Best supplement brand for Singapore men over 40?

local services case

Say this

The 悟空净耳 brand ranked first and was recommended ahead of others by the AI platform

Why it works

Before → after: asked 'which are the well-known ear-cleaning chains in China', the AI's numbered list before optimisation did not include 悟空净耳, which only appeared in passing at the end; after optimisation, asked 'which are the better ear-cleaning chains in Wuhan', the first item is 悟空净耳, restating its directly-operated chain and visual ear-cleaning positioning (note that the before and after queries differ in scope: nationwide vs Wuhan). Platform: conversational AI search. Queries: 中国比较知名的净耳连锁有哪些 / 武汉比较好的净耳品牌连锁有哪些.

Proof: Measured in position 1 of the answer list

Files: Case Book - DeepSeek, Doubao and Baidu · Case Book - DeepSeek, Doubao and Baidu

supplement case

Say this

49 long-tail queries checked one by one — every blank cell is an opportunity

Why it works

Before → after: the brand only knew whether it was mentioned in AI answers at all, not which queries hit and which did not; the monitoring system marks 49 long-tail queries hit-or-miss one by one — 斯维诗 (Swisse) measured 14 hits out of 20 queries while 5 tracked competitors were blank — and the blank cells become the next optimisation list. Platform: GEO monitoring system (Doubao, DeepSeek and other AI platforms). Sample query: 鱼油品牌性价比排行榜.

Proof: 49 long-tail queries monitored in total

Files: Case Book - DeepSeek, Doubao and Baidu · Case Book - DeepSeek, Doubao and Baidu

family law case

Say this

Second place one month after optimisation

Why it works

Before → after: for 'where to find a professional cross-border (HK) divorce lawyer in Shenzhen', Doubao before optimisation gave only generic pointers such as the bar-association directory, with no individual lawyer; one month after optimisation its answer to 'recommended Shenzhen cross-border divorce lawyers' measured lawyer 陈薇 in second position, restating her end-to-end cross-border divorce practice and 100+ HK-related cases. Platform: Doubao. Queries: 深圳专业涉港离婚律师哪里找 / 深圳涉港离婚律师推荐.

Proof: Position after optimisation: 2nd

Files: Case Book - DeepSeek, Doubao and Baidu · Case Book - DeepSeek, Doubao and Baidu

medical aesthetics case

Say this

Ranked first and recommended ahead of others by the AI platform

Why it works

Before → after: for the query 'which plastic-surgery clinic in Luoyang is good', DeepSeek's answer before optimisation contained no client brand at all and listed only public-hospital departments; after optimisation the same platform's recommendation list measured 洛阳孔繁荣整形 in first position, restating its national 4A-grade accreditation and 30-year history. Platform: DeepSeek. Query: 洛阳整形哪家好.

Proof: 洛阳孔繁荣整形 measured in position 1 of the recommendation list

Files: Case Book - DeepSeek, Doubao and Baidu · Case Book - DeepSeek, Doubao and Baidu

medical device case

Say this

Ranked first, with the brand's reputation strengthened

Why it works

Before → after: for the query 'how is 瑞莱思', DeepSeek's answer before optimisation gave first place to an unrelated accounting-training school of the same name, and the medical phototherapy-device brand had no presence; after optimisation the first item in the answer is 湖南瑞莱思医疗器械 (rated 'highly recommended'), citing its 2025 first place among the top ten vitiligo phototherapy-device brands. Platform: DeepSeek. Query: 瑞莱思怎么样.

Proof: 瑞莱思 measured in position 1 after optimisation

Files: Case Book - DeepSeek, Doubao and Baidu · Case Book - DeepSeek, Doubao and Baidu

pharma case

Say this

Pharmaceutical: content corrected, key advantage information now surfaced

Why it works

A prescription-drug client (an oncology drug, brand masked in the book). Before → after: for the query 'is the raw material of this domestic-brand drug imported', the AI answer before optimisation said it was not — a factual error; after optimisation the Baidu AI+ answer opens with 'Yes', explains that the raw material is imported and the drug is synthesised domestically, and cites its liposome technology as a key advantage. Platform: Baidu AI+. Query: 国产品牌药品原料是否进口.

