Real query screenshots
Keep AI Overview wording and visible source cards, with sampling date. Readers should view the screenshot before the written breakdown.
Periodic research · Real brands named by AI
GEO／AI Search case analyses
Using real Google AI Overview screenshots, we break down why brands are cited—query, descriptive sentences and source layers (Threads, Law Society, Michelin／OpenRice) are all public. These are not fictional client wins.
How this differs from simulated cases: Industry simulations teach diagnostic frames; this section only analyses real brands that appear in real queries.
Fixed breakdown for every issue
Screenshot evidence → brand by brand → why cited → GEO implications
One reading frame across categories: not “who is strongest,” but how answers are composed and which evidence layer they rely on.
Brand by brand
Capture how AI describes each brand and which trust labels it uses—not just a name list. The point is “which frame they were placed in.”
Source layers
Separate UGC, official directories, award platforms and local review sites so not every citation is treated as the same kind of authority.
Actionable implications
Map findings to GEO Guide criteria—signals peers can copy, and boundaries that cannot be faked.
How to read a report card
Read a case analysis without mistaking it for a recommendation ranking
Each report usually includes: the original query, sampling date, AI Overview screenshot, descriptive sentences for named brands, visible source cards, and GEO operational implications. Read by asking four questions in order: What is the query’s hidden intent? Which trust labels did AI use to filter brands? Which layer do visible sources belong to? If you are a peer, are controllable signals on your official site, directories or third-party platforms?
Do not read “being named” as an efficacy guarantee, service ranking or paid placement. Screenshots only prove that, in that sample, for that query, in that interface, these brands and sources appeared; the same query can change the next day. To assess your own brand, build a query set and baseline—do not only mirror one public report.
Start with intent
On the surface it asks “who is good”; underneath it may screen for hard sell, credentials or taste positioning.
Then look at labels
AI descriptions often use frames such as “less hard sell,” “listed in a directory,” or “ratings／reviews”—not abstract adjectives alone.
Check source layers
One answer may combine UGC, official directories and rating sites; weights differ by category.
Extract controllable actions
You can improve entity consistency, answer structure, truthful data and compliant wording—you cannot guarantee the next shortlist.
Screenshot evidence method
How we preserve “what the answer looked like,” not rewrite it later
Case-analysis credibility comes from an inspectable evidence chain—query, date, visible interface and source cards—not a summary sentence written afterwards.
| Step | What we do | Why it matters |
|---|---|---|
| Choose high-intent queries | Prioritise local decision questions over brand self-claims | Closer to real buyer shortlist behaviour |
| Sample a fixed interface | Record platform (e.g. Google AI Overview), visible region and date | Answers change with time and UI; they must be traceable |
| Save full screenshots | Include answer body and visible source cards／side panel | Lets readers check “why cited,” not only brand names |
| Record descriptions brand by brand | Transcribe AI wording and trust labels as shown | Stops answers being rewritten as marketing copy later |
| State limits | Mark non-client results; not efficacy／service endorsements | Prevents reports being misread as paid rankings or endorsement |
Screenshots are sample evidence, not a statistical census. Brands that do not appear are not necessarily “doing poorly”; brands that appear are not permanently cited. Method details: research methodology and editorial policy.
Category source-layer comparison
Beauty, lawyers, dining: same GEO, different evidence layers
Public reports show AI Overview leans on different trust infrastructure by category. Before copying a strategy, copy the right source layer.
| Category | Example query type | Common source layers | GEO implication (not a guarantee) |
|---|---|---|---|
| Local beauty | Decision questions such as “want a facial—who is good?” | Threads and other UGC／trust-label discussion | Real word-of-mouth frames in communities (e.g. less hard sell) may enter answers; empty efficacy claims on official sites may miss intent |
| Lawyers／legal services | Finding a lawyer; verifying practice information | Law Society and other professional／official directories | Verifiable credentials and directory consistency often matter more than promotional case studies |
| Dining／food shops | Recommendations, cuisine type and local food decisions | Michelin-style awards＋OpenRice and local review platforms | Awards, menu positioning and platform data need to align; official-site self-description is not enough |
That is this hub’s core teaching: do not force another category’s strategy onto yours. Beauty brands do not have a Law Society directory to lean on; law firms should not pretend Threads discussion equals statutory credentials. Full breakdowns are in each issue; for copyable signal lists see the GEO Guide list and Hong Kong GEO in practice.
