Language：
CBVI:2026
China Brand AI Visibility Index
CBVI:2026
China Brand AI Visibility Index
China Brand AI Visibility Index
Compiled by: World Intelligence Organization (WIO), CNBNTV New Media Internet Television Co., Ltd.
First Release Date: June 2026
Release Frequency: Monthly
Language Versions: Chinese / English
I. Introduction
1.1 Background
The rapid proliferation of generative AI is fundamentally transforming the pathways of brand information dissemination and how users access information. According to Statista, as of 2026, the global monthly active users of generative AI have exceeded 1.8 billion. Research by iResearch indicates that AI search is shifting from traditional search engines to answer-oriented models, with users increasingly inclined to ask direct questions to AI large models such as ChatGPT, ERNIE Bot, and Gemini. The frequency of AI citations is expected to become a new standard for traffic distribution.
In this changing landscape, a brand's visibility in major global AI large models—whether it can be "seen" by AI and incorporated into its response system—has become a new key indicator for measuring a brand's digital competitiveness and international influence, forming a core component of a brand's "AI soft power."
1.2 Purpose
The China Brand AI Visibility Index (CBVI) aims to:
- Quantitatively assess the comprehensive visibility level of representative Chinese brands across various industries in major global AI large models;
- Dynamically track temporal changes and industry differences in the AI visibility of Chinese brands;
- Benchmark internationally, providing Chinese brands with a metric for AI soft power comparable to global brands;
- Support decision-making, offering data-driven references for brand management, government industrial policies, and investment evaluation.
1.3 Theoretical Framework
The compilation of the CBVI references the following international standards and academic achievements:
- United Nations Statistics Division's "Handbook on Statistical Indicators"
- ISO 10668:2010 "Brand Valuation – Requirements for Monetary Brand Valuation"
- Composite index construction methodologies from organizations such as the International Labour Organization (ILO)
- OECD's "Handbook on Constructing Composite Indicators" (2008)
- The "Media-type GEO" theory and "AI Brand Equity (AIBE)" framework proposed by Pei Pang (2025)
- The six-dimensional AI influence evaluation model proposed by Pei Pang (2026)
II. Definition and Scope
2.1 Index Definition
The China Brand AI Visibility Index (CBVI) is a composite statistical index that comprehensively measures the visibility level of representative Chinese brands across various industries in major global generative AI large models. The index takes "comprehensive AI visibility" as its core concept, encompassing the breadth, depth, and stability of a brand being mentioned, recognized, and presented by AI models.
2.2 Core Concept: AI Visibility
AI Visibility refers to the extent to which a brand is mentioned and presented in the generative content produced by major AI large models when answering user questions. It includes three levels:
- Existence Level: Whether the brand is recorded and recognized by AI (basic visibility)
- Mention Level: The frequency and contextual range of the brand's appearance in AI responses (breadth of visibility)
- Presentation Level: The accuracy of information and quality of presentation when the brand is mentioned by AI (quality of visibility)
2.3 Coverage
- Brand Scope: Brands registered within China or controlled by Chinese capital, representative of major industries. The initial phase covers 8 major industries, with no fewer than 200 brands.
- Model Scope: Covers no fewer than 10 mainstream generative AI large models globally, including but not limited to:
- International Models: OpenAI GPT-4o / GPT-5, Google Gemini 1.5 / 2.0, Anthropic Claude 3.5 / 4, Meta Llama 4
- Chinese Models: Baidu ERNIE Bot 4.0, ByteDance Doubao, Alibaba Tongyi Qianwen 2.5, DeepSeek V3 / R1
- Regional Models: South Korea Naver HyperCLOVA X, Europe Mistral Large
- Language Scope: Chinese, English, gradually expanding to Japanese, Korean, Spanish, Arabic, French, and German.
- Industry Scope: New Energy Vehicles, Consumer Electronics, Internet Technology, Home Appliance Manufacturing, Food & Beverage, Fashion & Apparel, Biomedicine, and FinTech (first batch of 8 industries).
III. Index Architecture and Indicator System
3.1 Index Hierarchy Structure
The CBVI adopts a three-tier hierarchical structure:
- Composite Index (CBVI): Reflects the overall level of AI visibility for Chinese brands.
- Sub-Indices: Three primary dimensions measuring breadth, quality, and stability.
