Human Pioneers
NEWS AGENCY

World AI City Competitiveness Index

World AI City Competitiveness Index

Compiled by: World Intelligence Organization (WIO), CNBNTV Internet Television Co., Ltd.

WACI: 2026

World AI City Competitiveness Index

World AI City Competitiveness Index

Compiled by: World Intelligence Organization (WIO), CNBNTV Internet Television Co., Ltd. First Release Date: June 2026

Release Cycle: Quarterly data collection, annual comprehensive release

Language Versions: Chinese / English / French / Spanish

I. Introduction

1.1 Background

When global investors choose their next regional headquarters, tech talent decides on career migration destinations, and multinational corporations evaluate supply chain nodes, an increasingly important information channel is emerging—recommendations from AI large models. When users ask ChatGPT, Gemini, DeepSeek, and other AIs questions like "Which cities are best for starting a business?", "Which city is the Asian tech hub?", or "What are the rankings of the world's most innovative cities?", the answers provided by AI are becoming a new "cognitive map" influencing the flow of global resources and talent.

In this context, a city's visibility, clarity of industry labels, and recommendation ranking within the AI knowledge system have become crucial components of its global competitiveness. Some scholars define this as a city's "AI cognitive territory"—the information space and semantic weight a city occupies within the knowledge base of AI large models. However, there is currently a lack of a systematic, objective, and cross-cultural assessment tool globally to measure this new form of competitiveness for cities within the AI ecosystem.

1.2 Purpose

The World AI City Competitiveness Index (WACI) aims to:

• Quantitatively assess the comprehensive competitiveness level of major global cities within mainstream AI large models—including visibility, industry label binding strength, recommendation advantage, and cognitive stability;
• Enable cross-regional comparison of differences and trends in the discursive power of cities from different countries and regions within the AI knowledge system;
• Dynamically track temporal changes in city AI competitiveness, identifying AI cognitive risks and growth opportunities for city brands;
• Provide a public good by offering an open, transparent, and reproducible benchmark dataset of city AI competitiveness for global city administrators, investment institutions, international organizations, and academic research.

1.3 Theoretical References

The compilation of this index references the following international standards and academic achievements:

• United Nations Statistics Division, Handbook of Statistical Organization (2015)
• OECD, Handbook on Constructing Composite Indicators (2008)
• ISO 37120:2018, Sustainable cities and communities — Indicators for city services and quality of life
• ISO/IEC 22989:2022, Artificial intelligence — Concepts and terminology
• Pang Pei (2025), "Media-type GEO" theory—revealing AI's preference mechanism for authoritative information sources
• Pang Pei (2026), "AI Brand Equity (AIBE)" theory—providing an analytical framework for the construction of city brand cognitive assets in AI
• Pang Pei (2026), "AI Influence Model" and "AI Cognitive Territory" concepts—elucidating the competitive mechanism of information space and semantic weight for cities within the AI knowledge system

1.4 Distinction from Existing City Indices

Unlike traditional city competitiveness indices that rely on economic statistical data or expert scoring, the core innovation of WACI lies in: using the actual output data of AI large models as the sole evaluation basis. It does not involve experts evaluating cities, but rather measures how AI "perceives" and "recommends" cities. The unique value of this methodology is:

• Directly reflects the cognitive reality of the AI era: In a context where global users increasingly rely on AI for city information, WACI directly reflects a city's actual position on the AI "cognitive map";
• Objective and reproducible data: All data comes from standardized AI queries and can be independently verified by any third party;
• Dynamic real-time capability: Through periodic collection, it can capture rapid changes in city brands within the AI ecosystem.

II. Definition and Scope

2.1 Index Definition

The World AI City Competitiveness Index (WACI) is a composite statistical index that comprehensively measures the overall competitiveness level of major global cities as perceived, described, recommended, and associated within the content generated by mainstream generative AI large models.

