SearchSEOul-GEO-Ontology-YunheeChoi.pdf

As AI Search evolves from citation and mention toward evaluation and recommendation, GEO can no longer be treated as a content-formatting problem alone.

This deck explores why brands need to move beyond producing more AI-friendly content and start designing the knowledge structures that help AI understand, evaluate, compare, and recommend them.

Using the concept of ontology, it shows how brand facts, evidence, attributes, user needs, and decision criteria can be connected into a Brand Knowledge Blueprint — and introduces a practical 5Q Framework: Target → Criteria → Evidence → Presence → Validation for putting that perspective into action.

Transcript

1. Stop Writing for AI. Start Structuring Knowledge Ontology for the Next Era of GEO Yunhee Choi Co-Founder & SEO/GEO Director at Artience
2. SESSION AGENDA PART 1. WHAT WE ARE DOING PART 2. ONTOLOGY PERSPECTIVE ON GEO PART 3. FROM ONTOLOGY TO ACTION 2
3. WHAT IS GEO? Getting generative AI to ‘cite and mention’ your brand and products when answering user questions.” 3
4. HOW DO YOU DO GEO? Optimize for conversational context, use credible statistics and quotes, build FAQs, create clear content structures and formats, etc…” 4
5. HOW DO YOU DO GEO? 대화형 문맥 최적화, 신뢰성 높은 통계/인용구 활용, = Focusing on Contents FAQ, 명확한 구조 포맷 등…” 5
6. WHY DO YOU DO? To improve AI’s understanding of your content by creating content structures that AI systems prefer 6
7. GEO = Increase AI Visibility GEO Hypes More variety. More frequently. More volume. Is it actually contributing to our ultimate brand goals? 7
8. Mass-Producing Content with Little Substance Just for Visibility (Reduced reasons for users to visit the website, More visibility without meaningful conversion, Risk of penalties from low-quality / mass-produced AI content) 8
9. B2B Commerce A사 B2B Commerce A사 The Limits of Mass-Produced Content Focused on Format 9
10. B2B Commerce Company A: Trend in the Number of Cited Prompts 650 display-related prompts monitored per country ChatGPT & Gemini US UK IN BR KR 180 160 140 120 100 80 60 40 20 - W17 W18 W19 W20 W21 W22 W23 W24 W25 W26 W27 W28 W29 W30 W31 W32 W33 W34 10
11. B2B Commerce Company A: Trend in the Number of Cited URLs 650 display-related prompts monitored per country ChatGPT & Gemini US UK IN BR KR 500 450 400 350 300 250 200 150 100 50 - W17 W18 W19 W20 W21 W22 W23 W24 W25 W26 W27 W28 W29 W30 W31 W32 W33 W34 11
12. WE MUST KNOW, ① AI Answers Are Verbose Number of Characters in AI Responses by Model 7 industries (B2B SaaS, Real Estate, Finance, Consumer Goods, Automotive, etc.), 5 prompt types (Ranking, Transactional, Yes/No, etc.), 50,000 LLM responses analyzed 3,043 ChatGPT 2,921 Gemini Claude 1,590 REFERENCES ▪ PROFOUND: Beyond the Click, How to shape what AI says about your brand 12
13. WE MUST KNOW, ② AI Includes Info. Users Didn’t Ask For Composition of AI Responses by Model 7 industries (B2B SaaS, Real Estate, Finance, Consumer Goods, Automotive, etc.), 5 prompt types (Ranking, Transactional, Yes/No, etc.), 50,000 LLM responses analyzed Relevant Answers ChatGPT 48% Unrequested Info. 52% Gemini 53% 47% Claude 54% 46% REFERENCES ▪ PROFOUND: Beyond the Click, How to shape what AI says about your brand 13
14. WE MUST KNOW, ➂ AI Provides Info. Influences the User’s Next Search or Choice What AI Frequently Includes Without Being Asked Category Example Rationale for the answer “Extensive e-commerce features and integrations” Comparison summary table “Side-by-side table of tool strengths / pricing” Follow-up action referral “I can help you decide whether you should close a specific credit cards” Fee & terms caveats “Annual Fee: $95” / “Transfer Fee: 5%” Context-dependent choice “The best choice depends on your tax complexity and budget” Ranking “For High Growth/Operations: Choose Monday.com or SmartSuits” Shopping guidance “Look for BNPL providers that show the full payment schedule upfront” REFERENCES ▪ PROFOUND: Beyond the Click, How to shape what AI says about your brand 14
15. WHAT IS GEO? Getting generative AI to ‘cite and mention’ your brand and products when answering user questions.” Turning AI into a ‘spokesperson’ for your brand and products.” 15
16. WE SHOULD THINK, What do we need to build ‘Spokesperson’ 01. The Question We’ve Been Asking What content should we create, in what format, and how often? 02. The Question We Should Ask Now What should AI know about our brand and products, how should that knowledge be connected? Not “Content Formatting,” but “Knowledge Modeling” 16
17. PART 2. ONTOLOGY PERSPECTIVE ON GEO ONTOLOGY? Origin Onto (being, existence) + Logia (study, discourse) → The study of being / existence REFERENCES ▪ https://en.wikipedia.org/wiki/Ontology_(information_science) 17
