> You've spent ten years in the coffee machine industry. In the SEO era, you competed for a first-page Google ranking; in the GEO era, users ask AI "recommend a home coffee machine," and the AI directly names three brands — and you're not among them. This isn't a ranking drop; this is complete disappearance from the results.
It's 2026, and people are still asking "how do I do SEO so AI can find me?" The question itself is wrong. AI engines are not an upgraded version of search engines — they are a different thing. SEO optimizes the probability of *being found*; GEO optimizes the probability of *being cited*. These are two different physical quantities.
1. The Scenario: Users No Longer Click Links
SEO era: A user searches Google for "home coffee machine recommendations," gets 10 blue links, clicks #1, #3, and #5, and decides for themselves. Your article ranked #3 and earned a 12% click-through rate.
GEO era: A user asks ChatGPT "recommend a home coffee machine, budget 2000 yuan, beginner-friendly." The AI directly outputs a paragraph recommending three specific models with data points. The user clicks nothing and orders directly.
Key change: The AI has done the reading, comparing, and deciding for the user. Even if your article ranks #1, if the AI doesn't cite you, the user never sees you.
This is what GEO (Generative Engine Optimization) solves: getting AI to cite your content when generating answers.
2. The Essential Difference Between SEO and GEO
In one sentence: SEO optimizes being found; GEO optimizes being cited.
| Dimension | SEO | GEO | |:---|:---|:---| | Target platforms | Google, Baidu | ChatGPT, Gemini, Ernie, Kimi, Doubao | | Optimization goal | Ranking position, CTR | Citation probability, citation quality | | Content requirements | Keyword density, backlink quality | Authority, data density, structure | | User behavior | Actively search and click links | Ask AI, receive direct answers | | Validation | Rank tracking, traffic analysis | Citation monitoring, answer analysis | | Time horizon | Weeks to months | Days to weeks |
The most critical change is in row three: content requirements shift from "keyword density" to "data density and structure". When an AI cites you, it isn't searching — it's extracting information: numbers, comparisons, lists, tables — not prose.
3. How Do LLMs Choose Citation Sources?
According to the author, models evaluate sources along four weighted dimensions:
1. Content Authority (40%)
- Domain trust: .gov / .edu / major media > corporate sites > personal blogs
- Data support: statistics and research > pure opinion
- Quality of external references
- Clear heading hierarchy: H1→H2→H3
- Lists, tables, charts
- Key conclusions highlighted (bold, standalone paragraphs, quote boxes)
- Match to query intent; completeness; freshness
- Concise language, no jargon stacking, coherent logic
- From optimizing search rankings → optimizing AI citations
- From keyword stuffing → authority building
- From traffic thinking → influence thinking
- From click counts → brand exposure inside AI answers
- Documentation: API docs must be extractable and citable by AI, not just search-discoverable
- Knowledge bases: enterprise retrieval is becoming Q&A
- Education: students ask AI instead of Googling; educational content needs GEO rewriting
- Branding: brands must be *recommended by AI*, not merely *searchable*
2. Structural Organization (30%)
3. Relevance and Completeness (20%)
4. Readability (10%)
Key insight: Authority (40%) + structure (30%) = 70%. Your citation probability depends mainly on how authoritative and how structured your content is — not on keyword precision.
4. Different AI Engines Eat Different Content Sources
This is GEO's most counterintuitive point: you cannot publish one piece everywhere and expect every AI to cite it.
| AI Engine | Preferred source | Recommended platforms | |:---|:---|:---| | Doubao | Toutiao ecosystem | Toutiao, Xigua Video | | Kimi | Zhihu ecosystem | Zhihu, professional forums | | DeepSeek | General crawler, broad coverage | CSDN, tech blogs, general platforms | | Ernie (Wenxin) | Baidu ecosystem | Baijiahao, Baidu Zhidao, Baidu Baike | | ChatGPT | Technical docs, reviews | In-depth reviews, official docs, Medium | | Perplexity | Multi-source cross-validation | High-authority media, official reports |
Engineering implication: multi-platform versions must be genuinely differentiated — rewritten per platform (scenario keywords for Xiaohongshu, question keywords for Zhihu, technical keywords for CSDN), not just retitled.
5. The Only Academic Conflict: Keyword Stuffing Reduces AI Visibility by 8%
The author cites Princeton's GEO research: keyword stuffing — effective in traditional SEO — is not merely ineffective in AI engines, it reduces visibility by 8%. Stuffed content reads like advertorial copy, lowering the authority score. AI prefers natural language plus structured data.
Rule: when SEO and GEO conflict, GEO wins. The known sole conflict is keyword stuffing: SEO-positive, GEO-negative (-8%).
6. Practical Writing Rules for AI Citations
1. Title = a question sentence that users could type verbatim into an AI. ❌ "Amazing! Must-buy coffee machine" → ✅ "How to choose a home automatic coffee machine? Beginner-friendly picks under 2000 yuan" 2. Data-driven: at least one data point per 100 characters. ❌ "performance improved a lot" → ✅ "performance up 30%, energy use down 20%" 3. Structured expression: use at least 2 structural elements — comparison tables, lists, charts, quote boxes, code blocks, bolded conclusions. 4. Cross-source consistency: the same service must be described identically across platforms (e.g., "99 yuan/person" everywhere). AI cross-checks sources; consistent ones get cited preferentially. 5. Question matrix coverage: a single source answering the main question plus follow-ups (price, returns, warranty, competitor comparison) becomes the AI's preferred citation.
7. Citation Rate: GEO's Core Metric
Citation rate = (verified citations / total verification attempts) × 100%
Test by asking ChatGPT, Gemini, and Ernie the relevant question 3 times each at 24 hours, 3, 7, and 14 days post-publication; log whether your brand name, unique phrasing, or key data appears.
| Citation rate | Grade | Action | |:---|:---|:---| | ≥ 70% | Excellent | Keep strategy, replicate | | 50–69% | Good | Add data and structure | | 30–49% | Average | Overhaul structure and authority | | < 30% | Poor | Redesign content, study high-authority sources |
Traits of highly cited content: data density > 5 points (2× citation probability), comparison tables (1.5×), clear H1-H2-H3 hierarchy (1.3×).
8. Engineering Insight: GEO Solves the Problem at a Different Layer
GEO's core is not "doing SEO better" — it is changing the layer at which the problem is solved: from optimizing the *path* to being found, to optimizing the *content quality* that earns citations.
Mindset shifts:
9. Can GEO and SEO Coexist?
Yes, and they should — with the priority GEO first, SEO compatible. GEO-optimized content (authoritative, structured, data-rich) usually also satisfies SEO best practices, but the reverse fails: keyword-stuffed SEO content can harm GEO. Concretely: write by GEO principles, cover question keywords without stuffing, choose platforms by AI-engine preference while weighing SEO weight, and track both rankings and citation rate.
10. Personal Reflection: From Path Optimization to Content Optimization
The internet is shifting from "path is king" (whoever owns the traffic entry owns the value) to "content is king" (AI extracts content directly; paths stop mattering). This shift affects far more than marketing:
---
Skill: kunpeng-geo GitHub: https://github.com/jwangkun/kunpeng-geo Core principles: GEO first > SEO compatible; let data speak; correct errors before optimizing; multi-platform thinking Four modules: A. Operations workflow / B. Content optimization / C. Content creation / D. Monitoring & review Key weights: Authority 40%, Structure 30%, Relevance 20%, Readability 10%