The white paper uses the 'three laws of sources' to answer the core GEO question - what evidence AI accepts. This article breaks down entity anchor, cross-verification and real experience, and the three-pool source model behind 'about 85% of brand mentions come from third parties'.
The core of GEO is 'evidence'. So the next question is: what evidence does AI actually accept? The 2027 edition answers this with the 'three laws of sources' and upgrades it from opinion to a verifiable model.
I. GEO essence is evidence engineering for AI
The white paper summarises GEO essence as three verbs: crawl (technical) - cite (content) - verify (source). The first two decide whether you are 'present'; the third whether you are 'trusted'. Many get stuck at the third step because they only did content and official website, not source verification.
II. The standoff and reconciliation of two top-tier studies
In 2026 two studies gave opposite conclusions:
- Muck Rack (25 million cited links): third-party media 84%, brand-owned sites only 10% - conclusion 'AI trusts third parties, not self-narration';
- Profound (11.84 billion citations): brand-owned sites 57% - conclusion 'owned sites are still the largest single source pool'.
The difference is in statistical caliber: the former by domain type of cited links, the latter by page ownership of citations. Placed side by side, the real insight is: not 'official website vs media' as an either-or, but a 'twin engine' - the official website lets AI know you (entity anchor), media sources let AI believe you (cross-verification layer), community lets AI dare recommend you (real-experience layer).
III. The three laws of sources
- Law 1 (entity anchor): the official website is the brand's only 100%-controllable first-hand fact source. It decides whether AI can accurately answer 'who is this company, what does it do, what qualifications'. Without it, in AI eyes you are just a blur of citation fragments.
- Law 2 (cross-verification): AI does not rely on a single source. Research shows about 85% of brand mentions in AI answers come from third-party pages. Official website alone = talking to yourself; media alone = no name to point to.
- Law 3 (real experience): community and UGC decide whether AI 'dares recommend' you. They span about 50% across industries and are the fastest-rising layer, and also the main source of negative sentiment.
IV. The three-pool source model: missing one pool, no you
The white paper summarises the sources AI assembles an answer from into three pools:
- Owned pool: official website / product pages / documents / brand self-media. Entity anchor, 100% controllable, structurable; citation share 2%-74% by industry.
- Independent pool: media / industry sites / reviews / encyclopedias / directories. Cross-verification, the highest-weight trust signal for AI; press releases are nearly ineffective here (0.04%).
- Community pool: Zhihu / RED / forums / video / Q&A. Real experience, decides whether AI 'dares recommend', about 50%, weight still rising fast.
Typical failure symptoms of each pool alone: owned-only = talking to yourself, AI treats it as marketing; independent-only = no name to point to, AI mentions you but cannot say who you are; community-only = word-of-mouth without a source, AI dare not cite you on decision-class questions.
V. The counter-intuitive corollary: the official website is a 'cited object', not just a 'publishing channel'
If you only have official-website budget, do not rush to publish 100 articles - the more cost-effective move is to make the official website 'verifiable' first: qualifications, certifications, cases, parameters, sources, dates, all checkable. Because when third-party media cite you, they also verify facts against your official website. The official website is the 'cited object' of source building, not just a 'publishing channel'.
VI. Priority advice for enterprises
- Layer 1 (owned position) self-build: it is an asset that must stay in your hands;
- Layer 2 (industry / authoritative platforms) purchase: media relations and resources - self-building's time cost is far higher than buying;
- Layer 3 (social / community platforms) nurture: accumulates over time, cannot be bought short-term - keep investing, slowly nurture;
- Assets stay in your hands; resources can be bought with money.
Closing
The three laws of sources land on one line: make correct information easier for AI to cite, and wrong information harder for AI to systematically amplify. With official website, media and community all three pools complete, you are whole in AI answer.