We conducted a Turing Test in Finland using the Finnish language to evaluate ChatGPT 5.2’s performance in a non‑English cultural context. Because training data unevenly represent languages and cultures, we expected the model to underperform compared with earlier English‑US tests. To improve construct validity, we introduced model‑generated role prompting, allowing participants to interact with both human and model under identical conversational roles. Contrary to expectations, the model passed the Finnish Turing Test. Participants relied heavily on colloquial Finnish cues as markers of human authorship, which turned out to be the main source of error. We reconceptualize the Turing Test as a comparative method for assessing whether an AI can credibly exhibit membership in a particular social world, rather than a pure intelligence test. The approach captures model capability, prompted identity, participants’ insider competence, and AI literacy, providing a useful probe of the human‑machine boundary across domains.
Review