Guides on AI search visibility.

Three pages from the actual report format — executive summary, live AI answers, full five-pillar scorecard — shown on a fictional composite clinic so you can see the depth before you commission one.

346 grounded answers across six assistants and two languages. English and Chinese answers share only 3 of their top 20 sources; aesthetics and legal share none. Full dataset as CSV and JSON.

We asked ChatGPT, Claude, Gemini, and Perplexity real questions about Hong Kong clinics and law firms, in English and Chinese, and tallied who they actually name.

English and Chinese AI answers share only 3 of their top 20 sources. What that means for Hong Kong clinics and law firms, and why the language of the question changes the practice AI recommends.

ChatGPT and Claude are geofenced out of Hong Kong; Gemini arrived in March 2026. Which assistants local customers can actually open, and why a single blended visibility score is the wrong measurement.

Every AI crawler user-agent that matters, the three different jobs they do, and which market each serves — including the Chinese-model crawlers most references omit, and why DeepSeek cannot be governed by robots.txt at all.

The vocabulary of AI search visibility, defined and connected — GEO, LLMO, AEO, grounding, retrieval, share of voice, hallucination. Every term links to what it relates to and what it is confused with.

A working definition of GEO — the discipline of being cited, named, and recommended inside AI-generated answers — and why it matters for professional practices specifically.

Three overlapping terms, three different target surfaces. A precise comparison, with a table, and why most teams should treat them as one programme.

Honest comparison of the alternatives to a dedicated AI-visibility engagement — website-builder plugins, general SEO agencies, enterprise GEO platforms, freelancers, DIY — with the limitations of each.

Website builders and plugins that claim to optimise for AI: what they actually do, where they fall short, and when the audit-plus-remediation model fits better.

The four-step pipeline behind every grounded answer — retrieve, rank, synthesise, cite — and where a business is won or lost inside it.

What published research on retrieval-augmented generation and citation behaviour actually shows — and what it implies for how a practice should structure its information.

A practical method for measuring whether AI assistants mention and recommend your practice — including the questions to test and how to read the results.

A plain-text file that tells AI assistants what to cite about you. It works for the assistants that read it on request; it does not reach models built on broader training data.

AI answers send no referrer header, so GA cannot trace them. How to measure visibility against revenue when the attribution gap is structural.

DeepSeek and Qwen carry much of the Chinese-language usage in Hong Kong but publish no crawler user-agent robots.txt governs. What actually moves visibility for them.

AI answers usually name three practices. If yours is not among them you are invisible to the whole decision — visibility wins or loses at the shortlist moment.

The bigger the decision the more research precedes it, and the more of that research happens against an AI assistant. Why high-ticket high-trust services are the hardest hit.

Pick the one question your prospects actually ask, optimise for it, and measure share of recommendations before and after. A campaign shaped around a single prompt cluster.

A wrong price or a closed location stated as fact is not hallucination — it is conflicting sources and the model ranking one above the other. The fix is making your correct information the highest-ranked source.

DeepSeek and Qwen answer "is simplyai.work legitimate" with a different Hong Kong company's registry facts. Who actually operates simplyai.work, how entity resolution gets this wrong, and the checklist for any business with a colliding name.

The questions worth asking before engaging any GEO or AI-visibility provider — and the claims that should give you pause.

What to do in the first month: run the questions your prospects ask against the assistants your market can reach, log the errors, fix the fixable, and build the record the next campaign measures against.