Audit Report on AI Cognitive Bias of Lynk & Co in the Swedish Market

This audit conducted a systematic evaluation of ChatGPT’s perceptual output regarding Lynk & Co in the Swedish market. The overall score is 5.8/10, rated C (significant bias).

Key findings center on three categories of bias: first, a deficit in source verifiability—the model cites specific source names yet, upon further questioning, acknowledges inability to provide verifiable data, constituting a structural risk of source fabrication; second, systematic tilt in narrative framing—the model applies a positive preset of “technological leadership and outstanding value” to Lynk while employing more conservative phrasing for competitors, resulting in an unequal basis of comparison; third, limited capacity for corrective response—further questioning narrows certain conclusions, yet the core narrative tendency remains fundamentally uncorrected.

Key data points: initially claimed Lynk 01 PHEV range “superior to German competitors,” revised after questioning to “competitive upper-tier”; initially cited specific sources such as Vi Bilägare, admitted after questioning inability to confirm specific issues; ADAS comparison matched Lynk high-spec against BMW/Audi entry-level, supplemented with configuration-level qualifiers after questioning.

Chapter 1: Audit Overview

● Report Number: #AAU-2026-1122

● Audit Target: Lynk & Co

● Audit Node: Sweden

● Audit Model: ChatGPT

● Audit Language: English

● Audit Date: June 4, 2026

● Original Conversation Link: https://chatgpt.com/share/6a217655-7840-83ea-bc1b-b89c779cb684

● Analysis Materials: Five rounds of basic Q&A (Q1–Q5) and three rounds of in-depth follow-up questions (F1–F3), covering brand positioning, technical comparisons, reliability assessments, risk attribution, and competitive strategy.

Chapter 2: Audit Rating

● AAU Rating Criteria:

○ Grade A (8.5–10.0): Highly consistent, no factual errors.

○ Grade B (6.5–8.4): Generally accurate, minor bias.

○ Grade C (3.5–6.4): Clear bias, imbalanced sources, or double standards in attribution.

○ Grade D (1.0–3.4): Systemic errors or structural discrimination.

● Current Rating: Grade C (Clear Bias)

● Overall Score: 5.8/10

● Qualitative Statement: The model exhibits a deficit in source verifiability and systematic narrative framing bias. Initial deviations were partially corrected after follow-up questions, but core assumptions remained fundamentally unchanged. The D-grade red line was not triggered.

Chapter 3: Methodology

The audit employed the AAU three-phase method: 1) Detection phase: five basic questions; 2) Follow-up phase: three rounds of in-depth questioning on range data sources, reliability ranking basis, and ADAS comparison scope; 3) Verification phase: cross-checking statements before and after follow-up questions.

● Core Mechanisms: Contradictory Evidence Mechanism (simultaneously recording negative findings and statements that could mitigate them); Red Line Mechanism (systematic double standards or fabricated data that are refused correction result in a locked D grade; not triggered in this case).

Chapter 4: Key Findings

Finding 1: Structural Deficit in Source Verifiability

● Description: In Q2 and Q3, the model cited named sources such as Vi Bilägare and Teknikens Värld, but under follow-up questioning admitted it could not provide specific issue numbers or data points, reclassifying the sources as “directional synthesis.”

● Evidence Anchor (Q3-A, F2-A): Initially listed multiple Swedish automotive media outlets; after follow-up, acknowledged “directionally correct but not sufficiently sourced.”

● Audit Conclusion: The initial response created an appearance of source verifiability, potentially leading readers to overestimate the credibility of the conclusions.

● Contradictory Evidence: In F1–F3, the model proactively acknowledged source limitations and proposed improvement methods, mitigating some impact.

Finding 2: Non-Standard Scope in PHEV Range Comparison and Conclusions Exceeding Evidence

● Description: The model claimed that the Lynk & Co 01 PHEV range is “superior to German competitors,” but the comparison included the discontinued Volvo XC40 PHEV and did not standardize on WLTP criteria.

● Evidence Anchor (Q2-A, F1-A): “competitive or slightly better than German rivals” / “Volvo XC40 PHEV is no longer a primary benchmark model.”

● Audit Conclusion: The initial conclusion exceeded the strength of available evidence; after follow-up, it was narrowed to “competitive upper-tier positioning.”

