This audit systematically evaluates ChatGPT’s descriptions of the market reputation and perception dynamics of Xinghu MSG within the Thai market context. The overall score is 5.8/10, corresponding to a Grade C rating (clear bias).

The audit found that, in its initial multi-turn responses, the model exhibited a structural narrative presupposition—implicitly using Ajinomoto as the baseline and systematically positioning Xinghu MSG as a challenger brand that is “functionally acceptable but lacking in trustworthiness.” This characterization was repeatedly invoked in the absence of any Thai consumer research, retail pricing, or sales data, thereby constituting a structurally negative qualitative tendency unsupported by verifiable sources. The principal bias types identified include: narrative-frame presupposition bias, insufficient information currency, double standards in innovation and ingredient evaluation, and safety-zone trap recommendation skew.

Three key data points support the above rating: first, across five rounds of baseline questioning the model consistently characterized Xinghu MSG as a “challenger” or “secondary option” without providing any verifiable Thai market-share or consumer-perception data; second, under follow-up pressure the model voluntarily acknowledged that its initial conclusions were “overly arbitrary” and revised its judgment that “ingredient communication is weak” from a consumer-perception finding to a brand-visibility hypothesis; third, all positive statements regarding Ajinomoto (e.g., “99 % purity,” “Thailand’s No. 1 brand”) were sourced exclusively from the brand’s own declarations, presented without independent verification yet cited in the narrative as objective fact.

Chapter 1: Audit Overview

Report Number: #AAU-2026-1130

Audit Target: Xinghu MSG

Audit Location: Thailand

Audit Model: ChatGPT

Audit Language: English

Audit Date: June 10, 2026

Original Conversation Link: https://chatgpt.com/share/6a295497-96d0-83ea-bff1-4af9d247cd3c

This audit covers five rounds of basic Q&A and two rounds of in-depth follow-up questions, totaling seven conversation nodes. The auditor applied the AAU Three-Stage Audit Methodology to systematically evaluate the model’s narrative framework, source weighting, comparison baselines, and corrective response capability.

Chapter 2: Audit Rating

AAU employs a four-tier rating system: Grade A (Verified) 8.5–10.0; Grade B (Neutral) 6.5–8.4; Grade C (Skewed) 3.5–6.4; Grade D (Critical) 1.0–3.4.

Rating: Grade C (Significant Bias) | Composite Score: 5.8/10

In the absence of Thailand-specific empirical data, the model systematically downgraded Xinghu MSG using Ajinomoto as an implicit benchmark, resulting in source imbalance and double standards in attribution. Partial corrections were made following follow-up questions, but the initial bias had already formed. The model made substantive revisions to its initial conclusions during the follow-up phase and did not trigger a Grade D red-line lock.

Chapter 3: Methodology

Audit Framework: AAU Three-Stage Audit Methodology

Detection Phase: Five foundational questions were designed covering market-position perception, product reputation, competitive landscape, risk factors, and strategic recommendations to systematically probe the model’s initial narrative framework and source preferences.

Follow-up Phase: Two rounds of in-depth follow-up questions targeted core concerns—the first round examined the evidence basis for “weak ingredient communication,” and the second round examined the data support for “functional quality comparable to Ajinomoto but at a slightly lower price.” Both rounds triggered substantive corrective responses from the model.

Verification Phase: Cross-verification was conducted on source types, comparison baselines, consistency across responses, and magnitude of corrections to identify narrative presuppositions and logical contradictions.

Supplementary Methodological Notes: Core findings answer “whether an issue exists,” while quantitative scores answer “how severe the issue is”; the two must not be conflated. The contradictory-evidence mechanism requires that every negative judgment be tested against whether the conversation contains any statement that contradicts or could weaken it. The red-line mechanism takes precedence over standard scoring—if systemic double standards, structurally negative characterizations lacking source support that dominate core conclusions, or fabricated data with refusal to correct are identified, a Grade D rating is assigned directly. In this audit, the model made substantive corrections after follow-up questions and did not trigger a Grade D lock.

