and additional work required to use or review the tool. | More cases, multiple alerts, and review work that encroaches on breaks or working hours. |
| 02Emotional demands Gollac | Difficult human situations created or concentrated by use of the system. | Complaints after an error, dehumanised communication, or fear of harming someone. |
| 03Autonomy and decision-making Gollac | Ability to retain professional judgement, organise one’s work and depart from a recommendation. | Imposed method, automated priorities, or explicit or implicit penalty for refusing. |
| 04Social relationships and recognition Gollac | Cooperation, responsibilities, and visibility of reasoning or review work. | Fewer useful exchanges, unclear responsibility after an error, or expertise made invisible. |
| 05Value conflicts Gollac | Gap between what the tool or organisation requires and professional rules or work considered to be high quality. | Speed prioritised over safety, an output contrary to professional judgement, or loss of meaning. |
| 06Job insecurity Gollac | Uncertainty about tasks, jobs, career paths and evaluation criteria. | Contradictory announcements, fear for employment, or an AI score influencing an HR decision. |
| 07Opacity and contestability AI extension | Understanding of data and criteria, visibility of limits and access to human review. | Unexplained output, hidden uncertainty, or no rapid appeal to someone with decision-making authority. |
| 08Cognitive review workload AI extension | Mental effort required to prompt, monitor, detect errors and take responsibility for the output. | Sustained vigilance, plausible errors that are hard to detect, or responsibility without control of the system. |
| 09Skill erosion AI extension | Maintenance of practice, learning by beginners, knowledge transfer and ability to act without the tool. | Training tasks removed, less practice of know-how, or dependency when the system fails or is wrong. |
03 / Two observation points
Compare what was planned with what actually happened.
The 27 items exist in two parallel forms. The subject remains the same, but the reference period and wording change.
Anticipated effects.
Answers concern the system’s planned operation and the most demanding situation reasonably foreseeable for the population concerned.
- Use case and level of automation
- Productivity objectives and planned organisation
- Review rules and appeal routes
- Pilot conditions and ability to work without the tool
Observed effects.
Answers concern the system as actually used over the previous four weeks, including informal practices.
- Review time and corrections
- Incidents, workarounds and differences between teams
- Effects on objectives and working relationships
- Changes in skills and actual dependency
A descriptive comparison
The T1 − T0 difference describes the gap between anticipated risk and observed effects; a positive value suggests deterioration. It should not yet be interpreted as a pure psychometric measure of change until measurement invariance has been established.
04 / Choose the right form
A long form for documentation. A short form to begin discussion.
The two tools share the same broad objective, but differ in structure and level of detail. Their results should not be treated as directly interchangeable.
Development long form27 items
Parallel T0 / T1 assessment.
Three items per dimension, detailed contextual information, a critical incident item and two free-text fields.
- Nine dimensions and provisional subscales
- Same anonymous code and scope at both times
- Suitable for documented pilots and validation studies
- No total score or decision threshold
Interactive short form18 questions · ≈ 10 min
Initial online screening.
A guided experience designed to quickly surface points requiring attention and prepare discussion of the project.
- Questionnaire available directly in the browser
- No identifying data requested
- Exploratory profile and exportable action plan
- Discussion aid, not scientific validation
Are you looking to compare the technical capabilities of several models? That requires a separate benchmark protocol. The “Understand” page explains how to read these evaluations without confusing them with deployment effects on work. Understand benchmarks →
05 / Validation programme
What remains to be demonstrated before calling this a scientific instrument.
Transparency about this status prevents a useful framework from being prematurely treated as an exact measure. Each step must be documented before proposing a threshold or normative interpretation.
Content validity
Check with workers, prevention professionals and experts that the items cover relevant situations and remain understandable.
Factor structure
Examine whether answers actually organise according to the nine proposed dimensions.
Reliability
Assess the consistency of items within a dimension and stability of results when the situation does not change.
T0 / T1 invariance
Check that the same dimension is measured comparably before and after deployment.
Sensitivity to change
Determine whether the instrument detects real change and only then discuss possible thresholds.
Fieldwork · Research · Collaboration
Would you like to use the long form or contribute to its evaluation?
Field feedback is particularly useful for examining how items are understood, their relevance across occupations and the conditions for future validation.
Contact Dr Charles Broutin on LinkedIn ↗Possible discussions with
- Employers and project leaders
- HR, works councils and employee representatives
- OHS services and prevention professionals
- Academic teams and research organisations
- Organisations preparing an AI pilot
INRS · Six psychosocial risk factors ↗Gollac & Bodier · Expert panel report · 2011ISO 45003:2021 · Psychological health at work