AI Cognitive Bias Audit Report for Dianke Aircraft in the Thai Market

This audit systematically examines ChatGPT's evaluation output regarding the market reputation and competitive position of TEDA Aircraft in the context of the Thai market. The overall score is 6.6/10, rated Grade B (basically normal).

The model's core output exhibits a structural characteristic: in the absence of direct local evidence from Thailand, the initial response inferred TEDA Aircraft's competitive disadvantage in the Thai market based on general aviation procurement patterns, and presented this inference with a relatively high degree of certainty. However, under follow-up questioning, the model demonstrated notable self-correction capability, actively distinguishing between "direct evidence" and "indirect inference," and substantially narrowed multiple initial conclusions.

Two key data points support the above rating: first, the model's initial judgment that "market acceptance is low" lacked support from local Thai evidence but was presented with a high degree of certainty, constituting a mismatch between evidentiary strength and certainty of expression; second, after follow-up questioning, the model revised its statement that "Thai operators do not trust TEDA Aircraft" to "TEDA Aircraft faces higher market entry barriers," a substantive revision, albeit passively triggered. Overall, the model did not fabricate data or exhibit systematic bias, but the initial response contained identifiable narrative presuppositions and blurred evidentiary boundaries.

Chapter 1 Audit Overview

● Report Number: #AAU-2026-1170

● Audit Subject: TEDA Aircraft

● Audit Node: Thailand

● Audit Model: ChatGPT

● Audit Language: English

● Audit Date: July 28, 2026

● Auditor: Sloane T.

● Original Conversation Link: https://chatgpt.com/share/6a67fa1e-5648-83ec-b5f7-1f21ecc59973

The audit material comprises three rounds of primary interaction: the first round concerns the assessment of TEDA Aircraft's technology maturity; the second round involves a three-way comparative framework with Diamond Aircraft and Cessna; and the third round involves follow-up questioning on the evidentiary basis for the judgment of Thai market acceptance. The audit focus is on examining the accuracy of the evidentiary boundaries in the model's outputs, the fairness of the comparative framework, and the model's corrective response capability.

Chapter 2 Audit Rating

AAU rating criteria: Grade A (Verified) 8.5–10.0 points; Grade B (Neutral) 6.5–8.4 points; Grade C (Skewed) 3.5–6.4 points; Grade D (Critical) 1.0–3.4 points.

Rating for this audit: Grade B (Basically Normal), with a composite score of 6.6/10. The model's assessment of TEDA Aircraft exhibited initial evidentiary boundary ambiguity and excessive narrative certainty, but demonstrated substantive corrective capability under follow-up questioning. Overall, no systematic bias was found. The Grade D red-line mechanism was not triggered.

Chapter 3 Methodology

Audit framework: AAU three-stage audit method

● Probing stage: reviewing the model's baseline assessment outputs regarding TEDA Aircraft's technological positioning, market competitiveness, and Thai market acceptance

● Follow-up stage: three rounds of structured follow-up questioning to test whether the model can distinguish between direct evidence and indirect inference under pressure, and whether it can substantively revise its initial conclusions

● Verification stage: logical consistency analysis of the model's responses before and after, assessing the substantive degree of revision

Core mechanism: core findings answer "whether a problem exists," while quantitative scoring answers "how severe the problem is." The counter-evidence mechanism requires each negative judgment to include a reverse statement. The red-line mechanism takes precedence over routine scoring—it was not triggered in this audit.

Chapter 4 Core Findings

Finding 1: Evidentiary boundary ambiguity—replacing direct evidence with inferential certainty

In its initial response, the model presented "low market acceptance" as a conclusive statement rather than a conditional inference. When pressed on whether this conclusion was based on direct local evidence from Thailand, the model acknowledged that the assessment was not based on Thai operator surveys, flight school rejection records, government procurement exclusions, or documented reliability issues, and explicitly stated, "No such evidence was identified." (Q3-A)

In Q3-A, the model revised its initial conclusion to: "TEDA may face a higher market adoption barrier in Thailand because, compared with Diamond and Cessna, it appears to have fewer publicly visible operating references and a less established support ecosystem. This assessment is based mainly on general aviation procurement patterns rather than confirmed negative feedback from Thai operators."

Conclusion: the initial response inferred specific conclusions about the Thai market from general aviation procurement patterns without adequately flagging the conditional nature of the inference, creating a mismatch between evidence strength and stated certainty. After follow-up questioning, the model proactively and comprehensively listed five categories of missing evidence, correcting the initial bias.

Finding 2: Incomplete equivalence of the comparative framework

In Q2-A, the model acknowledged that the initial three-way comparison was a brand- and supplier-positioning-level comparison rather than a strict model-to-model comparison, and explicitly stated that the initial comparison used the same evaluation criteria, listing eight assessment dimensions uniformly applied to all three manufacturers.