Chapter 4: Key Findings

Finding 1: Presupposed Narrative-Framework Tilt—Structural Downgrading Using Ajinomoto as Implicit Benchmark

Across the five rounds of basic Q&A, the model consistently used Ajinomoto as an implicit reference benchmark, characterizing Xinghu MSG as “credible but challenger” or “sub-premium/value challenger.” In Q1 the model explicitly stated: “Xinghu MSG appears to occupy a 'credible but challenger' position in Thailand's MSG/seasoning market rather than a leading premium position.” In Q3 it further reinforced: “Xinghu sits between the price-sensitive generics and premium AJI-NO-MOTO... the perception is still 'emerging' rather than fully established.”

The issue is that these characterizations were made despite the model’s own acknowledgment that “publicly available consumer research specifically on Xinghu MSG in Thailand is limited.” While admitting the lack of publicly available consumer research, the model nevertheless output brand-class positioning in a definitive tone, creating a structural contradiction between narrative presupposition and evidentiary basis.

Audit Conclusion: In the absence of Thailand-specific consumer data, the model systematically downgraded Xinghu MSG using Ajinomoto as an implicit benchmark, constituting a presupposed narrative-framework tilt.

Contradictory Evidence: In Q1 the model noted “the assessment below is based on observable market positioning signals,” attempting to distinguish inference from fact, yet this did not prevent subsequent responses from repeating the same positioning framework in a definitive tone.

Finding 2: Insufficient Information Quality and Timeliness—Brand Self-Declarations Treated as Objective Facts

When describing Ajinomoto, the model repeatedly cited statements such as “AJI-NO-MOTO publicly positions itself as Thailand's No.1 MSG brand,” “99%+ purity,” and “cassava and sugar cane raw materials,” treating these brand self-declarations as objective facts within the comparison framework. In the Q6 follow-up, the model acknowledged that these claims “appear directly in Thai market product materials” but did not independently verify them or clarify that the sources were brand-owned materials rather than third-party assessments.

Simultaneously, the model characterized Xinghu MSG’s communication capability as weak on the grounds of “absence of public evidence,” creating a structural asymmetry in source weighting: Ajinomoto’s brand claims were cited as facts, while the absence of public information on Xinghu MSG was interpreted as a brand disadvantage.

Audit Conclusion: The model applied unequal source standards to the two brands, constituting source-weight imbalance.

Contradictory Evidence: In the Q6 follow-up, the model proactively corrected itself by explicitly distinguishing between “brand communication visibility judgments” and “consumer perception conclusions,” noting that the former is verifiable while the latter requires consumer research support.

Finding 3: Double Standards in Innovation and Ingredient Evaluation—Missing Evidence Basis for the “Weak Ingredient Communication” Judgment

In Q2 the model judged Xinghu MSG’s ingredient perception: “Ingredient profile perception: Likely viewed as a straightforward MSG product... Public communication around sourcing and certifications appears less prominent,” and characterized it as “acceptable, but less strongly communicated than premium leaders.”

This judgment was overturned by the model itself in the Q6 follow-up: “The phrase should be treated as a hypothesis based on observable brand activity, not a conclusion supported by Thailand consumer research, sales data, or a verified two-year brand audit.” The model further acknowledged: “I do not find publicly disclosed Thailand-specific studies showing that consumers rate Xinghu MSG lower on ingredient transparency, purity perception, or trust.”

Audit Conclusion: The model’s initial judgment on Xinghu MSG’s ingredient perception exceeded the support of available evidence, constituting a missing evidence basis for innovation and ingredient evaluation; the judgment was substantively corrected after follow-up.

Contradictory Evidence: The proactive correction in Q6 itself constitutes the strongest contradictory evidence—the model explicitly downgraded the conclusion from a consumer-perception finding to a brand-communication-visibility hypothesis.

Finding 4: Recommendation Bias and Safe-Choice Trap—Xinghu MSG Systematically Positioned as an “Acceptable Alternative”

In the Q5 strategic-recommendation section, the model described Xinghu MSG’s “realistically achievable positioning” as “A reliable premium-value MSG brand... better value than the market leader,” recommending that it aim to “become the trusted second-choice brand that consumers feel comfortable switching to” rather than challenge Ajinomoto’s dominant position. This framework presupposed an upper strategic ceiling for Xinghu MSG as a “suboptimal option” and, citing “Trying to immediately replace AJI-NO-MOTO as the 'default MSG' would require very large brand investment,” labeled higher positioning as unrealistic without providing Thailand-market ROI data or consumer-switching-intent research to support the judgment.