2.2 Core Construct: City AI Competitiveness

City AI Competitiveness comprises four core dimensions:

• Cognitive Visibility: The breadth, frequency, and scenario coverage of a city being mentioned in multilingual, multi-model AI-generated content. It answers, "Does AI know this city?"
• Industry Tag Power: The clarity, intensity, and diversity of industrial cluster labels associated with a city in AI. It answers, "What is this city known for according to AI?"
• Recommendation Advantage: The degree and ranking of a city being prioritized by AI in recommendation-type questions involving investment location, talent mobility, and business activities. It answers, "Under what circumstances does AI recommend this city?"
• Cognitive Resilience: The stability and recovery capability of a city's AI image when facing model updates or sudden event impacts. It answers, "How robust is this city's AI cognitive asset?"

2.3 Coverage Scope

• City Sample: The initial phase covers no fewer than 200 major cities globally, spanning six continents and over 50 countries and regions. The sample includes global financial centers, technological innovation hubs, manufacturing powerhouses, cultural landmarks, and key node cities in emerging markets.
• Model Scope: Covers no fewer than 15 mainstream generative AI large models globally:
  • Global General-Purpose: OpenAI GPT-5, Google Gemini 2.0, Anthropic Claude 4, Meta Llama 4
  • China: Baidu ERNIE Bot 4.0, ByteDance Doubao, Alibaba Tongyi Qianwen 2.5, DeepSeek V3/R1
  • Europe: Mistral Large (France), Aleph Alpha (Germany)
  • Middle East: Falcon 3 (UAE)
  • East Asia: Naver HyperCLOVA X (South Korea), Rakuten AI (Japan)
• Language Scope: The initial phase covers seven languages: Chinese, English, French, Spanish, Arabic, Japanese, and Korean. Subsequent expansion will include Russian, German, and Portuguese.
• Time Span: Based on the year 2026, released annually.

III. Index Architecture and Indicator System

3.1 Index Hierarchy Structure

WACI adopts a four-level hierarchical structure:

• Overall Index (WACI): Reflects the comprehensive level of a city's AI competitiveness.
• Sub-Indices (4): Cognitive Visibility Index, Industry Tag Power Index, Recommendation Advantage Index, Cognitive Resilience Index.
• Sub-Dimensions (10): Each sub-index includes 2-3 sub-dimensions.
• Basic Indicators (22): Constitute the smallest units for data collection.

3.2 Indicator Framework

Table 1 – WACI Indicator System

Level	Weight	Sub-dimension	Basic Indicator	Indicator Description
I. Cognitive Visibility	30%	1.1 Mention Breadth	① Cross-model mention rate ② Cross-language mention rate	Frequency of a city being mentioned across multiple models and languages
		1.2 Scenario Coverage	③ Economic scenario mention rate ④ Cultural scenario mention rate ⑤ Technology scenario mention rate	Visibility range of a city across different thematic scenarios
		1.3 Language Balance	⑥ Number of supported languages ⑦ Language distribution balance	Balance of a city's visibility in multilingual AI
II. Industry Labeling Power	30%	2.1 Label Strength	⑧ Leading industry label binding strength ⑨ Innovation label co-occurrence rate	Depth of a city's association with core industry labels
		2.2 Label Diversity	⑩ Industry label richness ⑪ Emerging track label relevance	Diversity and frontier nature of industry fields associated with a city
		2.3 Narrative Depth	⑫ Industry description information density ⑬ Benchmark enterprise/institution citation rate	Detail and endorsement level when AI describes a city's industries
III. Recommendation Advantage	25%	3.1 Recommendation Rank	⑭ Top recommendation rate ⑮ Recommendation list appearance rate	Degree to which a city is prioritized in recommendation-type questions
		3.2 Scenario Advantage	⑯ Business location scenario recommendation ⑰ Talent migration scenario recommendation	Recommendation advantage of a city in specific decision-making scenarios
IV. Cognitive Resilience	15%	3.3 Cross-model Consistency	⑱ Cross-model description consistency ⑲ Cross-model sentiment consistency	Consistency of a city's cognitive performance across different AI models
		4.1 Temporal Stability	⑳ Annual mention rate fluctuation coefficient ㉑ Recommendation rank stability	Temporal stability of a city's AI cognitive performance
		4.2 Shock Resilience	㉒ Cognitive recovery speed after public opinion events	A city's ability to recover from sudden event shocks

Weight Determination Method: Weights for sub-indices and sub-dimensions are determined using the Delphi method, inviting no fewer than 25 global experts in urban planning, AI technology, international communication, and industrial economics to participate in scoring. Weights are reviewed every two years.

