Client Cases

Measured. Not narrated.

Canlah AI's case page documents anonymised diagnostics for a Singapore omakase restaurant, a premium restaurant group, a children's education franchise entering Malaysia, a marketing-software company and two consumer AI apps, plus Canlah AI's own record: on 15 October 2026, ChatGPT used without signing in named Canlah AI in 9 of 16 answers to a Chinese-language question about which Singapore companies can get a brand recommended by AI search. Every card states the date, AI engines, market and denominator, and traces to answer records that can be re-checked.

Some cases are anonymised to protect confidentiality. Every number on this page traces back to a reviewable evidence archive — answer records, screenshots, timestamps.

Our own record Canlah AI · Singapore

ChatGPT named Canlah AI in 9 of 16 answers

We measure ourselves too. This is one buyer question, asked in Chinese in ChatGPT on the web without signing in.

The question (asked in Chinese)

新加坡有哪些公司能让品牌在 ChatGPT 等 AI 搜索的回答里被推荐？请列出几家，每家附一句理由。

“Which companies in Singapore can get a brand recommended in the answers of AI search such as ChatGPT? List a few, with one reason each.”

9 / 16

Answers that named Canlah AI

9 / 9

Of those, answers that cited canlah.ai as a source

What ChatGPT said about us (translated from Chinese; “…” marks a cut)

“Provides GEO services specifically, tracking ChatGPT, Gemini and Google AI answers…”

“Tracks how several AI engines answer about a brand and its competitors, then adjusts website content and outside sources based on the results; suited to brands that care about measurable results.”

How it was measured

ChatGPT on the web, not signed in, two browser profiles, Singapore connection. 15 October 2026, 04:32–13:53 Singapore time. 16 answers in the main sample: rounds with incomplete data were dropped by a fixed rule. Results differ by question, language and account type; each is reported with its own denominator.

Earlier record

August 2026: 5 buyer questions × 3 AI engines, 15 samples, not signed in. canlah.ai cited as a source: 0. Brand named: 1, and that one came from a third-party list. The questions and engines differ from October, so these are two separate measurements, not a before-and-after.

Flagship diagnostic Fine dining · Singapore

A high-end omakase restaurant (S$300+ per guest, anonymised)

Fully booked offline, near-invisible in AI answers. We measured exactly how wide the gap is — against a directly comparable peer.

Same price band, same segment — measured side by side

Peer Japanese restaurant — same price band, likewise no Michelin star

This restaurant

Mentioned by AI across 32 non-branded answers on three engines

34%

11 / 32

1 / 32

Google AI Mode checks

4 / 4

named every time

0 / 4

never named

2 / 7

Website pages indexed by Google

Restaurant structured data found on the site

Verification

Every finding was reproduced word for word in a real signed-in browser. The client can repeat the checks on their own phone.

Deliverable

A bilingual (EN/CN) diagnostic audit and a 90-day remediation plan, ordered by priority.

Group diagnostic Premium restaurant group · Singapore

A 20-year premium restaurant group (one parent site + four sub-brands, anonymised)

Two decades of offline reputation, none of it legible to AI engines or search crawlers. We measured the group and each brand separately.

3,452

Genuine reviews across its restaurants, each rated 4.5★ or higher

0 / 49

Recommendations received across 49 non-branded questions asked on three AI engines

16.8s

Mobile first-screen load time — the healthy threshold is 2.5s

Pages of the group portal indexed by Google

Deliverable

Ten bilingual diagnostic audits (group + four brands) with a complete evidence chain.

Landing-page rebuild Children's education franchise · Malaysia market

A children's education franchise brand (anonymised)

Buyer questions in the measured set

Page-level audit checks

Citation sources placed

A dead canonical tag had kept the page at zero indexation, and the WhatsApp contact link was broken — both diagnosed and fixed. A before/after comparison site documents every change, auditable commit by commit.

AI-answer diagnostics · software and apps

Where three brands stood before any work started

Three brands we audited in October 2026 (one client, two we audited ahead of any engagement), anonymised. Each card is a diagnosis, the starting point, not a result.

Diagnosis 13 Oct 2026

Marketing software · United States

A software company selling affiliate and influencer marketing to brands

0 / 185

Answers that named the company, across 32 buyer questions

137 / 185

Answers that named the category leader

0 / 130

“Best platform” and comparison articles that mention it

0 / 30

Buyer searches on Google where it reached the top 20

What we found

Asked about the company by name, 6 of 6 answers about its reviews pulled in unrelated businesses with the same name.

What we produced

A full diagnostic report and a six-month plan built on four buyer questions.

Diagnosis 11 Oct 2026

AI study app · built in Singapore, selling to American students

A Singapore-built AI homework and study app

0 / 186

Answers that named the app, across 31 buyer questions

137 / 186

Answers that named the most-recommended app

Times its own website was cited in those 186 answers

17 / 24

Answers about the app by name that brought in other businesses with the same name, or asked which one was meant

What we found

Smaller rivals were cited again and again through comparison and “best app” articles they publish on their own sites.

What we produced

A diagnosis and a six-month GEO plan for the American student market.

Diagnosis 11 Oct 2026

AI music app · United States

An AI song-generation app

3 / 178

Answers that named the app, across 30 buyer questions (ChatGPT: 0 of 89)

145 / 178

Answers that named the category leader

0 / 41

Comparison and “best of” pages cited by AI that mention the app

14 / 24

Answers about the app by name that cited similarly named websites

What we found

On the licensing question, 6 of 6 answers cited similarly named websites, and ChatGPT answered with a similarly named site's terms page.

What we produced

A diagnosis and a GEO plan built on four buyer questions.

Method for all three

ChatGPT (gpt-5.5) and Gemini (2.5 Flash) through their APIs with web search on, English, United States market, every question asked 3 times. Mentions were recounted from the raw answers with strict name matching. Buyer questions that already named a brand were left out of the headline count. The Google rankings and the 130-article count are one-time checks made the same day.

One pattern ran through all three: when buyers asked about the brand by name, AI mixed it up with other businesses or similarly named websites. Each plan covers answer pages, placements in the lists AI reads, telling the brand apart from similarly named sites, and a monthly re-measurement.

Clients we've served