AI Translation: Good Enough for What, Exactly?
Machine translation crossed a threshold. Knowing which content it crossed it for is what separates a saving from an embarrassment.
Machine translation quality improved enough in recent years that the old advice — never publish it unreviewed — needs qualifying rather than repeating. The honest answer is that it depends entirely on what the text is doing.
Where it is production-ready
- Internal documentation and support material, where clarity matters and register does not.
- Product descriptions with factual, structured content — dimensions, materials, compatibility.
- Customer support replies, reviewed by an agent who reads the target language even imperfectly.
- First drafts of anything, which is now a genuine time saving rather than a starting point that needs rewriting.
Where it still fails commercially
- Brand voice. Humour, rhythm and the deliberate rule-breaking that makes copy memorable do not survive translation, and the output reads competent and forgettable.
- Legal and contractual text, where a subtle shift in obligation is invisible in fluent output and expensive in a dispute.
- Marketing headlines. The literal translation of a good headline is almost never a good headline in the target language.
- Culturally specific references, idioms and anything where the equivalent phrasing is different rather than translated.
The workflow that works
Machine translate, then have a native speaker edit rather than retranslate. This is typically three to four times faster than translating from scratch and produces materially better results than either alone. Give the reviewer a glossary of terms that must not vary — product names, legal phrases, your own vocabulary — because consistency across a site is where machine output drifts most visibly.
SEO implications
Translated keywords are frequently not the terms the target market actually searches. Have someone do native keyword research per language rather than translating your existing list, and expect page structures to differ. A market with different objections needs different sections — that is localisation, and it is where the commercial return actually is. If you want this built rather than described, it is what our AI integration service covers.
- 3-4x faster than translating from scratch
- 1 glossary of fixed terms
- native keyword research per language
Frequently asked questions
Can we publish machine translation without review?
For low-stakes informational content, increasingly yes. For anything commercial, legal or brand-carrying, the review is cheap insurance — and it is editing now rather than rewriting, which is a much smaller cost than it used to be.
Does Google penalise machine-translated content?
Guidelines target unhelpful content rather than translation method. Well-reviewed machine translation that genuinely serves the reader is fine; bulk unreviewed translation across hundreds of pages is exactly the pattern that gets treated as spam.
More on this topic: Artificial Intelligence.
Keep reading
Want this built for your business? See what we do.