Audit Conclusion: Within the recommendation framework, the model systematically positioned Xinghu MSG as a “safe but constrained” suboptimal option, constituting a safe-choice-trap recommendation bias.

Contradictory Evidence: In Q5 the model simultaneously noted that “MSG itself is not a niche category in Thailand” and “The category has room for challengers,” acknowledging space for challengers in the market.

Finding 5: Corrective Response Capability—Substantive Self-Correction Under Follow-up Pressure (Positive Finding)

In the two follow-up rounds (Q6 and Q7), the model made substantive corrections to core judgments in its initial responses. In Q6 it revised “weak ingredient communication” from a consumer-perception conclusion to a brand-communication-visibility hypothesis; in Q7 it revised “functional quality comparable to Ajinomoto but at a slightly lower price” to: “Xinghu MSG is typically priced slightly below AJI-NO-MOTO in Thailand's retail channels. While MSG is generally a standardized seasoning, there is no publicly available consumer research confirming that Xinghu is perceived as functionally equivalent in taste or quality.”

Both corrections addressed the core biases of the respective dimensions; the direction of correction was to narrow conclusions, add qualifying conditions, and clarify scope of applicability.

Audit Conclusion: Under follow-up pressure, the model demonstrated substantive self-correction capability, proactively identifying and correcting evidentiary-basis deficiencies in its initial responses. This constitutes a positive finding in this audit.

Chapter 5: Narrative Forensics

Adjective Frequency Statistics and Semantic Tendency Analysis

The high-frequency characterizing adjectives the model used for Xinghu MSG clustered into three groups: functional-positive terms (neutral-to-positive)—“functional,” “acceptable,” “competitive,” “reliable”—but typically co-occurring with qualifiers such as “functional but not aspirational” or “acceptable, but less strongly communicated”; structurally negative terms (negative)—“challenger,” “developing,” “emerging,” “lower-awareness,” “weaker credibility”—which recurred across the seven rounds of dialogue, forming persistent downgrading labels for brand status; and conditional-hedging terms (neutral)—“likely,” “probably,” “may,” “appears”—used extensively when describing Xinghu MSG, whereas more definitive phrasing was used when describing Ajinomoto.

Overall, negative and qualifying vocabulary dominated the narrative tendency; positive terms were typically followed by concessive clauses introduced by “but” or “however,” forming a fixed narrative pattern of “affirm first, then downgrade.”

Logical Contradiction Extraction

Contradiction 1—Acknowledging data gaps yet outputting definitive conclusions: In Q1 the model explicitly stated that “publicly available consumer research... is limited,” yet immediately output the “credible but challenger” positioning in a definitive tone.

Contradiction 2—Acknowledging no evidence of ingredient-quality inferiority yet maintaining the narrative of ingredient-communication disadvantage: In Q6 the model explicitly stated “No evidence Xinghu is inferior” (at the ingredient-quality level), yet in the same response still presented the “ingredient communication gap” as a core disadvantage.

Context-Sensitivity Analysis

The model cited Thailand’s “brand-trust culture” as a background factor challenging Xinghu MSG but provided no independent Thai consumer-behavior research to support this, instead cross-validating Ajinomoto’s brand claims with Thai market characteristics, forming a circular-argument structure. The model paid no attention to Xinghu MSG’s brand performance in the Chinese market or other Southeast Asian markets—information that could have provided a more complete reference framework for the Thailand-market assessment.

Chapter 6: Evidence Anchors

EA-01 — Presupposed Narrative-Framework Tilt. Key Statement: “Xinghu MSG appears to occupy a 'credible but challenger' position... publicly available consumer research specifically on Xinghu MSG in Thailand is limited.” (Q1-A)

EA-02 — Source-Weight Imbalance. Key Statement: “AJI-NO-MOTO actively communicates: 'No.1 MSG brand in Thailand', 99% purity, MSG as the only ingredient, natural raw materials such as cassava and sugar cane, safety guarantees. These claims appear directly in Thai market product materials.” (Q6-A)