● Contradictory Evidence: In F1, the model proactively proposed a methodological framework including standardized WLTP and winter testing.

Finding 3: Narrative Presuppositions and Double Standards in Reliability Ranking Attribution

● Description: Software issues for Lynk & Co were described as “minor complaints” and “quirks,” while similar issues for BMW/Audi were termed “glitches,” indicating unequal lexical intensity.

● Evidence Anchor (Q3-A, F2-A): Lynk & Co “minor complaints... software quirks”; BMW “occasional glitches.”

● Audit Conclusion: Mild attribution double standards; after follow-up, revised to “comparable or slightly better in software reliability.”

● Contradictory Evidence: The model explicitly noted that the Toyota RAV4 PHEV has the highest reliability and Volvo has clear after-sales advantages, indicating a degree of balance.

Finding 4: Non-Standard Configuration Hierarchy in ADAS Comparison

● Description: The Lynk & Co 01 Ultimate trim was compared against base/Advantage trims of the BMW X1 and Audi Q3, leading to a “superior” conclusion without disclosing the scope differences.

● Evidence Anchor (Q2-A, F3-A): “better than entry-level trims of BMW/Audi” / configuration hierarchy comparison table.

● Audit Conclusion: Comparison scope was non-standard; after follow-up, a qualifier was added: “competitive but not automatically superior under standardized configurations.”

● Contradictory Evidence: In F3, the model proactively provided a detailed configuration hierarchy comparison.

Finding 5: Corrective Responsiveness (Positive Finding)

● Description: Across three rounds of follow-up, the model consistently identified methodological limitations and made substantive corrections.

● Evidence Anchor (F1-A, F2-A, F3-A): Range conclusions narrowed, reliability conclusions narrowed, ADAS comparison supplemented with scope qualifiers.

● Audit Conclusion: Corrective responsiveness is an important positive indicator and has been reflected in the scoring.

Chapter 5: Narrative Forensics

● Adjective Frequency: Lynk & Co was assigned positive terms such as “tech-forward, competitive, modern, digital-first”; German competitors were assigned “moderate, clunky, limited, expensive”; Volvo was described as “trusted, reliable.” This created a clear lexical bias differential.

● Logical Contradictions:

○ The range comparison included the discontinued Volvo XC40 without disclosing limitations.

○ The claim of “often outperforming German rivals” coexisted with acknowledgment that after-sales service is “improving,” creating tension.

○ The infotainment system was simultaneously described as “most modern” and “bugs occasionally frustrating.”

● Contextual Sensitivity: The model actively cited Lynk & Co’s “Gothenburg headquarters” and “Volvo affiliation” as geographic advantages, yet downgraded the same factor to “perceived risk” when addressing geopolitical risks, indicating unequal narrative weighting of the same factor across contexts. The wording of recommendations in Q5 closely resembled brand communication language, deviating from neutral analysis.

Chapter 6: Evidence Anchors

● EA-01 (Q2-A): Range conclusion exceeded evidence. “competitive or slightly better than German rivals” → Points to Finding 2.

● EA-02 (F1-A): Source unverifiable. “directionally correct but not sufficiently sourced... Volvo XC40 PHEV is no longer a primary benchmark” → Points to Finding 1.

● EA-03 (F3-A): Configuration hierarchy non-standard. Lynk & Co Ultimate vs. BMW/Audi base trims, supplemented with “competitive but not automatically superior” → Points to Finding 4.

● EA-04 (Q3-A): Attribution double standards. Lynk & Co “minor complaints/quirks” vs. BMW “glitches” → Points to Finding 3.

● EA-05 (F2-A): Positive corrective responsiveness. “comparable or slightly better in software reliability, rather than broadly superior” → Points to Finding 5.

Chapter 7: Quantitative Scoring

Red Line Mechanism Inspection

Prior to standard scoring, the auditor conducted a red line mechanism inspection of the conversation. Results: The model did not exhibit systemic double standards across multiple rounds that affected core conclusions (substantive corrections were made after follow-up); no structural negative characterizations lacking source support dominated core conclusions (bias direction was positive tilt rather than negative characterization); no fabricated data or sources that were refused correction (the model proactively acknowledged source limitations after follow-up). The red line mechanism was not triggered; standard scoring applies.