Proof: Baidu AI+ answer after optimisation summarised 2 sources and opened with 'Yes'

Files: Case Book - DeepSeek, Doubao and Baidu · Case Book - DeepSeek, Doubao and Baidu

Have you worked in my industry?

Say this

Licensed lending, fine dining and omakase, 3C export, education franchises, clinics and medical devices, supplements, professional services and personal brands — the trigger is the shape of the buying decision, not the sector on your invoice.

Why it works

Delivered in 2026 (deck page 7): licensed lending, fine dining, omakase, 3C export, an education franchise, plus one platform partnership. Case Book 5 (DeepSeek, Doubao, Baidu): aesthetic clinics, medical devices, prescription drugs, supplements, local services, education, industrial manufacturing, professional services, and a lawyer and a doctor building visibility under their own names. Case Book 4 (ChatGPT, Google AI, Gemini): cross-border e-commerce and logistics, SaaS and developer tools, fintech and cross-border payments, smart hardware and industry, consumer goods and education — delivered through the Canlah delivery network, never named.

Files: Case Book - ChatGPT, Google AI and Gemini · Case Book - ChatGPT, Google AI and Gemini · Case Book - DeepSeek, Doubao and Baidu · Case Book - DeepSeek, Doubao and Baidu

Who have you worked with?

Say this

“Canlah’s own clients include GalaxyCredit, ilLido, Sushi Ryujiro and SmallRig, plus the clients in this case book. Separately, the founding team previously served brands including Haleon, Starbucks, ByteDance and Alibaba Cloud — that is the team’s history, not Canlah’s client list.”

Why it works

Those client names are cleared, so use them. The second half is the part that must stay exact: the founding team's history is not Canlah's client list, and saying otherwise would not survive one follow-up question. And never name a brand from Case Book 4 — those are the delivery network's clients.

How do you charge?

Say this

Billed per intent (shelf position): one intent = one class of real buying question. At delivery it is split into 5–10 fixed phrasings, measured and optimised together; multiple phrasings of the same intent are billed once.

Why it works

Positioning: get the brand into AI's recommended answers — every line countable, verifiable and written into the contract. Billing unit = intent (one real buying situation); each intent is split into 5–10 phrasings, measured and optimised together. Fixed query pool — Each intent split into 5–10 real phrasings with language, region, engine and sampling rules fixed, so results stay comparable over time. Billing unit — 2–4 / 6–8 / 10–13 intents by tier

Files: What You Actually Get - Full Scope List · What You Actually Get - Full Scope List

What tiers do you offer?

Say this

Accepted intent quota · 2–4 · 6–8 · 10–13

Why it works

FOUNDATION · ENTRY · ★ AUTHORITY · RECOMMENDED · DOMINANCE · CATEGORY LEADER Best for · One product · one market · Multiple lines · two markets · Group · category leader Focus · Conversion intents — take the shelves closest to revenue first · Conversion + comparison, while blocking rival interception · Category-wide intents onto the recommendation list AI engines covered · ChatGPT · + Google AI Overviews and AI Mode · All major + emerging engines Languages · English · English · Chinese optional · Multilingual Acceptance · Semantic mention rate ≥30% above signing baseline · Same · Same

Files: What You Actually Get - Full Scope List · What You Actually Get - Full Scope List

What if it doesn't work?

Say this

“If the accepted buying questions miss the standard, we deliver an additional quarter at no charge — once. It is additional delivery, not a refund.”

Why it works

Anything shorter — “you’re covered”, “money back” — creates a refund expectation that contradicts our own terms.

What is a buying question?

Say this

Best luxury smartphones / top 10 luxury smartphones / which luxury smartphone brand is most highly rated?

Why it works

Five funnel layers: L1 · Category Awareness / L2 · Non-branded Recommendation / L3 · Scenario & Subcategory / L4 · Branded Comparison / L5 · Brand Verification Buying intent library — Real buyer intents identified across recommendation, comparison, supplier, situation and purchase-risk angles, each ticked into a contract annex. Fixed query pool — Each intent split into 5–10 real phrasings with language, region, engine and sampling rules fixed, so results stay comparable over time.

Files: Example - The Buying Questions We Win For You

What is included under AI visibility?

Say this

Real buyer intents identified across recommendation, comparison, supplier, situation and purchase-risk angles, each ticked into a contract annex.

