GEO

GEO: Generative Search and Brand Visibility Engineering

Advanced Studies · 2026

Research Team

Tongxin Li

Tongxin Li

Principal investigator

Assistant Professor and Presidential Young Fellow at the School of Data Science, CUHK–Shenzhen, and Visiting Assistant Professor at the National Hetao AI Academy. He received his PhD from Caltech and his bachelor’s and master’s degrees from CUHK.

His research spans generative engine optimization (GEO), artificial intelligence, online decision-making, reinforcement learning, and control and optimization, with a focus on reliable decisions under uncertainty, dynamic feedback, and constraints.

Fanzeng Xia

Fanzeng Xia

PhD Student

PhD student at the CUHK–Shenzhen AI Decision Lab, admitted in 2023. He holds a master’s degree in Computer Science from New York University and a bachelor’s degree in Computer Engineering through Jilin University and Queen’s University, Canada.

His experience includes research assistance at McGill University’s Intelligent Automation Lab, algorithm engineering at Tsinghua University’s NLP Lab, and full-stack software engineering at Amazon Web Services (AWS).

CUHK–Shenzhen AI Decision Lab · Full team ↗

Logistics

  • Format: 16 research seminars and 16 guided studios.
  • Materials: English editions followed by Chinese translations; two levels of study.
  • Dates and venue: To be announced. Materials are available for self-study.
  • Access: All core labs include an account-free, low-compute route.
  • Practical course: A guided path for vocational learners and beginners →

Content

What is this course about?

Generated answers increasingly mediate how people discover sources and decide what to investigate next. This course treats Generative Engine Optimization as an end-to-end scientific and engineering problem. We study the path from source eligibility and retrieval to context selection, generation, citation, attribution, and downstream decisions.

The syllabus connects information retrieval, language models, and decision science. We ask which sources an answer uses, what its citations actually support, and how to measure a change under repeated observations. Published studies are paired with small experiments, worked examples, and explicit limits on what the results establish.

Students finish with an auditable evidence package: a query and entity registry, versioned source records, stage-specific measurements, uncertainty and negative-result reporting, a reversible intervention, and a documented decision. The advanced track is designed for senior undergraduates, graduate students, and technical practitioners.

Prerequisites

Readiness diagnostic → · Complete syllabus →

Course Notes

Start with the English edition or its Chinese translation. Choose the Core Notes for theory and research, or the Practice Notes for guided skills development.

01 · ADVANCED STUDY

Core Notes

Twelve chapters, six appendices, and illustrated research examples covering retrieval, attribution, measurement, and controlled experiments.

English PDF ↗中文版 PDF ↗

Coursework

Eight cumulative labs build towards one auditable evidence package.

Assignments

  1. L01 Surface & source auditW01–W02 · 4 hours
    • Record a complete observation context.
    • Resolve visible source identities.
  2. L02 Query-set engineeringW06 · 5 hours
    • Define a query population and strata.
    • Detect duplicates and leakage.
  3. L03 Retrieval & reranking sandboxW03–W04 · 7 hours
    • Run or inspect sparse, dense, hybrid, and rerank stages.
    • Compute Recall@k and NDCG@k.
  4. L04 Claim–Evidence–Source graphW02 & W05 · 6 hours
    • Segment atomic claims.
    • Label entailment, coverage, source quality, contradiction, and absorption.
  5. L05 Controlled content interventionW09–W10 · 6 hours
    • Construct control/treatment assets.
    • Verify factual and accessibility equivalence.
  6. L06 Cross-surface & repeated-run measurementW07 & W13 · 8 hours
    • Analyze a repeated panel.
    • Keep stage metrics and missing states separate.
  7. L07 Drift, adversarial & governance reviewW14–W15 · 6 hours
    • Test synthetic abuse cases.
    • Map controls and residual risk.
  8. L08 Capstone evidence releaseW16 · 18–30 hours across the course
    • Assemble a complete evidence dossier.
    • Pass independent release checks.

Low-compute path for self-study

Every core lab includes an account-free route using supplied fixtures, saved observations, and local notebooks. The practical course offers a browser-and-spreadsheet starting point for learners who are new to coding.

Collaboration and AI use

Discussion and tool assistance are permitted when disclosed. Learners remain responsible for submitted claims, source checks, and interpretations. Keep the original observations, document corrections, and include a short revision note.

Assessment and review

Review begins from the rubric, recorded inputs, and supporting evidence. A second reader should be able to reconstruct the conclusion from the submitted materials. Each lab includes its own deliverables and checkpoints.

Capstone

Assemble an evidence package, a governance memo, and an oral defense. Explain what changed, how it was measured, and what remains uncertain. Capstone specification → Policies →

Schedule

Follow the sixteen-week sequence. Each week links to its lecture, notes, and associated lab; teaching dates will be announced separately.

GEO Advanced Studies · Teaching sequence
#DateDescriptionCourse MaterialsAssignments
01TBAWhat GEO is—and is notWhat is the actual object of optimization?4 h 20 min estimated total
02TBAEvidence before expressionWhich source can support which claim?4 h 30 min estimated total
03TBADiscovery, indexing, and retrievalCan the system obtain the source at all?5 h 10 min estimated total
04TBARanking and context selectionWhy does an eligible source enter the context?5 h 20 min estimated total
05TBAGeneration, citation, and absorptionWas a source merely displayed, or did it shape the answer?5 h estimated total
06TBAQueries as a measurement instrumentWhat distribution do the prompts represent?5 h 15 min estimated total
07TBAMetrics, uncertainty, and repeated observationWhen is a visibility change repeatable?5 h 30 min estimated total
08TBAFrom observation to identificationCan the change be attributed to the intervention?5 h 30 min estimated total
09TBAEvidence-rich content architectureHow can structure improve use without distorting facts?5 h 30 min estimated total
10TBAEntities, authority, and source ecosystemsHow does a fact become corroborated beyond one owned page?5 h estimated total
11TBAMultilingual, multimodal, and agentic surfacesWhat changes when the source is an image or the system can act?5 h estimated total
12TBAOptimization under competing queriesWhich gain survives a portfolio of intents?5 h 20 min estimated total
13TBAField experiments and monitoringHow do we learn under a changing platform?6 h estimated total
14TBAManipulation, poisoning, and defenseWhen does optimization become system abuse?5 h 30 min estimated total
15TBAFair attribution and mechanism designCan visibility incentives reward useful evidence?5 h 20 min estimated total
16TBACapstone synthesis and defenseWhat would make the result worth trusting?18–30 h capstone total estimated total