JUN &
GEO
Get your brand mentioned, cited, and recommended by AI. Buyers now ask ChatGPT, Perplexity, and Gemini for product recommendations before they visit a single website. If the answer names your competitors and skips your brand, the deal is lost before it starts. I am Jun Liu, a GEO consultant based in Cixi, Zhejiang, China, and the founder of AccessSmith. I measure how AI platforms talk about your brand — whether you are mentioned, cited, and recommended — then fix the gaps page by page and source by source. After 22 years in product management and 13 years in e-commerce, I build every engagement on one rule: visibility you cannot measure is visibility you do not have. My audits cover both layers of your presence — the brand itself and each product or SKU — across the six platforms your customers actually ask.
Years — ProductProduct management, 2003–
Years — E-commerceSelling online in China, 2010–
Angles measuredMentioned / Cited / Recommended
Platforms auditedPer platform, never averaged
MethodThree steps — Audit / Fix / Monitor
Specimen 01Every engagement follows the same order. First we find out exactly what AI platforms say about you today. Then we fix the pages, facts, and missing signals that keep you out of the answer. Then we keep measuring, because AI answers change every month and a one-time fix quietly decays.
Audit
I ask the questions your buyers ask, on all six platforms, and record whether your brand and each product are mentioned, cited, and recommended — with the full answers saved as evidence.
Fix
You get a prioritized fix list: pages to rewrite in answer-first form, wrong facts to correct everywhere they appear, and third-party sources to earn. I can execute the fixes with you or hand the list to your team.
Monitor
The same buyer questions run on a schedule, so you see your mention, citation, and recommendation rates move over time — and catch it early when an AI platform starts describing you wrong.
FrameworkThree angles × Two layers × Six platforms
Specimen 02What does a GEO audit measure? Three different things that most reports blur into one score. Mentioned means an AI platform names your brand when buyers ask for options in your category. Cited means the platform uses your own pages as sources and links to them. Recommended means the answer's text actually puts your brand forward as a good choice — a stronger signal than a citation, because AI answers often cite one vendor while recommending another.
Each angle is tested separately, for your brand and for every product or SKU, because the fixes differ: brand gaps need third-party signals, while product gaps need pages worth citing. Results are reported per platform, never averaged into a single vanity number.
MentionedSays you
Does the AI answer name your brand at all when a buyer asks for options? If you are not in the answer, nothing else on this page matters yet.
CitedSources you
Do your own pages appear in the answer's sources? Citations are the link between AI visibility and your website — they are what a buyer can click.
RecommendedPushes you
Does the answer text genuinely recommend you — “best for”, “top choice”, “we recommend”? Being cited while a competitor is recommended is the most common GEO failure.
| Platform | Brand layer | Product layer | ||||
|---|---|---|---|---|---|---|
| Platform | Mention | Cite | Recommend | Mention | Cite | Recommend |
| ChatGPT | — | — | — | — | — | — |
| Perplexity | — | — | — | — | — | — |
| Copilot | — | — | — | — | — | — |
| Gemini | — | — | — | — | — | — |
| Grok | — | — | — | — | — | — |
| Claude | — | — | — | — | — | — |
[1] This is the measurement structure, not a results screenshot. Cells stay empty on purpose: rates exist only after an audit runs your buyer questions on your brand. No estimated or illustrative numbers are published here. The brand layer measures your company name — AI awareness, share of voice, factual accuracy. The product layer measures each product or SKU on its own, because AI can know your brand and still never surface the product a buyer would actually purchase.
BlogResearch / Failures / Milestones
Field LogThe lab notebook, published. Every entry labels its evidence: published research is credited to its author; measurement findings are marked as ours. All nine entries below are published in full at the research blog, and every article is licensed CC BY-SA 4.0.