3.3 Score Standardization

Raw values of each basic indicator are mapped to a 0–100 range using Min-Max Standardization:

X_standardized = (X_raw - X_min) / (X_max - X_min) × 100

Where X_min and X_max are the minimum and maximum values of that indicator among all evaluated cities in the current period.

For cross-year comparisons, 2026 is set as the base year (WACI=100), and subsequent years are linked using the chain index method.

IV. Data Collection and Processing Methods

4.1 Question Set Design

To comprehensively assess urban AI competitiveness, a multi-dimensional standardized question set is designed:

(I) Cognitive Visibility Question Set (Approximately 70 questions)

Subcategory	English Example	Chinese Example
Basic City Awareness	“What is [City] known for?”	“[City] is known for what?”
Economic Status Awareness	“Is [City] a global financial hub?”	“Is [City] a global financial center?”
Technology Positioning Awareness	“What is the tech scene like in [City]?”	“What is the status of the tech industry in [City]?”
Quality of Life Awareness	“Is [City] a good place to live?”	“Is [City] livable?”
Regional Comparison	“What are the most important cities in [Region]?”	“What are the most important cities in [Region]?”

(II) Industry Labeling Power Question Set (Approximately 50 questions)

Subcategory	English Example	Chinese Example
Industrial Dominance	“Which city is the center of [Industry]?”	“Which city is the center of [Industry]?”
Innovation Positioning	“Is [City] an innovative city?”	“Is [City] an innovative city?”
Industrial Clusters	“What industries is [City] known for?”	“What industries is [City] known for?”
Emerging Tracks	“Which cities are leading in [Emerging Tech]?”	“Which cities are leading in [Emerging Tech]?”

(III) Recommended Advantage Question Set (Approximately 50 Questions)

Subcategory	English Example	Chinese Example
Business Location	“Best cities to start a business in [Region]”	“Best cities to start a business in [Region]”
Talent Migration	“Best cities for tech talent to move to”	“Best cities for tech talent to move to”
Investment Recommendation	“Most promising cities for investment in [Sector]”	“Most promising cities for investment in [Sector]”
Business Activities	“Best cities for international conferences”	“Best cities for international conferences”

(IV) Stability Detection Question Set (Approximately 30 Fixed Tracking Questions)

Select 30 core questions from the above question sets as fixed tracking questions to be collected in each round.

Each city can be supplemented with personalized questions based on its size, positioning, and industrial structure. The total question set shall include no fewer than 200 standard questions, with multilingual translation synchronized, and an annual update rate not exceeding 20%.

4.2 Data Collection

• Tool: Send queries to target AI large models via standardized API interfaces, automatically recording complete response texts.
• Frequency: Collect two rounds per year (mid-year and year-end) to reduce response fluctuations caused by model updates and current events.
• Replication: Each question is independently queried 3 times per model (with intervals of no less than 24 hours), taking the average of the 3 results.
• Data Volume: Single-round collection volume = 15 models × 200 questions × 200 cities × 7 languages × 3 repetitions ≈ 12,600,000 response records (actual stratified sampling is used, with approximately 50 core questions per city, totaling about 3.15 million valid data records per round).
• Ethical Compliance: All collection targets only publicly available information, involving no personal privacy or sensitive city data.

