What's this about?
Controlled recommender experiments find that assigning an agent to a booking platform rather than a traveler changes how it treats sponsored listings; this is role-prompt sensitivity, not an audit of OpenAI's new ad product.
What exactly is sponsorship bias?
A recommendation changes because a listing is paid for or because the agent serves the platform rather than the buyer. The paper examines controlled prompts; production effects still need direct measurement.
What's the community saying?
New arXiv preprint; no substantial community discussion sampled.
- Significance: the principal named in a prompt can shape agent incentives.
- Significance: disclosure alone may not ensure unbiased recommendations.
- Significance: test deployed products rather than infer their behaviour from this experiment.