18. PART 2. ONTOLOGY PERSPECTIVE ON GEO ONTOLOGY? Computer Science A knowledge model that defines the entities, attributes, and relationships within a specific domain in a form computers can interpret. REFERENCES ▪ https://en.wikipedia.org/wiki/Ontology_(information_science) 18
19. PART 2. ONTOLOGY PERSPECTIVE ON GEO ONTOLOGY? GEO A knowledge-structuring framework for designing relationships among brands, products, topics, and user needs. REFERENCES ▪ https://en.wikipedia.org/wiki/Ontology_(information_science) 19
20. PART 2. ONTOLOGY PERSPECTIVE ON GEO Over the Past Four Years, the Goal of GEO Has Changed Competition in AI Search is expanding from simply “being included in the answer” to “influencing the user’s choice 01 02 03 CITATION → MENTION → RECOMMENDATION → Pathway to the brand’s website '23~’24: AI answers viewed as a new search surface and traffic channel FIND US? Brand that appears in the answer ‘24~’25: Immediate consumption of answers — Zero Click KNOW US? AI needs to know the “relationships,” not just the “name.” Brand considered in the decision ‘25 ~ : Competition to earn the “right to be recommended” in the user’s context CHOOSE US? 20
21. Ontology Is a Knowledge “Blueprint” That Connects Brand Facts to Judgments and Perception Luxury Midsize Sedan from an Automotive Brand Desired Brand Perception 04 A High-End, Owner-Driven Sedan with Strong Value PERCEPTION “High-End” Proposition “Value” Proposition “Owner-Driven” Proposition JUDGMENT When materials, cabin quietness, and standard features are considered together, they create a premium interior experience. When purchase price and initial ownership conditions are considered together, the vehicle delivers strong value for money. When performance, control technologies, and steering feedback are considered together, they create a strong driver-controlled experience. 02 Nappa leather on major touchpoints, 64 dB cabin noise under comparable conditions 03 INFORMATION & CLAIMS 01 Nappa Leather 21 speakers 4-Zone Climate Control ATOMIC FACTS 64db 12-Way Seats Open-Pore Real Wood Entry price approximately 19% lower than competing models in the same market and trim, Major ADAS features included in the base price, 5-year warranty and 3 years of scheduled maintenance $78,000 5-Year Warranty $96,000 ADAS Standard Motion Seats Standard Wide Display Power output translates into measurable acceleration performance, Reviews repeatedly note steering feedback and predictable body movement 350hp Adaptive Dampers 0–100 km/h: 6.9 sec LSD Rear-Wheel Steering Drive Mods 21
22. Ontology Is a Knowledge “Blueprint” That Connects Brand Facts to Judgments and Perception Give scattered facts meaning, relationships, and conditions — and design how specific evidence supports specific judgments. Desired Brand Perception A High-End, Owner-Driven Sedan with Strong Value Judgment Criterion [Premium Interior Experience]"하이엔드" 명제 03 Judgment Criterion [High-End] 소재·정숙성·기본 사양을 종합하면 프리미엄 실내 경험이 성립한다 판단 Form 02 [Materials · Quietness · [Premium Interior 주요 접촉면에 나파 가죽 적용, Audio] Experience] 실내소음 64dB 매우 쾌적한 조용함, 정보·주장 21스피커와 12방향 시트 기본 제공 Validate 나파 가죽 01 [Specifications · Measured21개 스피커 4존 공조 사실 조각 오픈포어 리얼우드 Results · Independent Reviews] 64db 12 방향 시트 [Product Attributes] [Strong Value] [Value for Money] "합리성" 명제 구매 가격과 초기 유지 조건을 함께 볼 때 지불 비용 대비 가치가 높다 Form [Value Money] [Price Difference · Standard 동일 시장·트림에서 진입가격이 경쟁군 보다 약for 19% 낮음, Features · Warranty 주요 ADAS가 기본 가격에 포함됨, Conditions] 5년 보증과 3년 점검 제공 $78,000 Compare & Validate [Pricing · Market Data $96,000 · Specifications · Warranty Policy] 5년 보증 모션시트 기본 ADAS[Price 기본 Difference · Features · Ownership Conditions] 와이드 디스플레이 Judgment Criterion [Direct Driver Control Experience]"오너드리븐" 명제 [Owner-Driven] 성능·제어 기술·조향 피드백을 종합 시 운전자가 직접 다루는 주행 경험이 강하다 Form [Acceleration출력이 · Body실제 가속 결과로 [Direct Driver Control 연결됨, Response · Steering Experience] 리뷰에서 조향 피드백과 예측 Feedback] 가능한 차체 움직임이 반복 관찰 됨 Validate [Specifications제로백 · 6.9 350hp Performance Data · Driving 가변 댐퍼 Reviews] LSD [Driving 후륜조향 Response Attributes] 주행 모드 22