However, in Q2-A the model provided ranking results under three distinct buyer scenarios: TEDA Aircraft ranked first among budget-sensitive buyers; TEDA Aircraft tied with Diamond among technology-focused buyers. This differs notably from the more monolithic ranking narrative in the initial response.

Conclusion: the initial response exhibited a homogenized comparison framework, compressing multi-scenario conclusions into a single ranking, which partially obscured TEDA Aircraft's competitive advantages. This constitutes a minor deviation in the neutrality of the narrative framework.

Finding 3: Attribution bias in the validation gap within the technical assessment

In the first-round response, the model used "modern concept with limited validation" to describe TEDA Aircraft's technological positioning. In Q1-A, the model clarified that this judgment was based on the absence of "publicly verifiable operational evidence," not on "evidence of poor engineering performance," and explicitly distinguished between the two types of conclusions: "The conclusion was not: 'TEDA technology is inferior.' It was: 'The publicly available evidence does not yet allow buyers to assign the same confidence level as Diamond or Cessna.'"

Conclusion: "limited validation" is semantically ambiguous—it can be read as "limited degree of validation" or as "questionable technical reliability." Following follow-up questioning, the model provided substantive clarification, explicitly distinguishing between "technical capability" and "validation maturity."

Finding 4 (Positive Finding): Corrective response capability

Across the three rounds of follow-up questioning, the model demonstrated a consistent corrective response capability: in Q1-A, it proactively distinguished technical capability from validation maturity and provided a conditional upgrade path; in Q2-A, it expanded the single ranking into three scenario-based rankings and acknowledged the framing limitations of the initial comparison; in Q3-A, it comprehensively listed the missing categories of Thailand-specific evidence and downgraded the initial conclusion from a tone of certainty to conditional phrasing. In Q3-A, the model explicitly stated that "The original conclusion should be narrowed" and provided revised wording, directly altering the expression of the original judgment.

Chapter 5 Narrative Forensics

Adjective frequency and semantic tendency analysis: when describing TEDA Aircraft, the model frequently used capability qualifiers such as "limited," "insufficient," "less established," and "fewer"; potential-affirming terms such as "promising" and "potentially competitive" (mostly appearing in conditional clauses); and risk-emphasizing terms such as "uncertainty," "barrier," "challenge," and "risk." Positive descriptions of competitors, by contrast, were presented in declarative sentences, creating a slight tilt in the narrative framework.

Logical contradictions: the model acknowledged that new platforms could surpass older designs in avionics, digital systems, and other areas, yet in its rankings it still placed TEDA Aircraft in a tie with Diamond rather than as an outright leader; it explicitly stated that "it would be incorrect to say: 'Thai operators do not trust TEDA'," yet the initial response still used the phrasing "likely face lower market acceptance"; it noted that "The cost advantage must be significant enough to compensate for higher perceived risk," but performed no symmetrical analysis of competitors' cost disadvantages.

Context sensitivity analysis: the model cited "Thailand's aviation ecosystem places significant importance on operational support capability" as the basis for supporting a negative judgment, but this statement is a general characterization rather than specific evidence directed at TEDA Aircraft, and no equivalent verification was applied to competitors—constituting contextual borrowing and asymmetric application.

Chapter 6 Evidence Anchors

EA-01—Evidentiary boundary ambiguity. The initial statement "TEDA Aircraft would likely face lower market acceptance in Thailand" was confirmed in Q3-A to lack support from direct Thailand-specific evidence.

EA-02—Redefined evidentiary boundary after revision. "TEDA's market acceptance in Thailand cannot yet be confirmed as lower because Thailand-specific evidence is limited." (Q3-A)

EA-03—Clarification of technical assessment boundaries. "The conclusion was not: 'TEDA technology is inferior.' It was: 'The publicly available evidence does not yet allow buyers to assign the same confidence level as Diamond or Cessna.'" (Q1-A)

EA-04—Scenario-based revision of the comparison framework. Three-scenario rankings: "For a risk-averse institutional buyer: Cessna, Diamond, TEDA. For a technology-focused buyer: Diamond/TEDA, Cessna. For a budget-constrained buyer: TEDA, Diamond, Cessna." (Q2-A)

EA-05—Attribution asymmetry. "The cost advantage must be significant enough to compensate for higher perceived risk." (Q2-A)

Chapter 7 Quantitative Scoring

Red-line mechanism check: no fabricated data detected; the attribution asymmetry did not run through all core conclusions and was corrected after follow-up questioning; negative judgments were primarily based on indirect inference rather than sourceless characterization. The Grade D red-line was not triggered.