What this hub is not
Clear boundaries so research is not read as a paid wins wall
Not a paid ranking
Analysed brands do not appear in AI Overview because they paid AI Search Lab; reports only observe public query results.
Not fictional client wins
We do not narrate “we helped brand XX get listed.” Client work and public research stay separate; teaching simulations live in case studies.
Not efficacy or service endorsement
Naming a beauty brand, law firm or food shop is not a consumption recommendation or quality guarantee.
Not proof of permanent ranking
Screenshots correspond to the sample moment. After models, indexes and source pools change, answers can look entirely different.
Not a search-volume report
Topics are chosen for teaching and source-layer contrast—not claimed as Hong Kong’s highest-traffic terms.
Not a full market census
Each issue deep-dives a few brands and one query—not every supplier in the category.
Issue index
Published real-brand analyses
案例分析：想做 facial 邊個好？spa ph+、EVRbeauty、OASIS medical
以 Google AI Overview 真實結果拆解「想做facial，邊個好」：逐個分析 spa ph+、EVRbeauty、OASIS medical 被點名的原因，以及 Threads／UGC 如何成為引用來源。
案例分析：香港律師推薦——的近、柯伍陳、葉謝鄧
拆解 Google AI Overview 對「香港律師推薦」的真實答案：的近律師行、柯伍陳律師事務所、葉謝鄧律師行為何被點名，以及香港律師會目錄如何成為權威來源。
案例分析：香港美食小店——媽咪雞蛋仔、一樂燒鵝、呂仔記
以「香港美食小店」AI Overview 為例，逐個分析媽咪雞蛋仔、一樂燒鵝、呂仔記被引用的理由：米芝連背書、招牌產品描述、地區實體，以及 OpenRice 等本地平台角色。
Cadence and category coverage
How reports publish—and how to suggest the next category
This hub updates as periodic research: a new issue ships when there is a high-intent local query worth teaching from, savable screenshot evidence, and clear source-layer contrast value. Cadence prioritises quality and evidence completeness over daily volume. Each piece labels sampling date and, in title and excerpt, the category and query focus.
If you want a Hong Kong category (for example insurance comparison, home services, education centres, B2B SaaS) considered for a later public breakdown, raise it through Audit／service consultancy. We assess whether the query has buyer intent, whether the source layer has teaching value, whether screenshots can be shown publicly, and whether regulated content needs extra review. Selection is not a service commitment and does not guarantee your brand will appear in that sample.
After reading public reports, to measure your own brand: build a query set and baseline with the same method, then decide whether to fix entities, answer pages or external sources. Start with AI Visibility Audit; service options: Hong Kong GEO services; writing and structure: guides.
Next steps
After category observation, measure your brand with the same method
Case analyses explain why those brands entered answers; whether your brand is mentioned and described accurately needs a dedicated query set and baseline. Public reports give you a source-layer map; Audit gives you an actionable gap list.
FAQ about GEO / AI Search Case Reports | Real Brand Citation Breakdowns
Are these client success stories?
No. They are editorial analyses of public AI Overview results and do not imply commercial relationships with named brands.
How do they differ from industry scenario studies?
Scenario pages teach frameworks with simulated brands. This section uses real queries, screenshots, and brand names to explain one answer’s citation logic.
How often are reports updated?
We aim to add categories and queries over time. Each report dates the sample because Overviews change.
Can we request a category?
Suggestions are welcome. Inclusion depends on intent, screenshot verifiability, risk, and editorial priority.
Do these reports endorse the brands named?
No. Naming reflects what the AI surface showed in that sample—not a recommendation from AI Search Lab.
Want a baseline of how AI answers describe your brand?
Request a free Audit