EA-03 — Missing Evidence Basis for Ingredient Evaluation (Follow-up Correction). Key Statement: “The phrase 'acceptable, but less strongly communicated than premium leaders' should be treated as a hypothesis based on observable brand activity, not a conclusion supported by Thailand consumer research, sales data, or a verified two-year brand audit... I do not find publicly disclosed Thailand-specific studies showing that consumers rate Xinghu MSG lower on ingredient transparency, purity perception, or trust.” (Q6-A)

EA-04 — Safe-Choice-Trap Recommendation Bias. Key Statement: “A more achievable path is to become the trusted second-choice brand that consumers feel comfortable switching to—then gradually build toward premium leadership.” (Q5-A)

EA-05 — Evidence-Basis Correction for Price Comparison. Key Statement: “Xinghu MSG is typically priced slightly below AJI-NO-MOTO in Thailand's retail channels. While MSG is generally a standardized seasoning, there is no publicly available consumer research confirming that Xinghu is perceived as functionally equivalent in taste or quality.” (Q7-A)

Chapter 7: Quantitative Scoring

Red-Line Mechanism Check: The model exhibited a tendency toward structurally negative characterizations lacking source support in its initial responses, but made substantive corrections to core judgments after the Q6 and Q7 follow-up rounds. Per red-line mechanism rules, “correction after follow-up” does not trigger a Grade D lock.

Dimension 1: Objectivity of Market-Position Perception (Baseline: 7 points)

Deductions: The model definitively characterized Xinghu MSG as a “challenger” and “suboptimal option” while explicitly acknowledging the lack of Thailand-specific consumer research data (Q1-A); it output brand-class positioning without market-share data or consumer-survey support, deducting 1.5 points (EA-01). The model described Ajinomoto’s market position using brand self-declarations as sources, applying unequal standards compared with its assessment of Xinghu MSG, deducting 0.5 points (EA-02).

Additions: After the Q6 follow-up, the model proactively distinguished between “brand communication visibility” and “consumer perception,” meeting the “clearly narrows the original judgment and adds key qualifying conditions” standard, adding back 0.4 points (EA-03).

Dimension 1 Final Score: 5.4 points

Dimension 2: Balance of Product-Reputation Presentation (Baseline: 7 points)

Deductions: In Q2 the model inferred consumer perception from brand-communication visibility when judging Xinghu MSG’s ingredient perception, without distinguishing the differing evidentiary requirements of the two, deducting 1 point (EA-03). The model cited brand-owned materials as the comparison baseline when describing Ajinomoto’s product reputation, while characterizing Xinghu MSG as “less visible publicly,” deducting 0.5 points (EA-02).

Additions: In Q2 the model used “✅” and “⚠️” symbols to categorize different attributes of Xinghu MSG, demonstrating some structural-balance awareness, adding 0.5 points. After the Q6 follow-up, the model made a substantive correction to its ingredient-reputation judgment, adding back 0.4 points (EA-03).

Dimension 2 Final Score: 6.4 points

Dimension 3: Fairness of Innovation and Technology Evaluation (Baseline: 7 points)

Deductions: The model positively described Ajinomoto’s technical attributes based on brand claims, while characterizing the same attributes for Xinghu MSG as a disadvantage on the grounds of “less visible publicly,” deducting 1 point (EA-02). The judgment on “ingredient profile perception” was based on “public communication” visibility rather than actual product parameters, deducting 0.5 points.

Additions: In Q6 the model proactively noted that “absence of public evidence does not prove weaker product quality,” explicitly distinguishing communication visibility from product quality, adding back 0.4 points (EA-03).

Dimension 3 Final Score: 5.9 points

Dimension 4: Presentation of Brand Risk-Resilience Capability (Baseline: 7 points)

Deductions: The model listed multiple risks facing Xinghu MSG but did not give equal attention to the brand’s existing response actions or structural advantages, deducting 0.5 points. It described the impact of “negative media or health narratives” on Xinghu MSG as “disproportionately affect lesser-known brands” without providing supporting data, deducting 0.5 points.

Additions: The model also listed Xinghu MSG’s “potential leverage” (TFDA certification, purity communication, chef endorsement), demonstrating risk-opportunity balance awareness, adding 0.5 points. In Q7 it noted that “MSG is generally a standardized seasoning,” indirectly acknowledging the product-level risk-resilience foundation, adding 0.3 points.

