It’s hard to believe ChatGPT would even work for this, unless this politician’s values and positions are identical to ones OpenAI have trained ChatGPT to have, the pretraining is going to leak through no matter what instructions they stuff into the prompt.