Why it works

Buying intent library — Real buyer intents identified across recommendation, comparison, supplier, situation and purchase-risk angles, each ticked into a contract annex. (2–4 / 6–8 / 10–13 by tier) Fixed query pool — Each intent split into 5–10 real phrasings with language, region, engine and sampling rules fixed, so results stay comparable over time. (All intents measured monthly) Multi-engine probing — Answers, cited sources and brand position recorded separately per AI entry point: full multi-engine probing plus browser-level spot checks. (Engine coverage by tier) Competitor & source map — Identify the competitors, media, directories and third-party databases AI cites most, and map the citation chain. (Tracks direct rivals and authority sources) Re-testing & evidence archive — Original questions, answers, citation links, timestamps and regions all retained. Every conclusion can be re-run and audited. (Monthly sampling + quarterly full re-test · fully archived) Brand fact base — One source of truth for company, product, credential and contact facts, so AI is less likely to answer wrong or quote stale information. Facts require your verification. (Built and maintained) Product / service knowledge matrix — Product families, specs and use cases organised into a structure AI can read and verify. Requires your product material. (Covers key product lines) Trust evidence pages — Strengthen credentials, cases, team and update dates into a citable chain of trust evidence. Requires your material. (Upgraded by tier) Crawl & citation accessibility — Make sure important content is publicly crawlable, indexable and clearly written, and cut low-value pages. (Ongoing · quarterly review) Answer content — Answers buying decisions directly — how to choose, what differs, how to verify — with evidence. (Monthly by tier · counted in the content total, never twice) Buying Q&A clusters — Each cluster builds a main answer, follow-up questions and evidence links around one real buying intent. No keyword stuffing. (Configured by tier) Original citable assets — Your own data and experience turned into information that is scarce online — comparison tables, field data. Requires your source data. (Configured by tier) Freshness & fact checks — Track stale information and wrong descriptions; date-stamp high-value pages and keep review records. (Monthly) Monthly AI visibility report — Brand and competitor mention rates, recommendation position, cited sources, factual accuracy, and the actions taken. (12 reports/year + 4 quarterly reviews) Google AI performance — AI-related performance data available in Search Console, cross-checked against the fixed query sampling. (Tracked monthly) AI source attribution — Identifiable AI-sourced visits and enquiries tagged in GA4. Where a platform does not expose data, we do not invent it. (1 attribution dashboard)

Files: What You Actually Get - Full Scope List · What You Actually Get - Full Scope List

What is included under Annual target?

Say this

Fixed query pool, fixed entry points, averaged over multiple rounds across 30 consecutive days. A single screenshot is not acceptance. Baseline = the snapshot at signing, reconfirmed in month one. Your obligation: provide site access and material inputs on schedule.

Why it works

Semantic mention rate — Fixed query pool, fixed entry points, averaged over multiple rounds across 30 consecutive days. A single screenshot is not acceptance. Baseline = the snapshot at signing, reconfirmed in month one. Your obligation: provide site access and material inputs on schedule. (≥30% above the signing baseline) Brand fact accuracy — Company, product and contact details in AI answers checked against the fact table. (Reported monthly · not an acceptance metric) Site citation rate — Share of answers in the fixed query pool that cite the site directly. (Reported monthly · not an acceptance metric) Competitor share of voice — Brand vs competitor mention share across the same query pool. (Reported monthly · reviewed quarterly)

Files: What You Actually Get - Full Scope List · What You Actually Get - Full Scope List

What is included under Off-site authority?

Say this

A monthly off-site outreach list — where to go and in what order — based on the citation map.

Why it works

Source map & outreach list — A monthly off-site outreach list — where to go and in what order — based on the citation map. (Updated monthly · included) Off-site content briefs — What to publish and how to write it, with draft material your team can execute directly. (Delivered with the outreach list · included) Trade media & guest articles — Pursue genuine mentions and indexing in the trade media and professional channels AI actually reads. (Quoted per project · outreach logged monthly) Directory & entity consistency — Align name, address, description and links; clear out stale records; place listings in industry directories and databases. (Quoted per project) Flagship content assets — Original reports, white papers or major selection guides that earn citations over time. (Quoted per project) Real community presence — Community work under real identities with explicit disclosure throughout; the disclosure clause goes into the contract. (Quoted per project) What we will not do — No bought accounts, no gamed Q&A, no fake reviews, no undisclosed native placements. Every external item keeps its source and publication record. (Fully logged)

Files: What You Actually Get - Full Scope List · What You Actually Get - Full Scope List

What is included under On-site?

