AI can translate your words. It can’t make them land.

# linguistics

# translation

# blog

# ai-research

71% of global users distrust flat translations. Here's what AI alone can't fix.

June 8, 2026

There’s a belief spreading through enterprise leadership right now: AI has solved translation. The evidence seems compelling — costs are down, turnaround is faster, and the output passes a surface read. But our colleagues at Centific’s multilingual AI team have published a sharp rebuttal to that assumption, and it’s worth every OneForma contributor and content specialist taking note.

THE MYTH VS. THE REALITY

As the Centific team puts it, localization was never primarily a translation problem. It was always a cultural problem that translation technology helped address. AI has made the translation part faster and cheaper. It has not touched the cultural part — and those two challenges require fundamentally different solutions.

A model can convert words accurately from English to Brazilian Portuguese. What it cannot do is tell you whether the urgency of a call to action lands the same way, whether a villain still sounds threatening in German, or whether a joke that works in one market simply reads as bizarre in another. These are the failures that determine whether a global audience feels genuinely addressed — or processed.

“Content that has cleared the lowest bar — linguistic accuracy — has not been tested for cultural accuracy.”

WHAT THE NUMBERS SAY

The stakes are not abstract. Research cited by Centific finds that 71% of global users distrust culturally flat translations, and 57% will abandon a digital platform that lacks contextually accurate localized support. On the upside, brands that go beyond translation and adapt the full user experience — regional trust signals, market-specific messaging, localized user journeys — see conversion rates climb by as much as 70% over standard translated content. For multilingual ad campaigns, culturally adapted creative outperforms generic machine-translated assets by up to 86% in click-through and conversion.

WHAT THIS MEANS FOR OUR WORK AT ONEFORMA

As contributors who work at the intersection of language and AI, this matters directly to what we do. The human-in-the-loop step — the review, the cultural judgment, the feedback that gets fed back into the model — is not a workaround for AI’s limitations. It is the mechanism by which AI actually improves. Without structured human correction data, AI doesn’t learn from its cultural blind spots. It repeats them, across more content, in more markets, faster than any reviewer can catch.

The message is clear: AI handles volume and surface accuracy. People with cultural knowledge handle authenticity. Both are necessary, and neither substitutes for the other.

Read the full article

This article draws on insights from our colleagues at Centific. For the full analysis — including data on remediation costs, language-geography mismatches, and the ROI of full localization — read the original piece: Don’t buy the “translation is solved with AI” myth on the Centific Multilingual AI blog.

Comments (0)

Popular