Last updated: August 2026  •  ~9 min read

Written by the SMK Vietnam editorial team — a decade in SEO, content and AI-assisted marketing, running bilingual and trilingual campaigns for Vietnamese brands and for US, Japanese, Korean and European companies entering Vietnam.

Every foreign brand that lands in Vietnam asks the same question in month one: can we just run our English assets through AI and ship them? And every Vietnamese brand exporting to Japan, Korea or the US asks the mirror version of it. The honest answer, after running these pipelines for years, is yes for most of your words and absolutely not for the words that matter.

The mistake is treating this as a binary. It isn’t a choice between a €0.02/word machine and a $0.25/word copywriter. It’s a routing decision you make asset by asset — and getting the routing right is where the money is.

Directional estimates from SMK Vietnam client projects, 2025–2026. Treat as planning ranges, not benchmarks — your split depends on category, regulation and how much of your copy is emotive.

What is the difference between AI translation and transcreation?

Translation preserves the sentence. Transcreation preserves the reaction.

If your English headline is “Built for people who never sit still,” a translation engine gives you a grammatically perfect Vietnamese sentence about people who do not sit still. A transcreator throws the sentence away and asks: what does restlessness signal to a 28-year-old in Ho Chi Minh City? Ambition? Freedom? Being bận rộn in a way that reads as successful, not exhausting? Then writes a new line in Vietnamese that produces that feeling — and it may share zero words with the original.

There’s a third mode people forget: localisation — the plumbing. Currency in VND with the right separators, dates as DD/MM/YYYY, phone formats, Zalo instead of WhatsApp, Shopee/TikTok Shop instead of Amazon, addresses, sizing charts, and the legal footers Vietnam requires. AI handles almost all of this reliably if you tell it to; the failure mode is forgetting to ask.

Where is AI translation genuinely good enough in 2026?

Frontier models (Claude, GPT, Gemini) plus dedicated engines (DeepL, Google Translate) have crossed a real threshold on Vietnamese. Five years ago Vietnamese output read like a foreigner reading a phrasebook. Today, with a decent system prompt and a glossary, it reads like a competent junior copywriter — occasionally stiff, rarely wrong.

These asset types are safe to route machine-first, with a human doing a factual-accuracy pass rather than a rewrite:

• Product specifications and attribute tables — dimensions, materials, compatibility. Volume is high, creativity is zero.
• Help centre, FAQ and support macros — see our customer-service automation approach for how these feed chatbots.
• E-commerce listing bodies on Shopee, Lazada and TikTok Shop, where search-keyword coverage matters more than poetry.
• Long-form SEO and blog content — draft in the source language, translate, then have a native editor tighten the first 200 words and the H2s.
• Internal briefs, decks and reporting for cross-border teams. Nobody was ever fired for a slightly stiff Japanese status update.
• Email nurture bodies (not subject lines — subject lines are transcreation).

The unlock is not the model, it’s the context you hand it: a glossary of do-not-translate terms, a brand voice sheet, three examples of approved copy in the target language, and an explicit register instruction. Without those you get generic output and blame the AI. See our Claude playbook and ChatGPT playbook for the prompt scaffolding we use.

Where does AI translation fail — and what does it cost you?

Failures cluster in predictable places. Here’s the split we use to route work:

The specific ways it breaks on Vietnamese, in order of how often we see them:

1. Pronoun and register collapse. Vietnamese forces you to choose a relationship — bạn, anh/chị, quý khách, cô/chú. Models default to a flat, slightly formal bạn/quý khách mix that reads like a bank letter in a fashion campaign. This is the single most common giveaway that copy was machined.
2. Diacritic and tone drift in generated variants, especially when copy is regenerated or truncated for ad character limits.
3. Over-literal idiom. English marketing runs on metaphor. Vietnamese renders it word-for-word and it lands as nonsense or, worse, as unintentionally funny.
4. Length explosion. Vietnamese typically runs 15–30% longer than English. Your beautiful 40-character button label no longer fits, and your ad gets truncated mid-word.
5. Regulatory blindness. Models happily translate a health or financial claim that is legal in the US and not permitted in Vietnam. Check against local compliance rules before shipping.

What changes between Vietnamese, English, Japanese and Korean?

Each pair has its own trap. Treating “Asian languages” as one bucket is how foreign brands lose a quarter.

One rule that has saved more campaigns than any tool choice: never chain languages. If your source is English, translate English→Japanese and English→Vietnamese in parallel. Do not translate English→Vietnamese→Japanese because the Vietnamese file happened to be ready first.

What does the hybrid workflow actually look like?

Five steps. A four-language campaign kit — landing page, three ad sets, one email, ten listings — moves through this in about two to three working days.

What should you budget for a multilingual Vietnam campaign?

Rough planning numbers, not quotes. The point is the shape: your spend concentrates on a small number of high-visibility lines.

Half your localisation budget buys 5% of your words. That feels wrong until you remember those words are the ones on the hero image, the ad, and the button — the only text most of your audience will ever read. For how this fits a wider media plan, see our complete AI marketing stack for Vietnam.

Which mistakes cost the most?

Frequently asked questions