White Paper — 旗舰V1.0 · 2026-10-20
WP Measured, Not Guessed: A Field Methodology for AI Visibility (GEO)The full methodology in one document: two-part measurement, three rates, three-sample discipline, the cognition audit, fix operations, and the reporting standard. CC BY-SA 4.0. White Paper 2026-10-20Failures — 翻车记录03 PUBLISHED
F.01 What AI Thinks Your Factory MakesThree AI systems described one Cixi appliance OEM three different wrong ways. Mention 100%, citation 0% — every system knew the name; none knew the company. Failures 2026-10-20 F.02 Deleted in Five Days: Our Wikidata Item Didn't SurviveWe built a Wikidata item for a client factory on October 12. On October 17 it was deleted for failing the notability policy. A post-mortem against the policy's three criteria. Failures 2026-10-20 F.03 We Scored AI Answers From a Single Sample for Four VersionsFor four versions of our scoring system, every visibility number came from running each prompt once. What that cost, how we caught it, and what changed. Failures 2026-10-20Milestones — 达成记录03 PUBLISHED
M.01 The Two-Part GEO Audit: Exposure × CitationOur audit framework in two parts — exposure (awareness, share of voice, sentiment) and citation probability (citation rate, presentation quality) — measured on a 2×2 grid. Milestones 2026-10-20 M.02 From v1 to v5: Building an AI Visibility Scoring SystemA citation field that stored nothing, a third angle nobody was measuring, and a sampling fix. Five versions of the scoring system, and what each one taught. Milestones 2026-10-20 M.03 Five Minds, One PracticeKing on mechanism, Indig on measurement, Solis on execution, Fishkin on zero-click influence, Lily Ray on evidence. Where they agree, where they fight, and how the fights stay in the method. Milestones 2026-10-20Research — 技术研究03 PUBLISHED
R.01 Cited Is Not RecommendedLily Ray's retests: vendors cited by AI answers were still excluded from the recommendation in 69% of cases (Jun 2026), rising to 83% (Sep 2026). Research 2026-10-20 R.02 One Sample Lies: Why We Run Every Prompt 3 TimesKevin Indig's finding: the same prompt run three times produced consistent citations only 2.2% of the time. Prompt tracking is polling, not rank tracking. Research 2026-10-20 R.03 llms.txt: Necessary, Not SufficientOur measurement: near-zero direct effect on citation rates. We ship it as hygiene on every engagement — and refuse to sell it as a ranking lever. Research 2026-10-20AboutWho is Jun Liu?
Specimen 04Jun Liu is a GEO consultant in Cixi, Zhejiang, China, and the founder of AccessSmith, a GEO services practice for brands selling across borders. He spent 22 years as a product manager and 13 years in e-commerce, leading product categories that scaled to hundreds of millions in revenue.
That operator background shapes the work: audits end in fix lists with clear priorities, not slide decks. Jun publishes GEO notes in Chinese on his WeChat official account, JUN的AI出海笔记， and works with clients remotely, worldwide. JUN & GEO is the name he works under — Jun, plus GEO, the discipline he practices. His standard is the three-angle framework on this page: mentioned, cited, recommended — measured separately and reported honestly, including when the numbers are not flattering.
Read the full story →ServicesWork with me — via AccessSmith
Specimen 05Three ways to engage, in the order most clients use them. Every service is delivered by me personally through AccessSmith, with fixed scope and fixed price agreed before work begins.
GEO Audit
A complete AI visibility audit: three angles, two layers, six platforms. You receive the full answer evidence, your mention, citation, and recommendation rates, a competitor comparison, and a prioritized fix list you can act on immediately.
AI Visibility Tracking
Your buyer questions are stored in a tracking workspace and re-run on a schedule. You watch mention, citation, and recommendation rates per platform over time, instead of guessing whether last month's work changed anything.
Fix & Monitor
I execute the audit's fix list with you — answer-first pages, corrected facts, earned third-party signals — then monitor monthly and report what moved, what did not, and what we do next. No open-ended retainers without numbers.
FAQQuestions, answered directly
Q&A LogWhat is GEO?
GEO (Generative Engine Optimization) is the practice of making a brand visible inside AI-generated answers. When buyers ask ChatGPT, Perplexity, or Gemini for recommendations, the answer names some brands and ignores others. GEO audits measure whether your brand is mentioned, whether your pages are cited as sources, and whether the answer recommends you — then fix the content, structure, and third-party signals that determine those outcomes.
How is a GEO audit different from an SEO audit?
An SEO audit asks whether search engines can crawl, index, and rank your pages. A GEO audit asks a different question: when an AI platform answers a buyer's question, does it mention your brand, cite your pages, and recommend you? The two overlap on technical foundations, but GEO adds answer-level measurement across AI platforms and the third-party sources those platforms trust.
Which AI platforms do you measure?
Six platforms: ChatGPT, Perplexity, Copilot, Gemini, Grok, and Claude. These are the platforms buyers actually ask, so they are the platforms an audit has to measure. Results are reported per platform, never blended into a single average, because a brand can be strong on one platform and invisible on another.
What do I receive from a GEO audit?
You receive a report with three scores per platform — mention rate, citation rate, and recommendation rate — measured separately for your brand and for each product or SKU. The report includes the exact answers the AI platforms gave, the sources they cited instead of you, and a prioritized fix list: which pages to rewrite, which facts to correct, and which third-party signals to build.
How do we start working together?
Email [email protected] with your brand name, your website, and the products you sell. You will get a reply with the audit scope, the buyer questions we would test, and a fixed price before any work begins. All work is done remotely from Cixi, Zhejiang, China.
ContactStart with one email
Specimen 07Tell me your brand, your website, and what you sell. I will reply with the audit scope, the buyer questions I would test, and a fixed price. No call required — email is how I work with clients in every timezone.