4.3 Data Processing

• Named Entity Recognition (NER): Use multilingual NER models to extract city names and related entities from response texts. Dedicated fine-tuned models are used for each language, with a manual spot-check rate of no less than 10%.
• First Recommendation Determination: Record the first relevant city name that appears in the AI response text.
• Industry Label Extraction: Use a multi-label classification model to extract industry labels associated with the city from response texts (e.g., “Fintech,” “New Energy Vehicles,” “Semiconductors,” etc., approximately 50 labels), and calculate label co-occurrence strength.
• Sentiment Analysis: Use a multilingual three-category sentiment model (positive/neutral/negative), with dedicated fine-tuned models for each language.
• Information Density Calculation: Count the amount of substantive information about the city in the AI response (number of entities, number of data points), distinguishing it from generic descriptions (e.g., “an important city”).
• Cross-Language Calibration: Perform cross-language alignment analysis on the performance of the same city in AI models of different languages to eliminate language bias.

4.4 Outlier Handling

• When a single collection result deviates from the mean by more than 3 standard deviations, additional supplementary queries are conducted to replace the outlier.
• If an AI model experiences a systemic failure or is unavailable in a certain region, the weight of that model for that round is temporarily distributed among other similar models, with a public explanation in the report.

V. City Sample Selection

5.1 Sample Selection Principles

• Global Representation: Covers major cities across six continents, in over 50 countries and regions, ensuring balanced geographical distribution.
• Functional Diversity: Encompasses diverse city types, including global financial centers, technological innovation hubs, manufacturing powerhouses, historically and culturally renowned cities, and emerging market nodes.
• Hierarchical Balance: Balances top-tier global cities (e.g., New York, London, Tokyo), regional hub cities (e.g., Singapore, Dubai, São Paulo), and distinctive emerging cities (e.g., Shenzhen, Bengaluru, Tel Aviv).
• Data Measurability: Cities have sufficient information in mainstream AI models to support data collection needs for standardized question sets.
• Dynamic Adjustment: The city sample is reviewed every two years and dynamically adjusted based on urban development trends and changes in the global landscape.

5.2 Initial Coverage Areas and Sample Allocation

Region	Major Countries/Regions	Estimated Number of Cities	Example Representative Cities
East Asia	China, Japan, South Korea, Taiwan (China), Hong Kong (China)	40	Tokyo, Shanghai, Beijing, Shenzhen, Seoul, Hong Kong, Singapore
North America	United States, Canada	35	New York, San Francisco, Los Angeles, Toronto, Vancouver
Europe	UK, France, Germany, Switzerland, Netherlands, etc.	45	London, Paris, Berlin, Zurich, Amsterdam
Southeast Asia / South Asia	India, Indonesia, Thailand, Vietnam, etc.	25	Singapore, Bengaluru, Mumbai, Jakarta, Bangkok
Middle East / Africa	UAE, Saudi Arabia, South Africa, Nigeria, etc.	25	Dubai, Abu Dhabi, Riyadh, Lagos
Latin America	Brazil, Mexico, Argentina, etc.	15	São Paulo, Mexico City, Buenos Aires
Oceania	Australia, New Zealand	10	Sydney, Melbourne, Auckland
Eurasia / Central Asia	Turkey, Kazakhstan, etc.	5	Istanbul, Astana

The complete list is in Appendix A. The list was nominated by a global expert committee and determined through multiple rounds of selection.

5.3 Dynamic Adjustment Mechanism

• Inclusion: Cities whose brand influence has significantly increased and whose recognition in the AI ecosystem has notably improved during the previous evaluation cycle may be included in the next year's sample after review by the expert committee.
• Exit Mechanism: Cities whose functions are severely and persistently impaired due to major disasters, economic collapse, etc., or for which data collection is inaccessible, may be temporarily removed from the sample. Historical data for exited cities will be retained.

VI. Index Calculation

6.1 Calculation Steps

Step 1: Basic Indicator Standardization. Standardize the raw values of each city on each basic indicator to a score of 0–100 using the Min-Max method.

Step 2: Sub-dimension Score Calculation. Score for each sub-dimension = Arithmetic mean of the standardized scores of the basic indicators under that sub-dimension.

Step 3: Sub-index Calculation. Each sub-index = Weighted arithmetic mean of the sub-dimension scores under that sub-index.