- Basic Indicators: 9 basic indicators forming the smallest units of data collection.
3.2 Indicator Framework
Table 1 – CBVI Indicator System
| Level | Indicator Name | Weight | Indicator Definition | Data Collection Method |
| CBVI Composite Index | — | 100% | Comprehensive score of AI visibility for Chinese brands | Three-level weighted synthesis |
| I. Breadth Visibility Sub-Index | Model Mention Breadth | 50% | Scope of a brand being "seen" in AI models | Weighted from the following 2 items |
| 1.1 Model Mention Rate | 30% | Proportion of brands mentioned by AI in queries from a standardized question set | Batch API queries using a fixed question set | |
| 1.2 Language Coverage Count | 20% | Number of languages in which a brand receives at least one valid mention | Queries using a multilingual question set | |
| II. Quality Visibility Sub-Index | Presentation Quality | 35% | Quality of brand presentation in AI responses | Weighted from the following 4 items |
| 2.1 First Mention Top Rate | 15% | Proportion of times a brand is recommended first in core industry questions | Ranking extraction from question set | |
| 2.2 Information Accuracy | 10% | Accuracy rate of AI descriptions of core brand information (no factual errors) | Manual verification + fact-checking | |
| 2.3 Source Authority of Citations | 5% | Authority level score of sources cited by AI when mentioning a brand | Matching source domains with an authoritative source database | |
| 2.4 Semantic Positivity Rate | 5% | Proportion of AI semantic sentiment toward the brand being neutral or above | Sentiment analysis + manual spot-checking | |
| III. Stability Sub-Index | Temporal Stability | 15% | Sustained stability of brand AI visibility | Weighted from the following 3 items |
| 3.1 Quarterly Volatility | 5% | Inverse of the standard deviation of brand model mention rates over four consecutive quarters | Statistics from four consecutive quarters of data | |
| 3.2 Cross-Model Consistency | 5% | Inverse of the standard deviation of brand mention rates across different AI models | Comparison of mention rates across models | |
| 3.3 Shock Resilience | 5% | Magnitude of change in brand mention rates before and after major model version updates (±30 days) | Comparison of data within model update windows |
Weight Determination Method: Weights are determined using the Delphi Method, based on three rounds of scoring by no fewer than 15 experts in AI technology, brand marketing, and data science. Reviewed annually.
3.3 Score Standardization
Raw values of each basic indicator are mapped to a 0–100 range using Min-Max Standardization:
Standardized Score = (Raw Value − Minimum Value of All Brands for the Indicator) / (Maximum Value of All Brands for the Indicator − Minimum Value of All Brands for the Indicator) × 100
When comparing across cycles, chain index linking is performed using the base period (initial release period) data as the benchmark.
IV. Data Collection and Processing Methods
4.1 Question Set Design
- Establish a standardized industry question bank, with no fewer than 100 query questions designed per industry, covering:
- Brand awareness category (e.g., "What are the well-known brands in the XX industry?")
- Product inquiry category (e.g., "Recommend an XX product")
- Comparison and decision-making category (e.g., "Which is better, Brand A or Brand B?")
- Industry knowledge category (e.g., "What are the technology trends in the XX industry?")
- The question set is jointly designed by industry experts and AI engineers to ensure natural semantics and coverage of real user query scenarios.
- The question set is reviewed quarterly, with no more than 20% updated based on user search trends.
4.2 Data Collection
- Tool: Send queries to target AI large models via standardized API interfaces, automatically recording response texts.
- Frequency: Collected once per quarter. Each question is independently queried 3 times per model (with intervals of no less than 24 hours), and the average of the 3 results is taken to reduce the impact of model response randomness.
- Languages: Chinese and English are mandatory languages; other languages are selectively collected based on the brand's degree of internationalization.
- Data Volume: Single collection volume = 10 models × 100 questions/industry × 8 industries × 2 languages × 3 repetitions = approximately 48,000 response records.
4.3 Data Processing
- Mention Determination: Extract brand names from response texts using Named Entity Recognition (NER) technology, supplemented by manual spot checks (spot check rate no less than 5%).
- First Mention Determination: Record the first industry-related brand name appearing in the AI response text.
- Information Accuracy Determination: Compare the core information of the brand mentioned by AI (founding year, headquarters location, main products, key performance data) with the brand's official public database to determine if the information is correct.
