Pre-register question classes
Brand, category, comparison, local and decision queries; no cherry-picking questions based on favourable results after the fact.
Research and methods hub
AI Search research, sources and public methods
This hub gathers sourced third-party studies, AI Search Lab’s public methods, platform citation breakdowns and repeatable measurement frames. Observations without a complete sample and method are not packaged as original research findings.
Core library
From source distribution and technical conditions to measurement standards
GEO Guide list
Michelin-style editorial star ratings with public citability readiness criteria. Not a paid ranking; reviewed quarterly. Related brands disclose relationships and are not presented as independent endorsements.
Real brand case analyses
Periodic fixed-query sampling of AI answers, broken down by category source layers, with date, platform and limits labelled.
Technical GEO research
Crawler access, server-readable HTML, schema, answer units and entity priorities—how technical work improves citability conditions.
Five-layer technical framework
Retrieval access, structured data, answer extraction, entity evidence, and performance/rendering as an audit frame.
AI citation platforms and sources
How ChatGPT, Perplexity, Google AI Search, Gemini and peers display and select sources—dated, with limits stated.
Public sampling method
Query sets, test conditions, re-run methods, mention and citation definitions, bias and publication standards.
AI Search terminology
Definitions and boundaries for GEO, AIO, AEO, LLMO, mentions, citations and AI referral.
Evidence types
Do not blend three evidence types into one percentage
- Large third-party studies: Use them to understand overall source distribution; samples, dates, platforms and taxonomies from different vendors must not be summed as one ranking.
- Real brand case analyses: Category-level public sampling observations with date and limits labelled—not client results.
- Brand project sampling: Fixed questions to track a named brand—good for baselines, not market-wide claims.
- Simulated scenarios: Explain working methods and possible measurement—not client outcomes or research samples.
Research roadmap
Hong Kong Traditional Chinese and English AI Search baselines
The first planned public study compares brand mentions, cited sources and description differences for fixed Hong Kong buyer questions under Traditional Chinese and English prompts.
Record platform and date
Store each platform, model, language, login state, answer and cited URLs separately.
Publish aggregates and limits
Release aggregates without client-sensitive data; include field definitions, observation windows and what cannot be inferred.
FAQ about AI Search Research, Sources & Open Methods | AI Search Lab
What appears in the research section?
Platform analyses, technical GEO, glossary, public scoring methods, and Hong Kong teaching pieces that include sources, dates, methods, and limitations.
What happened to older AI Search wiki pages?
Legacy material was archived and is no longer the public research core. Still-useful facts are re-verified before returning as core pages.
Do you cite third-party research?
Yes—with source, study date, sample, and scope. English or global percentages are not treated as Hong Kong market results.
Do you publish raw client data?
No identifiable or confidential client data. Public pieces may share methods, anonymized aggregates, or authorized results with limits.
Do study findings hold across all AI platforms forever?
No. Models and indexes change. Findings apply to the stated date, query set, region, and test conditions—not permanent citation guarantees.
Want a baseline of how AI answers describe your brand?
Request a free Audit