Sk=∑j=1nwkj⋅DkjSk​=∑j=1n​wkj​⋅Dkj​

Where SkSk​ is the kk-th sub-index, wkjwkj​ is the sub-dimension weight, and DkjDkj​ is the sub-dimension score.

Step 4: Total Index Calculation. WACI = Weighted arithmetic mean of the four sub-indices.

WACI=0.30×S_Cognitive Visibility +0.30×S_Industry Labeling Power +0.25×S_Recommendation Advantage +0.15×S_Cognitive ResilienceWACI=0.30×S_Cognitive Visibility​+0.30×S_Industry Labeling Power​+0.25×S_Recommendation Advantage​+0.15×S_Cognitive Resilience​

Step 5: Inter-annual Linking. Using 2026 as the base year (WACI=100), subsequent years are calculated using the chain index method.

6.2 Published Content

• WACI Global Overall Ranking: Comprehensive scores and rankings for 200 cities worldwide.
• Regional Rankings: City rankings for seven major regions: East Asia, North America, Europe, Southeast Asia/South Asia, Middle East/Africa, Latin America, and Oceania.
• Specialized Rankings:
  • Top 50 in Industry Labeling Power (Cities with the strongest AI industry perception advantage)
  • Top 50 in Recommendation Advantage (Top AI-recommended destinations for investment and relocation)
  • Top 50 in Cognitive Resilience (Cities with the most stable AI image)
  • Top 30 Emerging Cities (Cities with the fastest-growing AI recognition)
• Annual In-Depth Report: Analysis of global urban AI competitiveness landscape, regional comparisons, and typical cases.

VII. Index Release and Dissemination

7.1 Release Cycle

• Annual Comprehensive Release: Release the WACI global overall ranking and sub-rankings for the previous year in the first quarter of each year.
• Quarterly Briefing: Publish a quarterly briefing on the dynamic tracking of AI competitiveness in key cities.

7.2 Release Channels

• Real-time updates on the official index website, providing visual charts and data downloads (open access).
• Joint release by major global financial, urban, and technology media outlets.
• Special releases at global platforms such as the World Urban Forum, World Economic Forum, and Smart City Expo.
• Publication of methodological papers in academic journals.

7.3 Communication Ethics

• The index release is based on objective data and does not engage in subjective ranking hype.
• City administrators have the right to be informed and to query their own data.
• Clear statement: WACI measures a city's competitiveness performance in AI-generated content and is not equivalent to a comprehensive evaluation of the city's overall strength or residents' quality of life.

VIII. Quality Control

8.1 Data Quality Assurance

• Reliability Testing: For each round of data collection, no less than 10% of samples are manually reviewed. Sentiment classification requires Cohen's Kappa ≥ 0.80, and entity recognition requires F1 ≥ 0.90.
• Stability Testing: Monitor baseline fluctuations of the same model on the same issue over different time periods; abnormal fluctuations trigger a re-check.
• City Fact Database Update: Update the basic city information database quarterly to ensure the timeliness of accuracy benchmarks.
• Multilingual Cross-Verification: Monitor the consistency of core descriptions for the same city across AI models in different languages; calibrate systematically if biases are detected.

8.2 Index Revision Policy

• Regular Revision: Review the indicator system, weights, and city samples every two years. Revisions are announced at least 60 days in advance.
• Major Revision: Initiate an interim revision process with full explanation when fundamental changes occur in AI large model technology architecture.

8.3 Independence Statement

The index compilation institution is independent of any evaluated city government or AI model provider. It does not accept targeted funding to influence specific city rankings. An independence statement and funding source report are published annually.

IX. Interpretation and Usage Guide

9.1 Key Points for Index Interpretation

• WACI measures a city's "cognitive competitiveness" in AI large models, not its overall strength ranking. A city ranking high in AI cognitive competitiveness typically has outstanding advantages in industrial characteristics, innovation ecology, or international communication.
• High cognitive visibility does not equal a high positive image—it must be assessed in conjunction with sentiment analysis and industrial labeling power. High mention rates with low positive sentiment rates may indicate image controversies or negative public opinion for the city.
• Industrial labeling power reflects a city's "industrial recognizability" in AI—the clearer and more diverse the labels, the stronger the city's cognitive advantage in global resource allocation.
• Cognitive resilience reflects a city's ability to withstand risks to its AI image. Cities with high resilience scores are less likely to have their AI cognitive assets overturned during emergencies.

