Social scientists increasingly deploy large language models as synthetic participants, using a short demographic persona prompt to answer surveys or play games. Yet persona prompting faces open questions: we do not know whether a prompt summons the intended person, whether persona descriptions and persona choices align, or whether a weak persona prompt can be strengthened. I investigate these questions through a proposed design and a small pilot. Drawing on the Twin-2K-500 dataset, I construct three persona prompts for each respondent: a minimal demographic card, the same card enriched with the person's own narratives, and the same card enriched with narratives from a matched stranger. Each persona completes a scripted interview about its life, ranked by a blind LLM judge, and answers a battery of decision items, both directly and after re-reading its own interview, scored against the person's own responses. In a pilot with ten respondents, the judge tends to prefer the minimal card's interviews, yet conversational quality shows no clear connection to later choices. Letting a persona re-read its own interview before deciding appears to help only when the interview is built from the person's own material. The full study will examine these patterns at scale.