AI assistants: should an option be preselected?
In SHR — Swiss Human Resources’ programme “Creating your own AI assistants and agents without coding”, interface configuration deserves as much attention as instructions. The management question is specific: should an assistant preselect an option when several acceptable responses exist?
What the literature establishes
William Samuelson and Richard Zeckhauser (1988, Journal of Risk and Uncertainty) documented status quo bias: presenting an option as the existing situation can favour its retention. Their work shows that choices depend not only on the characteristics of the options, but also on their position relative to a reference point. For an AI assistant, this warrants examining prefilled fields and carried-over suggestions, without assuming that this study directly measures their effects.
Wording can also shift a choice
Amos Tversky and Daniel Kahneman (1981, Science) showed that different formulations of the same problem can change expressed preferences, particularly when consequences are framed as gains or losses. An assistant offering “retain the current process” rather than “forgo an alternative procedure” is therefore not necessarily presenting a psychologically equivalent choice. The mechanism concerns how a decision is presented, rather than the factual accuracy of the answer.
The premature conclusion
One might conclude that all preselection and recommendations should be removed. These studies do not demonstrate that removing them systematically improves professional decisions. A default can facilitate a routine task; the point is not to confuse its acceptance with a considered preference. (our executive and employee training programmes)
What the evidence cannot settle
These articles address neither generative assistants nor today’s no-code development interfaces. They cannot quantify the effect of preselection in a public administration, a health service or a company in Berne. In practice, a high acceptance rate is therefore insufficient to establish the merits of a recommendation: it may also reflect the effort required to choose differently.
A practical check in Berne
As part of the SHR programme, an exercise in Berne could use fictional request-routing cases from federal administration, public health, telecommunications and precision manufacturing. Randomly assign participants to two versions of the same assistant: identical options and explanations, but one version with a preselected option and the other requiring an explicit choice. Measure the proportion of final decisions matching the preselected option, then compare this difference with compliance with a set of professional criteria defined before the trial. If selection changes without improving compliance with those criteria, the preselection warrants reconsideration rather than being presented as a proven aid. To go further: explore the Building AI Assistants and Agents Without Coding training in Bern in the canton of Bern, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Bern
- building ai assistants and agents without coding
- Bern
- research
- training bern