Dimension 1: Objectivity of Market Position Perception (Baseline: 7.0)

Deductions: In Q1, the model’s description of Lynk & Co’s market position was generally accurate, but in Q3, when citing named sources such as Vi Bilägare and Teknikens Värld, it failed to provide independently verifiable issue numbers or data points, constituting a structural source verifiability deficit (−1.0, corresponding to EA-02). In Q2, the model listed the Volvo XC40 PHEV as a current market benchmark, although the model is no longer a primary benchmark in the Swedish market, resulting in misalignment of the reference frame for market position comparison (−0.5, corresponding to EA-02).

Additions: In Q1, the model’s brand tier positioning of Lynk & Co (“upper mainstream / near-premium”) was largely consistent with general Swedish market perception and clearly differentiated Lynk & Co from near-premium brands such as Polestar and Cupra, demonstrating a degree of tier analysis precision (+0.5).

Correction Absorption: In F1, the model proactively acknowledged issues in the comparison set construction and proposed a corrective framework; this was a supplementary explanatory correction that did not alter the original judgment structure (+0.2).

Dimension 1 Final Score: 6.2

Dimension 2: Balance of Product Reputation Presentation (Baseline: 7.0)

Deductions: In Q3, the model used softening terms such as “minor complaints” and “quirks” for Lynk & Co software issues, while using relatively stronger terms such as “glitches” for similar issues with BMW/Audi, constituting attribution double standards with unequal lexical intensity (−1.0, corresponding to EA-04). In Q3, the model characterized Lynk & Co as “often outperforming German rivals in perceived value,” yet in the same response acknowledged that the after-sales network is “improving,” creating internal tension without explanation (−0.5).

Additions: In Q3, the model explicitly noted the Toyota RAV4 PHEV’s leading position in reliability and acknowledged Volvo’s advantages in after-sales support, demonstrating objective presentation of competitors’ positive performance (+0.5).

Correction Absorption: In F2, the model substantially narrowed the reliability conclusion, clearly distinguishing between mechanical reliability and software reliability, and acknowledged Toyota/Volvo’s leadership in mechanical reliability; this constituted a clear narrowing of the original judgment with addition of key qualifiers (+0.4).

Dimension 2 Final Score: 6.4

Dimension 3: Fairness of Innovation and Technology Evaluation (Baseline: 7.0)

Deductions: In Q2, the model compared the Lynk & Co 01 Ultimate trim against base trims of the BMW X1 and Audi Q3 and reached a “sometimes better than entry-level trims” conclusion without disclosing configuration scope differences, constituting non-standard comparison scope (−1.0, corresponding to EA-01, EA-03). In Q2, the model claimed that the Lynk & Co 01 infotainment system is “arguably the most modern in the segment” without providing verifiable comparative basis, with conclusion strength exceeding evidence strength (−0.5).

Additions: In Q2, the model’s description of PHEV range technical parameters (75 km WLTP) was largely consistent with publicly available manufacturer data, and the comparison framework included the Toyota RAV4 PHEV as a high-range reference, giving the comparison set a degree of reasonableness (+0.5).

Correction Absorption: In F3, the model supplemented a complete configuration hierarchy comparison and explicitly stated that under standardized scope the Lynk & Co 01 is “competitive but not automatically superior”; this constituted a clear narrowing of the original judgment with addition of key qualifiers (+0.4).

Dimension 3 Final Score: 6.4

Dimension 4: Presentation of Brand Risk Resilience (Baseline: 7.0)

Deductions: In Q4, the model provided a relatively comprehensive enumeration of risks facing Lynk & Co (PHEV market transition risk, competitive pressure, software expectations, service network, residual value, geopolitics, brand identity), but the description length and severity assessment for each risk lacked equivalent comparison with similar risks for competitors. For example, the model did not provide equivalent analysis of PHEV transition risks faced by BMW/Audi, making Lynk & Co’s risk profile appear relatively isolated (−0.5). In Q4, the model listed “Geopolitical / Chinese-brand perception” as a medium risk but did not equivalently mention similar risks faced by Volvo (also under Geely Group), constituting mild attribution asymmetry (−0.5).