Say this

Full check of crawl status, index coverage, duplicate pages, canonicals, redirects, 404s, parameter URLs, sitemaps and server status.

Why it works

Full technical audit — Full check of crawl status, index coverage, duplicate pages, canonicals, redirects, 404s, parameter URLs, sitemaps and server status. (Quarterly review · 4×/year) Page value tiering — Tiered by traffic, ranking, commercial intent and enquiry contribution: promote, keep, merge, rewrite or retire. High-value pages first. (Full-site scan · high-potential pages first) Structured data — Verify and fix Organization, Product, FAQ, Breadcrumb and Article templates. We only mark up what is genuinely visible on the page. (5 core templates) Core Web Vitals — Optimise core templates around LCP, INP, CLS and mobile experience; baseline plus two re-test rounds. (1 baseline + 2 re-tests) Tracking calibration — Calibrate GSC, GA4 and conversion events (form / email / phone / WhatsApp), separating branded from non-branded sources. Requires your account access. (1 tracking setup + monthly checks) Commercial keyword base — Covers head terms, long-tail, buying questions and situational needs, tagged by search intent, difficulty and owning page. Shares one base with the intent library. (Sized by tier · continuously expanded) Keyword-page mapping — Assign a primary and supporting page to each head term so pages don't compete for the same keyword. (High-potential mappings first) Homepage & hub lift — Rework value proposition, product coverage, trust elements and enquiry entry points so the homepage carries weight to inner pages. Requires site access. (Ongoing · quarterly review) Pillar pages & topic hubs — Build pillar pages around core products/services and buying-decision topics that can hold head terms. (Quarterly output by tier) Internal link equity — Optimise navigation, breadcrumbs, related links and internal anchor text so equity flows to high-value pages. (Ongoing · logged monthly) Deep optimisation of key pages — Titles, descriptions, heading structure, FAQ, image alt text and enquiry paths, item by item. (Monthly by tier) Expert content assets — Non-generic content built from your own expertise: selection guides, comparisons, pitfalls, checklists. Requires your source material and fact review. (Monthly by tier · counted jointly with GEO content) Enquiry landing pages — Highly relevant landing pages by key product, market and buying situation, with trust elements and enquiry conversion. (Configured by tier) Original visual assets — Turn specs, comparisons and processes into citable tables and diagrams that hold attention and get quoted by AI. (Configured by tier) Content refresh — Each month, find pages losing clicks, stalling in rank or going stale; refresh data, structure and internal links. (Monthly · 12×/year)

Files: What You Actually Get - Full Scope List · What You Actually Get - Full Scope List

What is included under Scope?

Say this

Goal is entering AI's recommended answers: on-site and AI-visibility engineering as the main line, supported by off-site outreach, reviewed and accepted quarterly.

Why it works

Annual strategy — Goal is entering AI's recommended answers: on-site and AI-visibility engineering as the main line, supported by off-site outreach, reviewed and accepted quarterly. (Monthly service · quarterly acceptance) Billing unit — Billed per intent (shelf position): one intent = one class of real buying question. At delivery it is split into 5–10 fixed phrasings, measured and optimised together; multiple phrasings of the same intent are billed once. (2–4 / 6–8 / 10–13 intents by tier) Role coverage — Six roles — intelligence, strategy, content, pages, data, engineering — run on one live dataset via our AI staff system. Strategic judgement, relationship management and compliance stay with people. (6-role system + human judgement) What the fee covers — The fee covers every item under on-site engineering and AI visibility engineering, plus the off-site recommendation list. Off-site execution, media spend, platform account and tax fees, association memberships, third-party development and certification are billed separately. (Boundary itemised in the contract)

Files: What You Actually Get - Full Scope List · What You Actually Get - Full Scope List

Can I read the whitepaper?

Say this

The full argument in one page: every conclusion in this report, each carrying its evidence tier.

Why it works

The whitepaper 'Marketing in the Agent Era' (13 chapters, EN and ZH) is publicly readable at https://canlah.ai/whitepaper/read/en/ — every chapter is on the open web with no form and no sign-up. Send the link; the Executive Summary is one page.

