Appendix 2: Sources cited in this book

This appendix lists the sources behind the data and the main arguments in each chapter, in chapter order. Every entry gives the source type, the chapter it supports and a URL, so readers can check and update them.

Chapter 1: The paradigm shift

Consumer trust in AI search

• Data cited: In 2025, 82% of consumers found AI search more helpful than traditional search; in 2026 the figure was 54%, a fall of 28 percentage points in one year.
• Source: Fractl × Search Engine Land, 2026 AI Search Consumer Trust Study.
• URL: https://searchengineland.com/new-ai-search-data-visibility-trust-480089

Gartner forecast on search engine traffic

• Data cited: By 2026, traditional search engine visits will fall 25%.
• Source: A Gartner forecast, cited in a research report from CITIC Construction Investment.
• URL: https://www.stcn.com/article/detail/3550064.html

Scale of generative AI users in China

• Data cited: As of December 2025, China had 602 million generative AI users, a penetration rate of 42.8%.
• Source: the 57th Statistical Report on China's Internet Development (CNNIC).
• URL: https://m.gmw.cn/2026-03/04/content_38627125.htm

Rising weight of authoritative sources in AI citations

• Data cited: In May 2026, YouTube was cited 188,863 times (up 56.4% month over month), Wikipedia 83,191 times (up 55.2%), NIH 91,227 times (up 41.3%).
• Source: Meltwater GenAI Lens, tracking more than 8 million AI citations.
• URL: https://www.meltwater.cn/blog/ai-search-visibility-april-may-2026

How consumers handle AI search

• Data cited: 86% of consumers check the primary sources themselves when the information matters; 72.4% of users switch engines after AI gets a citation wrong twice in a row.
• Source: Fractl × Search Engine Land, 2026 study.
• URL: https://searchengineland.com/new-ai-search-data-visibility-trust-480089

Chapter 2: The core principles of Impact GEO

The citation weight gap between authoritative media and independent accounts

• Data cited: Content from national-level media is 8 to 12 times more likely to be cited by AI than content from ordinary sources. One national media outlet can carry more citation weight than a hundred independent accounts combined.
• Source: Industry research data, cited in GEO industry analysis articles.
• URL: http://www.xtrb.cn/syunf/2026-08/25/content_1288492.htm

Share of AI citations going to independent accounts and brand websites

• Data cited: Earned media (third-party sources) account for 82% to 94% of all AI citations; brand-owned websites account for a single-digit share.
• Source: AuthorityTech, 2026 study.
• URL: https://authoritytech.io/

Keyword density and citation rate

• Data cited: Pages above 8% keyword density are 17% less likely to be cited by AI than pages at a natural 2%–3%.
• Source: Industry test data. Widely repeated across GEO industry analysis articles; the original publication is still to be confirmed. Readers are referred to the testing section of the GEO industry whitepaper.

Chapter 3: The white-hat GEO method

The DSS principles (white-hat GEO method)

• Data cited: The three core principles of Semantic Depth, Data Support and Authoritative Source.
• Source: iResearch × Yuanyi Information, 2026 Generative Engine Optimization (GEO) Whitepaper.
• URL: https://report.iresearch.cn/report/202602/4787.shtml

Structured content and citation probability

• Data cited: Content in structured formats is 63% more likely to be cited directly by AI than plain text.
• Source: Volcano Engine developer community analysis report (April 2026), cited in GEO industry articles.
• URL: https://www.hongshu18.com/article-detail/BPL38gEB

Cited sources, statistics and direct quotes

• Data cited: Content with cited sources is 34.4% more likely to be cited by AI, content with statistics 32.1%, content with direct quotes 29.7%.
• Source: Industry test data. Widely repeated across GEO industry analysis articles; the original publication is still to be confirmed. Readers are referred to the testing section of the GEO industry whitepaper.

Named authors in AI citations

• Data cited: In May 2026 Google changed how AI Overviews displays citations and made named authors a search ranking variable.
• Source: Google announcement and Search Engine Land reporting.
• URL: https://searchengineland.com/seo-pulse-preferred-sources-expand-gmail-brand-lift-pichai-on-ai-overviews/577173/

Platforms cutting back low-quality content

• Data cited: DeepSeek compressed the sources it reads closely from 10-15 down to 4-5, and citation rates for more than 70% of mass-distributed press-release GEO content fell sharply.
• Source: DeepSeek's May 2026 algorithm update, analyzed by several GEO industry outlets.
• URL: https://www.hongshu18.com/article-detail/BPL38gEB

Schema markup and citation rate

• Data cited: Content without structured markup loses about 47% of its weight in AI citation on average.
• Source: Industry test data. Widely repeated across GEO industry analysis articles; the original publication is still to be confirmed. Readers are referred to the testing section of the GEO industry whitepaper.

Chapter 4: Incentives in the content ecosystem

Weight allocation across the five scoring dimensions

• Data cited: Media tier and endorsements (25%), content expertise (22%), social and citation signals (20%), site health (18%), content freshness (15%).
• Source: Public information from the major AI platforms plus industry testing. The weights are an industry observation; actual platform weights may differ, and readers should check each platform's own documentation.

Freshness and AI citation

• Data cited: AI engines rarely cite content that has not been updated in about two years, and the correlation between metadata freshness and citation rate reaches 0.68.
• Source: Industry research. Widely repeated across GEO industry analysis articles; the original publication is still to be confirmed. Readers are referred to the testing section of the GEO industry whitepaper.

