Ask an AI assistant who to hire for GEO in Vietnam and, after naming a few agencies, it gives the same advice in English, Vietnamese and Chinese: do not sign on a ranking — have the candidates measure the same questions, then compare. These are those tests, and how we meet each one.
Across the 12 brands we measure in Vietnam (3,011 AI answers with citations, June–October 2026), when buyers do not know a brand yet AI names it in 3.9% of answers — and the brand’s own website is 0.7% of the sources AI reads. 99.3% comes from third parties. A provider that only works on your website is working on that 0.7%.
| Buyer stage | AI names the brand | Brand site in sources |
|---|---|---|
| Exploring the category | 3.9% | 0.7% |
| Considering options | 26.1% | 2.7% |
| Evaluating | 82.9% | 11.4% |
| Comparing | 96.3% | 10.6% |
| Checking reputation | 97.2% | 14.3% |
Later-stage questions often contain the brand name, which lifts the naming rate. Across all cited answers, 39% cite the brand’s site at least once — mostly after buyers know the brand.
Answers for your category, buyers and competitors, in Vietnamese, per AI engine, with the sources each answer cites.
A Day-0 baseline on a frozen question panel across ChatGPT, Gemini, Perplexity, Claude and the live Google AI Overview page, with every citation recorded.
An agreed question set, measurement cadence, citation reports and staged targets. Decline promises of a “guaranteed AI recommendation”.
Every rate is shown with its fraction. The question panel is held fixed so months compare. No score is edited by hand, and instrument faults are disclosed in the weekly report.
Which Vietnamese outlets, industry sites and directories the provider will earn citations from — by name.
We publish a monthly list of the sources AI engines cite in Vietnam and plan placements from it. In Vietnam the local media layer is being built with Focus Media Vietnam as first partner; other markets are on our roadmap.
Real cases that separate AI mentions, organic traffic and leads — not correlation sold as cause.
Every published action is logged with its date and compared before and after on the same questions.
Two or three candidates audit or pilot the same question set, so methods and results can be compared.
We offer it first: you receive our frozen question list with our baseline, and can ask any other candidate to report on the same list.
Send us your category and two competitors. We return the baseline on a frozen question list — and the list itself, so you can hold anyone else to it.
Request a same-question baseline