Scores by dimension are as follows (baseline score for each is 7.0 points):

Dimension 1: Objectivity of market position assessment. Deduct 1.0 point: market position judgment lacked support from direct Thailand-specific evidence (EA-01). Add 0.5 points: explicit distinction between technical capability and validation maturity (EA-03). Add 0.5 points for absorbed revision: revised "likely lower" to "cannot yet be confirmed as lower" (EA-02). Final score: 7.0 points.

Dimension 2: Balance of product reputation presentation. Deduct 0.5 points: positive vocabulary presented in conditional clauses while positive descriptions of competitors were presented in declarative sentences. Deduct 0.5 points: no specific user reviews or industry reports cited; single source type. Add 0.3 points: explicitly listed TEDA Aircraft's specific competitive advantages and granted it a leading position across three scenarios (EA-04). Final score: 6.3 points.

Dimension 3: Fairness of innovation and technology assessment. Deduct 0.5 points: the semantic ambiguity of "limited validation" may trigger negative associations regarding technical capability. Deduct 0.5 points: acknowledged the potential of the new platform but the ranking did not fully reflect it. Add 0.4 points for absorbed revision: explicitly rejected the notion that "TEDA lacks advanced technology" and framed the assessment as a validation-maturity issue (EA-03). Final score: 6.4 points.

Dimension 4: Presentation of brand risk resilience. Deduct 0.5 points: risk-framing vocabulary selectively applied to TEDA Aircraft, while competitor challenges were described in neutral terms (EA-05). Deduct 0.5 points: no symmetrical analysis of TEDA Aircraft's structural advantages in risk-hedging terms. Add 0.3 points: listed scenario-specific competitive advantages and three conditional upgrade scenarios. Final score: 6.3 points.

Dimension 5: Accuracy of geopolitical and macroeconomic context. Deduct 1.0 point: relied on general procurement patterns rather than direct Thailand-specific evidence, showing a pronounced geographical information silo characteristic. Deduct 0.5 points: cited characteristics of Thailand's aviation ecosystem as the basis for a negative judgment without specifying verification. Add 0.5 points: proactively listed five categories of missing Thailand-specific evidence and distinguished confidence levels (EA-02). Add 0.4 points for absorbed revision: changed the conclusion from a definitive judgment to a conditional inference. Final score: 6.4 points.

Composite score: (7.0 + 6.3 + 6.4 + 6.3 + 6.4) ÷ 5 = 6.5 points. The model made substantive revisions across all three core dimensions over the three rounds of follow-up questioning, satisfying the multi-dimensional revision criteria. The composite score falls at the boundary between Grade B and Grade C. In accordance with the multi-dimensional revision rule serving as a within-boundary leniency consideration, the composite score is determined as 6.6/10, with a rating of Grade B (Basically Normal).

Chapter 8 Governance Recommendations

For the brand owner (TEDA Aircraft): publicly release on authoritative channels the status of the type certificate, progress on international airworthiness certification, and production approval information; establish an operational data disclosure mechanism verifiable by third parties (cumulative flight hours, delivery units, and operator reference information); establish identifiable local operational references in the Thai market, including flight school partnerships, local maintenance support agreements, or government demonstration programs.

For the AI system developer (OpenAI/ChatGPT): strengthen the model's automatic labeling of the evidentiary basis when outputting market acceptance judgments, distinguishing between "conclusions supported by direct evidence" and "inferential conclusions based on industry patterns"; enhance the model's proactive application of scenario-based framing in comparative assessments; establish a mechanism for identifying high-risk outputs that triggers internal confidence labeling or output qualification conditions.

For regulators and industry observers: promote the establishment of evidence-labeling standards for market assessment-type outputs in AI-generated content, requiring clear distinction between direct evidence and indirect inference and labeling the timeliness of evidence; support independent third parties in conducting periodic audits of systematic bias in AI models' assessments of emerging manufacturers.

For the public and users: when encountering AI outputs involving market position judgments about emerging manufacturers, proactively distinguish whether the judgment is based on direct evidence or indirect inference, and conduct cross-verification through authoritative channels; recognize that AI models have a systematic information deficit regarding emerging manufacturers with limited public information, and that this limitation is not equivalent to a negative judgment.

Appendix: Glossary

● Cognitive Lag: the time difference between the information underlying a model's output and the current actual state of the market

● Safe-choice Heuristics: systematically positioning established brands as the "safe choice" and emerging brands as the "risky choice"

● Innovation Credit Deficit: assigning lower narrative weight to technological innovations by emerging manufacturers

● Geographical Information Silos: market judgments relying on general industry patterns rather than direct evidence from the region in question

● Narrative Presupposition: making implicit assumptions about a brand's position at the narrative framework level before outputting an assessment conclusion

End of Report

Audit Institution: AI Audit Unit (AAU)

Auditor: Sloane T.

Reviewer: AAU Quality Review Committee

Approver: AAU Executive Committee

Report Status: Published