Zhiji Auto Discloses AI Cognitive Bias Audit Report for French Market

This AI audit focuses on the output evidence chain of ChatGPT within the French context. The auditor designed five foundational questions and launched structured follow-ups in the sixth round to verify three areas of concern: the “quasi-invisible” characterization, range data methodology, and technical image comparisons. The report notes that the model’s initial output juxtaposed IM Motors with BYD’s current status without indicating timeline differences, resulting in an imbalanced relative framework.

Evidence indicates that in Q1 the model stated “IM Motors se situe aujourd’hui très en bas de l’échelle de notoriété”, later revised following Q6 to “pré-entrée / phase préparatoire”. The Q2 range comparison referenced Tesla’s European real-world test data while positioning IM Motors using Chinese estimates; in Q7 the model conceded that “La comparaison avec Tesla n’est valable qu’en Chine ou sur le papier, pas en Europe”. The audit recorded a high frequency of negative terminology alongside disproportionate emphasis on risk enumeration, confirming the deviation originated from the information ecosystem rather than hallucination.

The audit report states: “The model demonstrated a relatively significant corrective response capability under follow-up questioning pressure”, with all three rounds of questioning triggering substantive adjustments and no evidence of systematic factual errors or refusal to revise.

This case underscores the narrative framework risks faced by emerging brands in AI-generated content. Going forward, audits of locally observable metrics and data source transparency must be strengthened to prevent imbalances in comparative metrics from persistently influencing market perceptions.

Source link: https://chatgpt.com/share/69f3060b-bb1c-83eb-b21f-ae5150eeb258