The translation-quality problem

AI translation has improved dramatically since 2022, but it is not yet at the level where you can deploy it for guest-facing content without human review. The remaining 8-15% quality gap is exactly where embarrassment lives: contextual subtleties, idioms, cultural references, and tone that a human native speaker spots in 5 seconds and AI misses.

What "translation quality" actually means

Three dimensions, all of which matter:

AI is at 95%+ on grammatical accuracy, 80-90% on idiomatic fluency, 70-85% on cultural appropriateness. The last dimension is where AI most often fails in ways that matter.

Where translation fails most often

The pragmatic workflow

For routine guest communications, AI-generated translation reviewed by a native speaker on the team produces shippable output at 5-10× the speed of full manual translation. For high-stakes content (formal apology to a serious complaint, contract language, marketing collateral), use AI to draft and a professional translator to finalize.

Never deploy AI-only translation for any content where a single embarrassing phrase will reach a wide audience — public marketing, press releases, large-group communications. The asymmetric risk is too high.