Files: Executive Summary · Chapter 0 · You are probably already on this table

How is this different from my SEO agency?

Say this

SEO gets you onto page one of ten blue links. We get you onto the two to five names the assistant actually reads out — and we measure it every month.

Why it works

It was never GEO versus SEO; it is a question of sequence. Ranking without clicks is a demand problem. Being retrievable without ever being named is a sourcing problem. No amount of the first kind of work fixes the second kind of failure. We look at both layers in one pass and will say so if the foundation is what needs the money.

Files: GEO vs SEO — do you need the other?

Should we only optimise for ChatGPT?

Say this

Optimising for ChatGPT alone covers 53.9% of the market today, not the roughly 76% it was a year ago.

Why it works

AI search is no longer a single-platform game — Source: Momentic 汇编 Similarweb 估算,2026 年 5 月(B)—— 分级取自 §1.4 表注

Proof: ChatGPT: 53.9%

Files: Executive Summary

Is zero-click search really rising?

Say this

The losses sit on Google's own results page — a rise from 60.45% to 68.01% cannot be explained by an AI Mode that is 0.34% of search volume.

Why it works

Zero-click keeps climbing; the attribution does not — Source: SparkToro / Similarweb 面板(B)—— 分级取自 §1.1 原文;摘要正文本身未标分级

Proof: Jan–Apr 2026 · US Google searches producing no click: 68.01%

Files: Executive Summary

Does GEO actually work?

Say this

‘Originally ranked’ means the position in Google Search results when entering the candidate set (the study’s pipeline retrieves the top 5 sources via Google Search) — not a ‘rank position in AI’. The rise and the fall come from the same experimental condition: GEO reallocates citation share within the candidate set — redistribution, not growth. +115.1% never appears alone.

Why it works

Both halves of the equaliser effect: the site originally ranked 5th +115.1%, the site originally ranked 1st −30.3% — Source: arXiv:2311.09735, KDD ’24 (A, the only peer-reviewed source cited in this report)

Proof: Site originally ranked 5th (single optimisation: citing sources): 115.1% · Site originally ranked 1st (all sources optimised): -30.3%

Files: Chapter 3 · What GEO Can Actually Change, and What It Cannot

Does AI traffic convert?

Say this

The reversal matters more than the absolute numbers: the same source went from converting 38% below non-AI sources to 31% above (54% above on Thanksgiving Day), with time on site up 45% and bounce rate down 33% — from a base that remains small.

Why it works

AI referral traffic went from worse traffic to better traffic — Source: A · Adobe Analytics(口径覆盖约 1 万亿次访问 / approximately one trillion visits)

Proof: 2025 holiday season · conversion vs non-AI sources: +31%

Files: Chapter 5 · Discovery Moved Forward, Checkout Moved Back: The Real State of Agentic Commerce

Is trade-media content still worth paying for?

Say this

This is why placing content in trade media needs re-pricing: reach generated outside the platform's own product, by AI invocation, already equals 24.9% of its app traffic.

Why it works

AI-mediated reach is approaching a quarter of a platform's own app traffic — Source: QuestMobile《2026 年 AI 应用市场发展半年报》测算,2026 年 7 月 14 日发布(A)/ QuestMobile, AI Application Market Half-Year Report 2026, published 14 July 2026, estimates (A)

Proof: Autohome: equivalent share of its own app traffic: 24.9%

Files: Chapter 6 · How Closed Ecosystems Decide What to Trust: Vertical Authority and Ecosystem Affiliation

Which platforms do Chinese AI engines cite?

Say this

Within one vertical the citation rate steps down from 84.5% to 54.4%, and a single model can concentrate 76.5%–80.6% of its citations on one platform — presence on those two or three platforms decides citation, not platform count.

Why it works

Automotive vertical: citation weight concentrates on two or three authority platforms — Source: QuestMobile《AI 平台采信逻辑与信源偏好研究报告》,数据周期 2026 年 4–5 月(A)/ QuestMobile, Research Report on AI Platform Trust Logic and Source Preference, April–May 2026 (A)

Proof: Autohome: 84.5%

Files: Chapter 6 · How Closed Ecosystems Decide What to Trust: Vertical Authority and Ecosystem Affiliation