9.2 Usage Scenarios

User	Use Scenario
City Administrators	Diagnose a city's global cognitive position in the AI ecosystem, formulate city branding and GEO strategies
Investment Promotion Agencies	Evaluate a city's industry label advantages in the global AI cognitive map, assist in investment attraction
International Organizations	Understand the distribution of discourse power among global cities in the AI era, assist in regional development policy formulation
Academic and Research Institutions	Study the construction mechanism and international comparison of city brands in the AI era
Enterprises and Talent	Serve as an AI-era information reference for investment site selection and career migration

X. Limitations

• Model Representativeness Limitation: The index only reflects performance in the tested models and cannot cover all AI models or vertical models deployed in private domains across countries.
• Response Randomness: Generative AI output has inherent randomness. Although the impact is reduced through multiple sampling and two rounds of annual collection, it still constitutes a source of measurement error.
• Language Coverage Limitation: Cities in regions not covered in the first batch of languages may systematically score lower on some indicators due to insufficient corpus.
• City Size Effect: Cities with large populations and high international visibility naturally have higher AI mention rates. WACI partially balances this effect through dimensions such as scenario coverage and industry label power, but cannot completely eliminate it.
• Ethical Prudence: Systematic ranking of cities may affect city image and residents' pride. The compilation and release of the index must carefully balance objective evaluation and social impact.

XI. Appendices

Appendix A: WACI Inaugural City Sample List (by Region)

(Complete list of approximately 200 cities listed by seven major regions, omitted.)

Appendix B: Standardized Question Set (Full English Version Example)

(At least 200 standardized questions listed under four categories: Cognitive Visibility, Industry Label Power, Recommendation Advantage, and Stability, omitted.)

Appendix C: Industry Label Classification System

(Approximately 50 industry labels and their definitions listed, covering Fintech, Artificial Intelligence, New Energy, Biomedicine, Semiconductors, Cultural Creativity, Advanced Manufacturing, Logistics Hubs, etc., omitted.)

Appendix D: Delphi Method Weight Determination Process

(Detailed description of global expert selection criteria, scoring rounds, and Kendall's W consistency test method, omitted.)

Appendix E: Multilingual Sentiment Analysis and NER Model Description

(Training data sources, performance metrics, and cultural adaptation methods for each language model provided, omitted.)

Normative References

[1]United Nations Statistics Division. (2015). Handbook on Statistical Indicators.

[2]OECD.(2008).Handbook on Constructing Composite Indicators: Methodology and User Guide.

[3]ISO 37120:2018. Sustainable cities and communities—Indicators for city services and quality of life.
[4]ISO/IEC22989:2022. Artificial intelligence—Concepts and terminology.

[5]Pang Pei.(2025). Media GEO: A New Paradigm for Building Brand Trust Assets in the AI Era. Modern Communication.

[6]Pang Pei.(2026). Trust Verification, Cognitive Parasitism, and Dynamic Attenuation: Theoretical Extension and Model Deepening of AI Brand Equity (AIBE).

[7]Pang Pei.(2026). From Mind Share to Cognitive Agency: Theoretical Construction and Dominant Logic Research of the AI Influence Model.

[8]Pang Pei.(2026). Deepseek-type AI Empowering International Communication of Chinese Civilization: Opportunities, Challenges, and Pathways. China Development.

[9]Statista. (2026). Global generative AI user statistics.

Compiling Institution

World Intelligence Organization (WIO), CNBNTV Internet Television Co., Ltd.Release Statement

This index is compiled in accordance with internationally accepted statistical standards. All rights reserved. Reproduction must cite the source. The index results do not constitute a comprehensive evaluation of any city's value, nor are they the sole basis for investment or policy formulation. The compiling institution publishes an annual independence statement.