Google Preferred Sources data

• Data cited: Google Preferred Sources covers 345,000 sources, and users are about twice as likely to click a preferred source as an ordinary link.
• Source: Google announcement, reported by Search Engine Journal.
• URL: https://www.searchenginejournal.com/seo-pulse-preferred-sources-expand-gmail-brand-lift-pichai-on-ai-overviews/577173/

The C2PA provenance standard

• Data cited: C2PA released Content Credentials 2.3, which binds production information into the file cryptographically.
• Source: C2PA official website.
• URL: https://c2pa.org/the-c2pa-launches-content-credentials-2-3-and-celebrates-5-years-of-impact-across-the-digital-ecosystem/

The GEO service trust assessment

• Data cited: The China Academy of Information and Communications Technology (CAICT), together with the AIIA Security Governance Committee, launched China's first GEO service trust assessment; the first nine companies passed and were certified.
• Source: Xinhuanet report "AIIA trusted GEO symposium held in Beijing".
• URL: https://www.xinhuanet.com/digital/20260518/5b81c242155a44da9a69d69f90410c57/c.html

Chapter 5: Governance and standards

The T/CAPT 026—2026 association standard

• Data cited: China's first association standard for trusted GEO dissemination. It sets up the A/B/C/D four-tier credibility rating and specifies three-zone separation, end-to-end traceability, a circuit-breaker mechanism and the L1/L2/L3 service capability tiers.
• Source: Issued by the China Association of News Technology Professionals, standard number T/CAPT 026—2026.
• URL: https://www.ttbz.org.cn/standardDetail/8617629717344255b65b44426dc77ce8.html

Bodies that took part in drafting

• Data cited: Drafted with participation from Xinhuanet Convergence Media Future Research Institute, the Xinhua News Agency State Key Laboratory and the Guangxi Daily.
• Source: Xinhuanet and Guangxi News reporting.
• URL: https://v.gxnews.com.cn/ (Guangxi Daily reporting)

The CAC's "Qinglang" campaign against AI application chaos

• Data cited: The campaign lists AI data poisoning carried out by tampering with training corpora, fabricating authoritative data or using GEO technology for malicious marketing as a priority target.
• Source: The CAC's "Cyberspace China" account, reported by CCTV.com, Guangming Online and others.
• URL: https://m.gmw.cn/2026-05/01/content_38744439.htm

CAA GEO standardization work

• Data cited: GEO standardization work began in full in March 2026, focused on end-to-end compliance standards. President Zhang Guohua called for "a healthy pattern where good money drives out bad".
• Source: Xinhuanet client reporting.
• URL: https://app.xinhuanet.com/news/article.html?articleId=0e5ab2a0858ccd79b752198b7bc7d26f

The UK CMA ruling against Google

• Data cited: In June 2026 the CMA required Google to attribute content sources and links more clearly in AI-generated search results and to give publishers an opt-out option from AI search, without demoting publishers in ordinary search for opting out.
• Source: Competition and Markets Authority (CMA) announcement.
• URL: https://www.gov.uk/find-digital-markets-measures/google-search-publisher-conduct-requirement

DeepSeek source compression

• Data cited: DeepSeek cut the sources it reads closely at the final stage from 10-15 down to 4-5, and citation rates for more than 70% of mass-distributed press-release GEO content fell sharply.
• Source: DeepSeek's May 2026 algorithm update, analyzed by GEO industry outlets.
• URL: https://www.hongshu18.com/article-detail/BPL38gEB

Cross-verification on AI platforms

• Data cited: The odds of content being blocked as a single point of failure reach 95%, and content that contradicts itself across platforms sees its odds of being featured fall 82%.
• Source: Industry test data, cited in GEO industry analysis articles.
• URL: https://cloud.tencent.com.cn/developer/article/ (Tencent Cloud developer community analysis)

Research report on a trusted GEO ecosystem

• Data cited: CAICT led and the AIIA published China's first Research Report on Building a Trusted Generative Engine Optimization (GEO) Ecosystem.
• Source: Xinhuanet reporting.
• URL: https://www.xinhuanet.com/digital/20260518/5b81c242155a44da9a69d69f90410c57/c.html

Preface and chapter 1 case

The Apollo-9 smart band poisoning case

• Data cited: Exposed at CCTV's 2026 3·15 gala. Industry insiders used the "Liqing GEO optimization system" to invent the Apollo-9 smart band; two hours after a dozen advertorials went out, two major AI models were recommending it as the standard answer.
• Source: CCTV's 3·15 gala, reported by CNR Online, the China Internet Joint Rumor Refutation Platform and others.
• URL: https://news.cnr.cn/dj/20260316/t20260316_527553512.shtml
• URL: https://www.piyao.org.cn/20260403/ad5ea72461a34cd3bf9a27d7f252fa3e/c.html

Notes on sources

1. Every source listed here is public material actually consulted while writing the book. Some industry test data (keyword density, Schema markup, scoring weights and similar figures) is repeated across many GEO analysis articles without a clear original publication; those entries say so.
2. The full text of the association standard is available on the national association standards information platform (www.ttbz.org.cn). Policy documents are available from the CAC's "Cyberspace China" account and from Xinhuanet, Guangming Online and other official outlets.
3. For academic research data (AI citation source statistics, trust surveys and the like), readers should obtain the original report for the full methodology and sample information before judging how far the figures apply.
4. This appendix is updated with each revision of the book. Corrections and additions are welcome.