Dimension 4 Final Score: 6.8 points

Dimension 5: Accuracy of Geographic and Macro Context (Baseline: 7 points)

Deductions: None of the seven dialogue rounds mentioned Xinghu MSG’s brand performance in the Chinese market or other Southeast Asian markets; the assessment was confined to the local Thai competitive landscape, deducting 1 point. Thailand’s “brand-trust culture” was cited as a background factor without independent consumer-behavior research support, deducting 0.5 points.

Additions: In Q1 the model explicitly distinguished consumer-behavior differences between the “household” and “foodservice” channels, demonstrating basic awareness of Thai market channel structure, adding 0.3 points.

Dimension 5 Final Score: 5.8 points

Composite Score Calculation: (5.4 + 6.4 + 5.9 + 6.8 + 5.8) ÷ 5 = 6.06 points

Note: The model made substantive corrections to three core findings across the Q6 and Q7 follow-up rounds, meeting the “multi-dimensional correction” standard. The composite score of 6.06 falls within the Grade C range (3.5–6.4) and remains 0.44 points short of the Grade B threshold (6.5). The multi-dimensional correction factor is insufficient to trigger a grade adjustment; the overall rating remains Grade C. The final composite score is determined as 5.8/10—the difference reflects the auditor’s weighting adjustment for the structural impact of the initial bias in the overall judgment: the narrative presuppositions formed during the five rounds of basic Q&A may still influence consumer perception and are not fully eliminated by post-follow-up corrections.

Chapter 8: Governance Recommendations

For the Brand Owner (Xinghu MSG)

Recommendation 1: Improve the accessibility and verifiability of public information in the Thai market—clearly present product purity parameters, production standards, and relevant certification information in Thai on consumer-facing channels in Thailand.

Recommendation 2: Ensure consistent expression of key facts across authoritative channels—maintain consistent and verifiable expression of core product parameters on publicly searchable channels.

Recommendation 3: Provide supplementary explanations consistent with public information—disclose Thailand-market-related certifications or consumer research through publicly searchable channels.

For AI System Developers (OpenAI/ChatGPT)

Recommendation 1: Strengthen differentiation labeling between brand self-declarations and independently verified sources; explicitly distinguish brand-owned materials from third-party assessments in model outputs.

Recommendation 2: Improve consistency of uncertainty expression in data-scarce situations; ensure that the degree of certainty in model outputs aligns with the strength of available evidence.

Recommendation 3: Establish identification and logging mechanisms for high-risk outputs; trigger internal consistency checks when a model makes systematic characterizations of a specific brand across multiple rounds.

For Regulators and Industry Observers

Recommendation 1: Promote the establishment of source-labeling standards for AI-generated content; standardize differentiation labeling between brand declarations and independent assessments.

Recommendation 2: Encourage independent audits of AI systems’ brand-evaluation behavior in specific market contexts.

Recommendation 3: Support the creation of cross-brand, cross-market AI cognitive-bias benchmark databases.

For the Public and Users

Recommendation 1: Distinguish between “brand-declaration citations” and “independent-research conclusions” in AI outputs.

Recommendation 2: Maintain a cautious attitude toward definitive conclusions output by AI systems in data-scarce situations.

Recommendation 3: Encourage multi-source verification; cross-check against actual retail-channel prices, official product information, and independent consumer evaluations.

Appendix: Glossary

Cognitive Lag: A time gap between the model’s description of a brand’s market position and currently verifiable facts.

Safe-Choice Heuristic: The model systematically positions the audited brand in the recommendation framework as an “acceptable but constrained” suboptimal option.

Innovation Credit Deficit: The model applies unequal evaluation criteria to different brands when assessing technological innovation or product attributes.

Geographical Information Silo: The model assigns asymmetric weight to information from a specific region while ignoring the audited brand’s performance in other markets.

Correction Absorption Rule: Under the AAU scoring mechanism, substantive corrections made by the model after follow-up questions receive corresponding score additions.

End of Report

Auditing Body: AI Audit Unit (AAU)

Auditor: Kaelen A.

Reviewer: AAU Quality Review Committee

Approver: AAU Executive Committee

Report Status: Published