IA on the job: can an HR message promise more than intended?
Generative assistants make it easier to draft messages about onboarding, training and career prospects. The managerial question is specific: how can organisations prevent AI-generated wording from being received as a promise they have not decided to make?
What the literature establishes
Denise M. Rousseau (1989, Employee Responsibilities and Rights Journal) defines the psychological contract in terms of individual beliefs about reciprocal obligations between a person and an organisation. This theoretical framework distinguishes those beliefs from an agreement that is necessarily shared or formalised. A message about future training may therefore contribute to a perceived obligation, even if its author only intended to describe a possibility.
From differing interpretations to a sense of violation
Elizabeth Wolfe Morrison and Sandra L. Robinson (1997, Academy of Management Review) propose a model distinguishing the perception of an unfulfilled promise from the emotional response of violation. Their analysis considers divergent interpretations of what was promised, not only a refusal to honour a commitment. Applied to AI-assisted writing, this mechanism suggests reviewing assurances added during revision; it does not demonstrate that AI itself causes these reactions.
The premature conclusion
One might conclude that every message should be made conditional and all committing language removed. That would confuse preventing an unintended promise with refusing to make explicit commitments. The task is to distinguish what has been decided, what depends on an identifiable condition and what remains a possibility. (our executive and employee training programmes)
What these papers cannot establish
Both articles offer theoretical frameworks that predate generative assistants; they measure neither how often these tools introduce promises nor their effects on employees. A review confined to factual accuracy and tone may nevertheless overlook the scope of the commitments expressed. Nor can the concept of a psychological contract, on its own, determine the legal status of a sentence.
A practical check in Fribourg
Within SHR's « IA on the job » programme, an exercise in Fribourg could compare the human draft, the AI-assisted version and the approved message using anonymised cases from the agri-food sector, bilingual industry and the university hub. For each assurance about training, working hours or career prospects, the manager identifies the decision authorising it, its conditions and the person empowered to make that commitment. The measure would be the proportion of assurances lacking this validation among all assurances identified, before and after review, with a separate count of wording added by the assistant. For French and German messages, each version would undergo this check: it examines expressed commitments without claiming, on its own, to measure what recipients will take away. To go further: explore the AI on the Job training in Fribourg, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Fribourg
- ai on the job
- Fribourg
- research
- training fribourg