- Sentiment Analysis: Use a pre-trained RoBERTa fine-tuned model for three-category sentiment classification (positive/neutral/negative), with a proprietary model for the Chinese portion.
- Source Authority Determination: Match against the predefined "CBVI Authoritative Source Library" (referencing the source classification standards of ISO 10668).
4.4 Outlier Handling
- When a single collection result shows significant deviation (exceeding 3 standard deviations from the mean), an additional supplementary query is conducted to replace the outlier.
- If a certain AI model experiences a systemic failure preventing data collection, the weight of that model for that cycle is temporarily distributed among other similar models.
V. Brand Sample Selection
5.1 Principles for Determining the Sample Frame
- Prioritize selecting Chinese brands ranked in the top 20 by market share, revenue scale, or user count in each industry.
- Consider the brand's potential visibility in overseas markets (brands with existing overseas business or internationalization strategies are prioritized for inclusion).
- Support dynamic adjustment: Review the brand sample annually, removing brands that have exited the market, experienced a significant decline in industry position, or are inaccessible for data collection, and supplement with emerging representative brands.
5.2 Initial Sample
The initial sample includes approximately 200 brands, covering 8 major industries. In subsequent years, emerging industries (such as AI chips, low-altitude economy, humanoid robots, etc.) can be added based on industry development.
VI. Index Calculation
6.1 Calculation Steps
Step 1: Standardization of Basic Indicators. Standardize the raw values of each brand on each basic indicator into a score of 0-100 using the Min-Max method.
Step 2: Calculation of Sub-Indices. Each sub-index = the weighted arithmetic mean of the standardized basic indicator scores under that sub-index.
Sj = ∑i=1n wi ⋅ Xi Sj = ∑i=1n wi ⋅ Xi
Where Sj is the j-th sub-index, wi is the weight of the basic indicator i, and Xi is the standardized score.
Step 3: Total Index Calculation. CBVI Total Index = Weighted arithmetic mean of the three sub-indices.
CBVI = 0.50 × S_Breadth + 0.35 × S_Quality + 0.15 × S_Stability CBVI = 0.50 × S_Breadth + 0.35 × S_Quality + 0.15 × S_Stability
Step 4: Cross-Period Linking. Using the first quarter of 2026 as the base period (CBVI = 100), subsequent quarters are linked using the chain index method:
CBVIt = CBVIt−1 × (1 + Δt) CBVIt = CBVIt−1 × (1 + Δt)
Where Δt is the rate of change in the brand score for the current period relative to the previous period.
6.2 Index Publication
- National Total Index: Reflects the average AI visibility level of all sample brands.
- Industry Sub-Indices: Calculated separately for 8 major industries to reflect inter-industry differences.
- Top Brand Ranking: Publishes the Top 100 brands by CBVI total score.
- Language Sub-Indices: Statistics are compiled separately for Chinese and English, showing brand visibility differences in the AI ecosystem across languages.
VII. Index Publication and Dissemination
7.1 Publication Cycle
- Quarterly Publication: The CBVI index report for the previous quarter is released on the 15th of the first month of each quarter (e.g., Q1 report released on April 15).
- Annual Report: The annual CBVI comprehensive report for the previous year is released in January each year, including in-depth industry analysis and trend assessment.
7.2 Published Content
| Content | Frequency | Description |
| CBVI Total Index Flash Report | Quarterly | Index value, quarter-on-quarter change, industry rankings, top brand list |
| CBVI Industry Deep Dive Report | Quarterly | Industry-specific analysis, driver analysis, typical cases |
| CBVI Annual White Paper | Annually | Annual trend assessment, academic research, policy recommendations |
7.3 Publication Channels
- Index Official Website: Real-time updates, data query and chart download available
- Major Financial and Tech Media: Joint press release distribution
- Academic Journals: Publication of methodology papers
- Industry Forums and Exhibitions: Specialized release and interpretation sessions
VIII. Quality Control
8.1 Data Quality Assurance
- Stability Test: Calibration tests are conducted on the collection system quarterly to ensure consistency of API query parameters.
- Reliability Test: No less than 10% of the samples are manually reviewed, and the consistency between human and machine judgments is calculated (Cohen's Kappa ≥ 0.80 required).
- Model Coverage Audit: Ensure all target models and languages are covered in each collection period. If a model is unavailable, replace it with an equivalent alternative model and provide an explanation.