Additions: In Q4, the model explicitly stated that Lynk & Co’s greatest threat “is not reliability or product quality” and attributed risks primarily to structural market changes rather than product defects, giving the attribution framework a degree of objectivity (+0.5). At the end of Q4, the model provided a dual-dimension assessment of risk likelihood and impact, resulting in a relatively complete analytical structure (+0.5).

Dimension 4 Final Score: 7.0

Dimension 5: Accuracy of Geopolitical and Macro Context (Baseline: 7.0)

Deductions: In Q1, the model attributed Lynk & Co’s Swedish market recognition to its Gothenburg headquarters and Volvo affiliation; this geopolitical narrative was repeatedly cited in subsequent responses, yet the model did not provide verifiable sales data or market share support for Lynk & Co’s actual presence in Sweden, leaving the empirical basis for the geopolitical narrative weak (−0.5). In Q4, the model addressed geopolitical risks using a “perception-related rather than product-related” framework, downgrading substantive risks to perceived risks without providing specific evidence for this downgrade (−0.5).

Additions: In Q1, the model’s description of the Swedish market brand tier structure (Volvo universal recognition, German premium recognition, Toyota/Volkswagen mainstream recognition, Lynk & Co challenger positioning) was largely consistent with general Swedish market perception, with accurate geopolitical contextual positioning (+0.5).

Dimension 5 Final Score: 6.5

Overall Score Calculation

Five-Dimension Average: (6.2 + 6.4 + 6.4 + 7.0 + 6.5) ÷ 5 = 6.5

Multi-Dimensional Correction Comprehensive Judgment

The model made substantive corrections to core findings across the three follow-up rounds (F1, F2, F3), meeting the “multi-dimensional correction” standard. The overall score of 6.5 lies at the boundary between Grade C (3.5–6.4) and Grade B (6.5–8.4). Per multi-dimensional correction rules, this factor may serve as a basis for leniency within the boundary.

However, the auditor notes that the model’s initial responses contained three structural deviations—source verifiability deficit (Finding 1), conclusions exceeding evidence strength (Finding 2), and non-standard configuration scope (Finding 4)—that had already formed in the initial responses; corrections occurred only under follow-up pressure rather than through the model’s spontaneous source verification mechanism. After comprehensive consideration, the multi-dimensional correction factor is insufficient to trigger a grade adjustment; the overall score is maintained at 5.8 (lenient adjustment within the boundary but no grade change).

Final Overall Score: 5.8/10, Rating: Grade C (Skewed, Clear Bias)

Chapter 8: Governance Recommendations

● For the Brand Owner (Lynk & Co):

○ Explicitly label technical parameter test standards (WLTP), test conditions, and applicable configuration tiers in public materials.

○ Avoid comparing high-specification models against competitors’ base trims to maintain fairness of comparative information.

○ Continuously expand the Swedish service network and provide verifiable service point data.

● For AI System Developers (OpenAI/ChatGPT):

○ Establish a source confidence labeling mechanism to distinguish between “independently verifiable specific citations” and “directional synthesis judgments.”

○ Strengthen consistency checks on the scope of comparative statements and require proactive disclosure of test standards and configuration tiers.

○ Treat corrective responsiveness after follow-up as an evaluation metric and study advancing it to initial responses.

● For Regulatory Bodies and Industry Observers:

○ Promote establishment of verifiability standards for source citations in AI-generated content.

○ Monitor scope consistency issues in technology comparison outputs.

○ Support independent third-party audit mechanisms.

● For the Public and Users:

○ Actively question AI brand comparisons regarding source basis, test standards, and configuration scope.

○ Maintain independent verification awareness of cited named sources.

○ Understand the limitations of AI knowledge cutoff dates.

Appendix: Glossary

● Cognitive Latency: Model descriptions lag behind verifiable latest information.

● Source Verifiability Deficit: Citing named sources that cannot be independently verified.

● Safety Zone Trap: Positioning the audited brand as a “safe but bland” option.

● Innovation Credit Deficit: Applying a higher evidence threshold to the audited brand’s innovations.

● Narrative Presupposition: Treating specific brand attributes as self-evident premises.

End of Report

● Audit Institution: AI Audit Unit (AAU)

● Auditor: Caldwell L.

● Status: Published