There is a growing assumption in organisations that AI has solved translation.
The logic feels sound. Generate a draft instantly, run a quick human check, and move on. Faster, cheaper, more efficient.
But in practice, many teams are discovering something different. The work has not disappeared. It has shifted.
And in many cases, it has become slower.
The illusion of speed
AI translation tools can produce output in seconds. That part is not in question.
The issue begins immediately afterwards.
Unlike human translation, which arrives with embedded judgement and an understanding of risk context, AI output arrives without accountability. It presents language that looks complete, but AI has no awareness of compliance requirements, brand implications, or downstream consequences.
This means every line must be verified.
Not skimmed. Verified.
- Is the meaning accurate in this specific context?
- Does the terminology align with industry standards and compliance frameworks?
- Does the tone reflect the organisation’s established voice and style?
- Does any phrasing introduce ambiguity, reputational exposure, or regulatory risk?
- Has the AI arbitrarily introduced or removed content?
What looked like a time-saving step becomes a line-by-line validation exercise rooted in risk prevention.
Research from the European Commission has highlighted that while machine translation can improve productivity in low-risk, high-volume contexts, human revision remains essential for accuracy, usability, and compliance—particularly in specialised or regulated domains.
In other words, speed at the front end creates more effort at the back.
Translation is not substitution
Part of the challenge lies in how translation is still perceived.
Many non-specialists assume it is a process of substitution. Replace one word with another. Swap languages, preserve meaning.
Specialist translators are all too familiar with that impatient exhortation: “Just translate it!”
But meaning does not move in neat, word-level units.
It is shaped by context, culture, regulatory environment, authorial intention, audience expectation.
A translated phrase that is technically correct can still be misleading. A translated sentence that reads fluently can still distort intent or fall short of compliance standards.
This is particularly visible in areas such as financial reporting, sustainability disclosures, legal documentation and high-profile advertising or marketing, where wording is not just descriptive but performative. It carries obligations. It becomes part of the organisation’s compliance infrastructure, with direct implications for brand protection, brand viability, and stakeholder interpretation.
Guidance from ISO 17100 emphasises that translation is a structured, multi-step process requiring qualified human review to ensure accuracy, appropriateness, and domain alignment.
AI does not remove that requirement. It increases the need for it.
The hidden cost of revalidation
When organisations rely on AI-generated translation, they often underestimate the mental effort placed on the reviewer.
Reviewing a human translation is typically an act of refinement. In most cases, the structure, intent and terminology are already in place.
Reviewing AI output is different. It requires constant vigilance.
Every sentence raises questions:
- Is this correct, or does it only appear to be correct?
- Has nuance been flattened or altered in a way that introduces risk?
- Are there inconsistencies that could affect clarity, compliance, or credibility?
- Has AI introduced content that should not be there?
- Has AI removed elements that are necessary for accuracy, compliance or meaning?
- Has AI re-interpreted content according to its own “understanding” of the original prompt?
This kind of checking takes more time than translating a text from scratch.
It also introduces exposure. Studies in human factors, including work by Mary L. Cummings on automation bias, show that people tend to over-rely on automated outputs, particularly when they appear fluent and well-formed. This reduces the likelihood that errors are rigorously challenged.
In translation, that fluency can mask subtle but critical issues—precisely the kind that undermine risk prevention efforts.
Voice and style do not survive by accident
Beyond accuracy, there is another layer that AI struggles to preserve: voice and style.
Organisations invest heavily in how they come across to their audiences. Their tone signals credibility, authority, and intent. It is a core component of brand protection.
AI does not understand that voice. It approximates it.
Without careful intervention, AI-translated content can drift. It may become more generic, more formal, or simply inconsistent with established brand guidelines.
Over time, this creates fragmentation.
The organisation begins to sound different across markets. Messages that should reinforce each other start to diverge.
This is not just a stylistic issue. It is a brand risk.
Consistency of voice is not cosmetic. It is part of how organisations maintain trust, coherence, and credibility across borders.
Compliance is where speed breaks down
As already noted, the risks become more pronounced in regulated environments. In sectors such as finance, healthcare, and energy, wording is often tied directly to legal and regulatory frameworks. Small shifts in phrasing can have material implications.
The European Banking Authority, for example, emphasises that disclosures must be clear, accurate, and not misleading, recognising that ambiguity can distort interpretation and decision-making.
AI-generated translations, even when fluent, can introduce unintended ambiguity, inconsistent terminology, or deviations from accepted compliance language.
This is where the idea of “quick post-editing” begins to collapse.
Because once compliance is in play, there is no shortcut. Every sentence must be checked not only for linguistic correctness, but for regulatory alignment, legal defensibility, and reputational impact.
Where AI does help, and where it doesn’t
None of this is to suggest that AI has no role.
It can be valuable for:
- Rapid internal understanding of foreign-language content
- Early-stage drafts in low-risk contexts
- Supporting terminology research
- Acting as a kind of super-thesaurus
But when the output is client-facing, public, or subject to compliance scrutiny, the equation changes.
The real question is not speed
Organisations often approach translation decisions with a simple question: how fast can we get this done?
A more useful question is: how much interpretation risk are we willing to carry?
Because that is what sits beneath translation. Not just words, but meaning. Not just meaning, but consequence.
AI can generate language quickly. But ensuring that language is accurate, compliant, aligned with voice and style, and safe from a branding and regulatory perspective still takes time. And that time does not compress easily. If anything, it expands.
Translation has never been a word-for-word exercise.
It is a decision-making process about how meaning communicates itself—safely, clearly, and consistently.
AI has just changed the tools.