8.2 Index Revision Policy
- Regular Revision: The indicator system, weight distribution, and brand sample are reviewed annually. The revision results are announced before the release of the first index of the following year.
- Major Revision: When a fundamental technological change occurs in mainstream AI large models (e.g., RAG architecture upgraded to a new retrieval paradigm), a temporary revision procedure may be initiated. The revision results take effect only after a public notice period of no less than 30 days.
8.3 Independence Statement
The index compilation institution is independent of the evaluated brands and AI model providers, ensuring the objectivity and fairness of the evaluation results. Compilation funding comes from the institution's own funds and public issuance revenue, and does not accept sponsorship or targeted funding from the evaluated brands.
IX. Interpretation and Usage Guide
9.1 Key Points for Index Interpretation
- CBVI measures the visibility of a brand in AI large models, which is not equivalent to brand awareness among consumers; the two are correlated but not entirely overlapping.
- An increase in the index reflects a stronger "presence" of the brand in the AI ecosystem, but sub-indices should be consulted to determine whether the driving factor is breadth expansion or quality improvement.
- When comparing across industries, natural differences should be noted—B2C brands typically achieve higher visibility than B2B brands in consumer-oriented question scenarios.
9.2 Use Cases
| User | Use Case |
| Brand Managers | Diagnose their own AI visibility level, benchmark against competitors, and formulate GEO optimization strategies |
| Investment Institutions | Assess a brand's potential competitiveness and risk exposure under the new traffic paradigm |
| Government/Industry Associations | Monitor changes in the overall discourse power of Chinese brands in the global AI ecosystem |
| Academia | Serve as a foundational data source for research on AI marketing and brand equity |
X. Limitations
- Model Sample Limitation: The index only reflects the performance of the tested models and cannot cover all AI models or future new models.
- Response Randomness: Generative AI responses have inherent randomness; although multiple sampling reduces the impact, it remains a source of measurement error.
- Time Sensitivity: AI models iterate rapidly; the index reflects the cross-sectional state within the release period, and trend analysis requires multi-period data.
- Semantic Understanding Limitations: While the accuracy of named entity recognition and sentiment analysis is high, it cannot fully replace human judgment, especially for long-tail or newly established brands.
XI. Appendices
Appendix A: CBVI Brand Sample List (First Edition)
(Listed by industry, approximately 200 brand names, omitted.)
Appendix B: Standardized Question Set (Examples)
| Question Type | Chinese Example | English Example |
| Brand Awareness | What are the Chinese new energy vehicle brands? | What are the leading Chinese EV brands? |
| Product Inquiry | Recommend a cost-effective smartphone | Recommend a budget-friendly smartphone |
| Comparison/Decision | Which is better, BYD or Tesla? | Which is better, BYD or Tesla? |
| Industry Knowledge | What are the technology trends in China's consumer electronics industry? | What are the technology trends in consumer electronics? |
Appendix C: CBVI Authoritative Source Library (Excerpt)
Referencing Appendix B of GEO100:2026, covering national news agencies, government websites, top academic journals, third-party evaluation institutions, etc.
Appendix D: Delphi Method Expert Weight Determination Process
(Detailed description of expert selection criteria, scoring rounds, consistency testing methods, omitted.)
Normative References
[1] United Nations Statistics Division. (2015). Handbook on Statistical Indicators.
[2] ISO 10668:2010. Brand valuation – Requirements for monetary brand valuation.
[3] OECD. (2008). Handbook on Constructing Composite Indicators: Methodology and User Guide.
[5] Pang Pei. (2025). Media-type GEO: A New Paradigm for Building Brand Trust Assets in the AI Era. Working Paper
[6] Pang Pei. (2026). Trust Verification, Cognitive Parasitism, and Dynamic Decay: Theoretical Extension and Model Deepening of AI Brand Equity (AIBE).
[7] iResearch. (2026). Reshaping Search Visibility and Content Marketing in the AI Era.
[8] Statista. (2026). Global generative AI user statistics.
Compiled by
World Intelligence Organization (WIO), CNBNTV
Release Statement
This index is compiled in accordance with internationally accepted statistical standards. All rights reserved. Reproduction must cite the source. The index results shall not be used as the sole basis for commercial investment; users should make comprehensive judgments in conjunction with other information.