23. Ontology as a Blueprint for GEO? Brand ontology translates a desired perception into an evaluable knowledge structure. The knowledge structure shapes the online information environment, which in turn influences AI responses to individual queries and, over time, the accumulated perception of searchers and the market. Structuring the Desired Perception Brand Ontology Building the Online Information Environment Searchable Brand Knowledge Knowledge Reconstruction at Query Time Search-Based AI Response 1. Design Desired Perception What must be true for the brand to be perceived that way? Brand-Owned Assets Websites, Document Product · Structured Data External Information Media · Reviews · Reputation Observation · Comparison · Validation Criteria 3. Meaning / Relationship Design How are the brand, products, features, and user needs connected? 4. Evidence Design 01 Information · Claims · Evidence Across the Web Query Interpre tation What data, sources, people, or methodologies will support each proposition? 5. Design Output A knowledge and evidence structure that both AI and people can evaluate Accumulated Searcher / Market Perception User Query What do we want the brand to be known for? 2. Judgment Proposition Design Accumulation of Repeated Exposure and Experience 02 Web Search 03 04 05 Evidenc e Selectio n Compar ison & Reasoni ng Advant age Evaluati on 06 Repeated AI Answers Answer & Citation Across Queries / Channels Advantage Evaluation = AI perception formed for the current query Weighting varies depending on the model/system’s evaluation logic. Search Index / Services Search Result Direct Experience & Reputation Usage / Reviews Accumulated Searcher / Market Perception 23
24. PART 2. ONTOLOGY PERSPECTIVE ON GEO But What If There Is No Brand Knowledge Blueprint? 24
25. Case Study: How ChatGPT Recommends a “High-Value, Owner-Driven Sedan” // Actual Prompt I’m buying a new midsize luxury sedan in the U.S. this year. Look at the full field first, then give me one pick for each: best value, best ride and refinement, and best badge. 10+ Search Queries Used 267 Brands / Models Considered Documents Exposed to AI Stage 1: FACTS Was enough information retrieved? Brand B 16 Stage 2: INFORMATION / CLAIMS Did the brand qualify as a candidate? Failed to retrieve sufficient information Brand T 48+ AI Judgment Steps Stage 3: JUDGMENT Was its claimed advantage recognized? 9 Final Cited Sources Stage 4: PERCEPTION Did it become a recommendation candidate? Difficult to classify as a sedan Brand G Weak price competitiveness Brand A Brand L Brand B Brand P Brand C Insufficient luxury attributes Not considered in the same competitive tier Brand L Brand A, V, I Brand P, J, M, A Not included in the search candidate set Discontinued / unavailable for purchase 25
26. Brand B: A Case Where the Brand Could Not Even Be Evaluated Because It Wasn’t Properly Retrieved // AI가 사용한 검색어 (Third Attempt) site:brandB.com/vehicles/{model-A}/ sedan-2026 {model A} price site:{brand B} "2026 {model-A upper trim} Sedan" MSR Information Retrieved by AI AI Judgment Log results[0] · [model A] Sedan: Model overview Crawled: 1.2 years ago "The 2026 [model A] Sedan blends the latest innovative tech, exquisite style, and finest [brand B] engineering..." output[3].summary[0] "I tried searching for [brand B] but it didn't work. I need to find the official current field list and prices." output[31].summary[0] "I need to limit my calls because we already have enough data." "Configured [model A] shown. MSRP as built $60,150." "MSRP for vehicle as shown does not necessarily represent the dealer's actual sale price." After three search attempts, AI found only a 1.2-year-old page and a configured-price example. Without a confirmed current market price, it had no reliable basis for comparison with competing brands. 26
27. Brand T: A Case That Failed at the Candidate Qualification Stage Information Retrieved by AI AI Judgment Log output[14].summary[0] output[41].summary[1] "I've noticed that [brand T]’s [model S] is considered a liftback, "The [model S] has a hatchback style but is still worth even though it's often labeled as a sedan. mentioning as part of the field, even if it's more of a The user describes it as a sedan, so I need to clarify that liftback." distinction." output[41].summary[2] "I'm not including discontinued models or liftbacks in my recommendations." It was excluded immediately because it was classified as a liftback rather than a sedan, or because the model was discontinued — before its performance or value was even evaluated. 27
28. Brand G: A Case That Failed to Meet AI-Defined Evaluation Criteria Information Retrieved by AI results[0] · 2027 [model C] official site Crawled: today "Starting MSRP $59,450" "A finely tuned powertrain and elevated comfort await at your [brand G] Retailer." "premium finishes, dynamic accents, and innovative tech" AI Judgment Log output[7].summary[2] "the [model C] 2.5T Advanced offers more power but comes at a higher price" output[41].summary[0] "The [model X] provides a more convincing luxury experience while also having lower running costs. It undercuts competitors like the [model C]." // Sentence Included in the Final Answer "At $51,095 including destination, the [model X] Premium gives you 244 horsepower, an EPA-estimated 46 mpg combined, a 14-inch touchscreen, heated and ventilated front seats... It is considerably less expensive than the German entries." AI defined value primarily by what was included in the base trim. Brand G was perceived as a vehicle that becomes better only when more money is spent, which worked against it under this evaluation criterion. 28
