Trait or Bias? Why item keying is indispensable when comparing language model responses across languages

Cross-linguistic differences in large language model (LLM) questionnaire responses are increasingly interpreted as evidence of culturally patterned behaviour. Such inferences, however, assume that observed score differences reflect variation in the target construct rather than artefacts of measureme