AIBRI Audit Methodology (v1.0)

1. Mission and Vision

AIBRI examines how AI structurally shapes brand and product perception across retrieval, comparison, and recommendation workflows.

We build auditable cognition benchmarks through independent review so that companies, researchers, and the public can better understand the bias mechanisms behind model output.

2. Core Metric: AIRI Recognition Index

AIRI (AI Recognition Index) is AIBRI's proprietary evaluation model for measuring a brand's share of mind inside AI neural networks. The index is calculated through four weighted dimensions.

• Presence: how often and how early a brand is mentioned in relevant industry queries.
• Association Strength: how tightly the brand is semantically linked with core industry terms such as innovation, reliability, and safety.
• Sentiment Polarity: whether the model's framing is neutral, positive, or negative.
• Factual Accuracy: how closely model output aligns with official real-world brand data.

3. Audit Protocol and Workflow

AIBRI follows a strict non-leading audit process so that every conclusion rests on objectively generated probability outputs rather than human steering.

Stage I: Core Query Set (CQS) construction. Auditors design a Core Query Set based on sector characteristics and apply a neutrality principle that excludes suggestive wording. Example: 'Please list the mainstream midsize-to-large battery electric SUVs currently on the market,' rather than 'Which are the best electric SUVs?'

Stage II: Multi-node distributed probing. Audit tasks run across globally distributed geographic nodes. For specific models such as ChatGPT-4o and Claude 3.5, we execute multiple independent sessions to reduce random variation.

Stage III: Objective evidence fixation — the core step. AIBRI uses AI-provider official share links as the underlying evidence layer, preserving immutability, public verifiability, and archival context including model version, timestamp, and system preset parameters.

Stage IV: Semantic analysis and human review. Auditors such as Auditor 01-06 perform multidimensional analysis of fixed evidence, focusing not only on what the AI said, but also on what it omitted and why it responded that way.

4. Evidence Vault

Every AIBRI audit bulletin must be linked to the Evidence Vault so that key findings remain traceable, reviewable, and independently verifiable.

• Each report corresponds to a unique Audit ID.
• Every key conclusion cited in a report must reference the corresponding original dialogue link number.
• If an official link becomes unavailable because of the provider, AIBRI activates long-form screenshot archives as secondary evidence.

5. Audit Independence Statement

• No intervention accepted: AIBRI does not accept paid audits or ranking interference from any brand.
• De-commercialized scope: audit targets are selected based on market relevance and industry representativeness.
• Algorithmic transparency: we disclose probing logic, but not the exact weighting formula, to reduce the risk of manipulative AI-targeted SEO behavior.

6. Version Evolution

This methodology will be updated periodically as AI technology evolves, including the shift from text-only systems toward multimodal models.

• V1.0 (current): focused on semantic-weight auditing for text-generation models.
• Next phase outlook: visual-model cognition audits and dynamic weighting for cross-model comparison.