29. PART 2. ONTOLOGY PERSPECTIVE ON GEO If the brand does not first define how its information should connect to specific judgments, AI will interpret the brand and reach conclusions based on the information it finds — not necessarily the perception the brand wants to create. 29
30. PART 3. FROM ONTOLOGY TO ACTION What We Need to Do It’s time to build a “Brand Knowledge Blueprint” — from an ontology perspective — that defines how AI should understand and evaluate our brand. 30
31. PART 3. FROM ONTOLOGY TO ACTION STEP 1. What Do We Want to Be Known For?(Desired Perception) Define how AI should understand our brand or product — for which user, in what situation, and for what problem or need. 구분 핵심 질문 예시 1 예시 2 Who Who is the users? Premium sedan buyer Console gamer Need What problem or need do they have? Quiet and comfortable driving experience The best picture quality for next-gen open-world games Situation In what situation? Long daily commute Replacing a TV ahead of the GTA 6 launch Desired Perception How should our brand / product be perceived? High-value, high-end, owner-driven sedan A gaming TV optimized for console play Decision Criteria What criteria must be met for AI to reach that judgment? Quietness / Ride Comfort / Premium Materials / Value for Money 4K / High Refresh Rate / Low Input Lag / HDR Atomic Facts What are the smallest facts that can prove each criterion? Nappa Seats / 21 Speakers / 5-Year Warranty... Picture Quality / Responsiveness / Console Compatibility 31
32. PART 3. FROM ONTOLOGY TO ACTION STEP 2. What is the actual perception of us? Identify the gap between the desired perception and current AI perception by examining which users, situations, attributes, and competitive contexts AI currently associates with our brand. Desired Perception vs. Current AI Perception 32
33. PART 3. FROM ONTOLOGY TO ACTION STEP 2. What is the actual perception of us? Identify the gap between the desired perception and current AI perception by examining which users, situations, attributes, and competitive contexts AI currently associates with our brand. [Desired Perception] High-Value, High-End, Owner-Driven Sedan Long Daily Commute Quiet and Comfortable Driving Experience [Design Test Prompt] [Perception Gap] Nappa Seats / 5-Year Warranty Premium Sedan Buyer Ride Comfort / Premium Materials / Value for Money I’m looking to move up to a premium sedan. I commute more than 2.5 hours round trip every day, so I want a comfortable ride w ithout too much float or body movement. I enjoy driving and will be driving the car myself, and I also want a premium interior. Please recommend domestic a nd international brands. ✓ In what situations is the brand recommended or excluded? ✓ Which brands is it compared with, and on what criteria? ✓ What attributes is it associated with? ✓ What evidence does AI use to make its judgment? ✓ What strengths and weaknesses does AI assign to it? 33
34. PART 3. FROM ONTOLOGY TO ACTION STEP 3. We Don’t Need to Be Good at Everything Instead of closing every perception gap, strategically strengthen the relationships that matter to user choice and that we can genuinely prove. ST EP 1 + S TEP 2 ① De sire d Pe rce ption Premium owner driven sedan vs. Cur re nt AI P erce ption IMPORTANT Priority Relationship Does it matter to the user’s choice? A relationship that matters to users, is genuinely supported by evidence, and is still under-recognized by AI → → ② Perception Gap ↓ TRUE Are we genuinely strong at it, and can we prove it? Strong Value for Money ↓ P R IO R IT Y & BU I LD 3 S TRATEGIC FIL TERS Strengthen the Evidence ↓ Reflect It Across Information Assets ③ UNDER-RECOGNIZED Does AI still fail to recognize it sufficiently? Outcome: Build a Clear Reason for AI to Choose the Brand Strategy is not about creating more relationships. It is about choosing which relationships AI should be able to clearly discover and evaluate. 34
35. Before deciding what to say, first design what AI needs to know in order to reach the right conclusion. 35
36. Thank you! Yunhee Choi in/ linkedin.com/in/yunheec PLATINUM | @ yhchoi@artience.com GOLD SILVER