AI on the job: who carries the empathy in an AI-written message?
In financial services, insurance, tech and the European headquarters of international groups in Zurich, some exchanges require acknowledging a difficulty without being able to meet the request. The managerial question is specific: does AI-generated empathetic wording help employees maintain that position, or impose a stance that is harder to sustain?
What the literature establishes
Arlie Russell Hochschild (1979, American Journal of Sociology) offers a framework for understanding how social rules define the emotions appropriate to a situation and the efforts made to conform to them. This work distinguishes the visible expression of an emotion from work on what one actually feels. Applied to AI, this distinction cautions against equating a warm message with an emotionally easy interaction.
The mechanism behind the wording
Alicia A. Grandey (2000, Journal of Occupational Health Psychology) conceptualises emotional labour through emotion regulation, distinguishing in particular between changing emotional expression and changing feelings. Her model suggests a hypothesis for writing assistants: they may make it easier to display an emotion without changing what the employee feels. The text is then available, but the person must still be able to stand behind it in the next exchange.
The premature conclusion
One might conclude that AI should be excluded from sensitive exchanges. These papers do not justify that conclusion: a suggestion may also help someone find respectful words in a difficult situation. The useful distinction concerns wording the employee can genuinely stand behind, rather than empathy assessed solely from the text. (our executive and employee training programmes)
What the evidence cannot establish
These two articles offer theoretical frameworks; they test neither generative assistants nor their effects in Zurich companies. They therefore cannot establish whether an AI-generated reply increases or reduces emotional strain. An assessment confined to the politeness of the message would, however, overlook whether its expressed stance can be maintained when the customer replies.
A practical check in Zurich
As part of SHR’s “IA on the job” programme, a pilot could focus on responses to complaints in a Zurich team, using an authorised tool and without transferring customer data to an unapproved service. Over a period defined in advance, record the proportion of AI-assisted drafts whose tone needs changing because the employee could not stand behind it in a direct exchange, distinguishing these changes from factual corrections. Comparing this proportion with that for messages written without AI on comparable requests, using voluntary and aggregated feedback, provides a local check without claiming causality. The training discussion then has a concrete criterion: which respectful formulations can we actually sustain when speaking to the customer? To go further: explore the AI on the Job training in Zurich, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Zurich
- ai on the job
- Zurich
- research
- training zurich