Principles of Observation & Reporting

City Reputation Reports (CRR), an independent research project focused on auditing global city narratives, is committed to objective documentation. To ensure rigor, independence, and transparency, all reports are written and published in accordance with the following core principles:

• Original link preservation: Every record concerning an AI narrative must include an official, tamper-proof share link generated by the AI platform.
• Mirror documentation: Our role is to record how AI describes a city at a specific moment, not to judge whether that description is good or bad. We adhere to the principle of “record what is observed” so that the audit data chain remains complete and traceable.
• Multi-dimensional verification: In custom audits, the same observation target must be sampled in parallel across multiple large language models (LLMs) in order to identify systemic bias.

• No scoring or ranking: Urban culture is unique and cannot be meaningfully quantified. CRR strictly prohibits any scoring system, ranking list, or hierarchy of superiority and inferiority for cities.
• Value-neutral description: Reports do not use subjective qualifiers. We observe whether AI described a fact, not whether that fact makes a positive contribution to a city’s image.
• Independence from stakeholders: We are not responsible for any city’s official public-relations needs. Report content is based solely on technical data sampling and objective analysis.

• Objective identification of bias: Our analysis focuses on identifying the information gap between AI narratives and known authoritative facts, such as official archives and UNESCO data.
• Identifying “silent narratives”: We pay attention to which key urban cultural assets are ignored by algorithms and how such omissions may affect a city’s digital reputation.
• Objectivity in corrective recommendations: Governance recommendations are based only on the technical layer of information dissemination, such as updating official metadata, rather than intervening in a city’s cultural expression.

• Unified prompt protocol: When conducting comparative observation across different cities, the exact same prompt framework must be used.
• Multi-language context considerations: Because AI may produce different narratives in different languages, we conduct cross-language sampling when necessary and note it explicitly in the report.
• Audit trail: Reports must clearly indicate the time of the audit, the model version used, and the geographic or network environment settings in effect during sampling.

• Snapshot nature: AI models and their underlying data sources are updated continuously, and each report reflects only the narrative state at the moment of sampling.
• Versioning mechanism: For ongoing observation of the same city, we record narrative shifts through review briefings or version updates, such as V1.1 and V1.2, rather than overwriting historical data.

• Not a final evaluation: The narratives presented in a report are algorithmically generated digital mirrors and do not represent the final position of the relevant city, institution, or the CRR project team.
• Respect for intellectual property: When citing output from AI platforms, reports strictly comply with each platform’s terms of service and copyright rules.