> "we ask ChatGPT to answer ideological questions by proposing that, while responding to the questions, it impersonates someone from a given side of the politic

> "we ask ChatGPT to answer ideological questions by proposing that, while responding to the questions, it impersonates someone from a given side of the political spectrum. Then, we compare these answers with its default responses, ie, without specifying ex-ante any political side, as most people would do. In this comparison, we measure to what extent ChatGPT default responses are more associated with a given political stance." The result would also be consistent with some other conclusions than value bias.