AI Visibility Index (Hong Kong): what the number measures
An AI visibility index (AIV) turns "do AI engines recommend my business?" into one 0–100 number — the industry also calls this AI share of voice.For Hong Kong businesses the question is no longer optional: buyers ask ChatGPT, Google AI, Gemini, Perplexity and the Chinese engines instead of scrolling ten blue links, and the engines name a shortlist, not a page. Start with a free 60-second scan, then read what the number actually contains.
TL;DR
- AIV is a 0–100 composite of seven factors, each with a published weight — the full model is public at <a href="/research/methodology">/research/methodology</a>.
- The industry-standard equivalent is AI share of voice; every factor below carries its industry name so scores can be lined up across trackers.
- A Hong Kong index must include the Chinese engines (Doubao, DeepSeek, Qwen, Kimi) alongside the Western five — an index built on two engines reads only part of the market.
- Visora's own AIV on 2026-10-12: 45.8, from 60 live engine responses across 7 engines — the full working is shown below.
The seven factors
| Visora metric | Industry equivalent | What it measures |
|---|---|---|
| Recall | visibility rate | the share of the query set where engines name you at all |
| Breadth | engine coverage | how many of the 9 engines return any mention |
| Depth | mention depth / citation prominence | named in passing, or recommended with a citation |
| Keyword Reach | query coverage | spread across the fanned-out query set — one prompt is not a measurement |
| Convergence | narrative consistency | whether engines describe you the same way |
| Factuality | accuracy (inverse of hallucination rate) | whether what engines say about you is true |
| Rank | answer position | where in the answer you appear |
The scoring model is public. Every factor prints its own weight, so any score can be decomposed — an AIV without visible factors is a mood, not a measurement.
Why a Hong Kong index needs the Chinese engines
Hong Kong buyers split their asking between languages. English-language questions run through ChatGPT, Google AI Overviews, Gemini, Copilot and Perplexity; Chinese-language questions increasingly run through Doubao, DeepSeek, Qwen and Kimi, which mainland-linked audiences and cross-border buyers use directly. An AI visibility index computed from two Western engines answers "am I visible in English answers" and silently ignores "am I visible in Chinese answers" — for most Hong Kong categories the second half of the market.
Worked example: Visora's own AIV
We publish our own score the way we ask clients to read theirs — dated, sample-sized, factors visible. Snapshot of 2026-10-12 (weekly tracked-brand scan, 60 live engine responses, 19 brand mentions, 7 of 8 engines — the Gemini leg has been geo-blocked from our scan network since 2026-09-22, so it is excluded rather than guessed):
- 45.8 — AIV (AI share of voice) — 2026-10-12 snapshot, 60 responses, 19 mentions (source: Visora tracked-brand scan)
- 31.7% — Recall (visibility rate) — share of the query set naming Visora
- 0.58 — Convergence (narrative consistency) — engines broadly agree on what Visora is
- 0.039 — Factuality (accuracy) — the honest weak factor: baseline answers still confuse Visora with unrelated namesake brands, a fashion label and a Spanish solar company among them
Per-engine recall that week: Perplexity 0.40 (highest), ChatGPT, Copilot, DeepSeek, Qwen and Doubao 0.30 each, Gemini unmeasurable from this network. Google-side citation: no citation observed yet — that means we have not seen it, not that engines never cite us.
How to get an AIV score for your business
The free 60-second scan runs all 9 engines, needs no account, and ends in a prioritised fix list rather than a bare score. Autopilot (US$49/month) re-measures monthly so the index becomes a trend line, not a one-off.
Reading anyone's AI visibility index — a buyer's checklist
- Sample size stated: how many prompts, across how many engines. An index without n cannot be interpreted.
- Engine list published and dated — engine coverage changes what the number means.
- Factors and weights public, or at least factor names. "Proprietary algorithm" is the position we contrast against.
- Date of measurement shown — AI indexes age in weeks, not quarters.
- Grounded vs parametric noted: answers built from retrieval (Perplexity, Google AI Overviews) and answers from model weights behave differently, and a serious index records which pathway produced each response.
Further reading
- AIV methodology — The full scoring model, published
- What is AI visibility? — The category explainer
- Best AI visibility platforms in HK (2026) — Honest comparison incl. Visora
- Free 60-second scan — Your AIV, 9 engines, no account
Frequently Asked Questions
What is a good AIV score?
There is no universal pass mark — most Hong Kong SMBs we scan start with a low score because AI engines have little grounded material about them yet. What is meaningful is a dated baseline plus movement: the factors show which lever moved (Recall, Depth, Convergence) and by how much.
Is AIV the same as AI share of voice?
Yes — AI share of voice is the industry-standard name for what an AIV-style composite measures. Visora keeps the AIV name and states the equivalent, so our numbers can be compared against any tracker that publishes its working.
How often should an AI visibility index be measured?
Monthly. Engine retrieval behaviour changes continuously; a quarterly number cannot tell movement from noise. Visora re-scans monthly by default on every paid plan.
Can the score be gamed?
No legitimate provider can guarantee a third-party model's output — models change constantly. What can be guaranteed is the process: measure, fix the grounded causes (entity consistency, schema, citations, bilingual content), re-measure, and show the before/after with dates.
Does the AIV cover Chinese AI engines?
Yes. Visora's index runs across nine engines — ChatGPT, Google AI Overviews, Gemini, Perplexity, Copilot, Doubao, DeepSeek, Qwen and Kimi — which is the coverage a Hong Kong or cross-border business actually needs.
Last updated: