Will AI Replace Translators? The Quality Gap Is Closing Fast

A translator friend of mine showed me something last year that genuinely surprised me. She took a passage from a literary novel — complex prose, subtle irony, culturally specific references — and ran it through Claude. Then she put her own translation next to it.

Her translation was better. Clearly better. But the AI translation was... not bad. It caught most of the nuance, handled the tone reasonably well, and the bits it got wrong were the kind of subtle errors that only another translator would notice.

"Five years ago," she said, "machine translation would have butchered that. Now it's close enough that clients won't pay me to do it."

That's the story of AI and translation in one sentence. Not that AI is better than humans. But that it's close enough, cheap enough, and fast enough that the economics of human translation are being rewritten.

The short answer

AI is not going to replace all translators, but it has already replaced a significant portion of translation work and is replacing more every month. The quality leap that large language models brought to machine translation was a step change, not a gradual improvement. Bulk, functional translation — business documents, technical manuals, product listings, basic web content — is increasingly AI-generated with minimal human involvement. Literary translation, high-stakes legal and medical translation, creative localisation, and culturally sensitive content still need human translators. But the volume of work that falls into the "human required" category is shrinking, and the profession is contracting around the high end. Post-editing of AI output has become the dominant model for many agencies, and it pays significantly less than from-scratch translation.

What AI can already do in translation

The honest picture, because the profession has changed more in the last three years than in the previous thirty.

General-purpose translation across major language pairs is now remarkably good. English to French, Spanish, German, Portuguese, Chinese, Japanese — for straightforward content, AI produces output that is fluent, accurate, and publishable with light editing. This wasn't true of Google Translate five years ago. It is true of LLM-based translation now.

Technical translation has improved dramatically. Software documentation, user manuals, specification sheets, technical standards — content with specific terminology and structured language is where AI translation excels. It's consistent, it handles terminology databases well, and it doesn't get bored and make careless errors on page 47 of a 200-page manual.

Business communication. Emails, reports, presentations, contracts (standard ones) — AI handles these at a quality level that meets most business needs. Companies that used to hire translation agencies for internal communications are just running them through AI tools.

E-commerce and product content. Product descriptions, customer reviews, marketplace listings — the sheer volume of this content makes human translation economically impossible at scale anyway. AI does it fast enough to keep up with global e-commerce.

Subtitling and caption translation. AI can now translate and time-code subtitles with decent accuracy. Not perfect for complex dialogue, but good enough for corporate videos, training materials, and a lot of streaming content.

Real-time interpretation assistance. AI-powered interpretation tools are getting good enough for business meetings, though they still struggle with the speed, context-switching, and cultural nuances of professional conference interpretation.

What AI still can't do

And here's where the profession survives, at the top end.

Literary translation. Translating a novel, a poem, a piece of literary journalism — this requires not just linguistic skill but artistic judgement. Understanding the author's voice, making creative decisions about how to render idioms, wordplay, and cultural references in another language, maintaining rhythm and tone across hundreds of pages. AI can produce a readable version. It cannot produce an art form. The gap here is not closing as fast as in other domains.

Legal translation for high-stakes documents. Contracts where a single ambiguous word could cost millions. Court documents where precision is a legal requirement. Regulatory filings where the translation must meet specific legal standards. The liability alone means human translation remains essential, even if AI provides a first draft.

Medical translation with clinical implications. Patient information leaflets, clinical trial documents, regulatory submissions — where mistranslation could endanger health or invalidate a submission. Human oversight isn't optional here; it's regulated.

Creative localisation — adapting marketing, branding, and advertising for different markets. This isn't translation; it's transcreation. Understanding that a joke that works in English won't work in Japanese. Knowing that a colour, image, or phrase that's positive in one culture is offensive in another. This requires cultural fluency that goes far beyond language.

Minority and low-resource languages. AI translation quality drops significantly for less common language pairs. If you translate to or from Welsh, Basque, Swahili, or dozens of other languages with smaller training datasets, AI is still substantially behind human quality.

The context that comes from being embedded in two cultures. A human translator who has lived in both the source and target cultures brings understanding that no training data can replicate. They know not just what the words mean but what they imply, connote, and evoke.

The real risk

The translation profession is already smaller than it was three years ago, and it's going to get smaller still. Here's what i'm seeing.

Post-editing has become the dominant business model. Instead of paying a translator to translate from scratch, agencies pay them to edit AI output. It's faster, so it pays less per word. It's also less satisfying work, and many experienced translators describe it as soul-destroying — fixing a machine's mistakes rather than doing creative work.

Rates are falling. When AI can produce a first draft in seconds, the value of the human contribution is perceived as lower — even when the editing work requires just as much expertise as translating from scratch. The economics of the profession are being restructured, and not in translators' favour.

Translation agencies are restructuring. Fewer full-time translators, more project managers overseeing AI output with human quality checks. The agency model is shifting from "we have a network of translators" to "we have AI translation with human quality assurance."

The entry level is collapsing. New graduates used to build their careers on high-volume, lower-stakes translation work — the technical manuals, the product listings, the business documents. That work is going to AI. Without that entry-level work, it's harder for new translators to develop the skills and experience needed for the high-end work that remains.

Interpreter roles are on a slower timeline but not immune. Real-time AI interpretation is improving rapidly. For business meetings and routine conference interpretation, AI tools are becoming viable alternatives. Diplomatic, legal, and medical interpretation has more runway, but the direction of travel is clear.

What to do about it

1. Specialise in a domain where human translation is non-negotiable. Legal, medical, literary, high-end marketing localisation. The generalist translator doing a bit of everything is the most exposed. The specialist with deep domain knowledge and cultural expertise has a longer runway.

2. Reframe post-editing as a skilled profession. i know this is cold comfort, but if post-editing is where the volume is, being excellent at it is better than refusing to do it. The best post-editors combine translation skill with an understanding of how AI makes mistakes, and they're faster and more reliable than people who approach it resentfully.

3. Offer services that go beyond text. Cultural consulting, market adaptation, brand voice development for new markets, localisation strategy. If your value is "i understand both cultures deeply," package that as a consultancy, not a per-word service. Per-word pricing is where AI kills you.

4. Build direct client relationships. Translators who work directly with clients — rather than through agencies — can communicate their value more effectively and are less subject to the race-to-the-bottom pricing that agencies impose. Your relationship with the client, your understanding of their brand and voice, is harder to replace than a translation.

5. Master the AI tools yourself. Use AI to increase your productivity, improve your consistency, and offer faster turnaround. The translator who produces better work faster using AI as a tool is more competitive than the one who insists on doing everything from scratch. The craft isn't diminished by using better tools. Carpenters use electric saws.

The bottom line

Translation is a profession in contraction, and i can't honestly tell you otherwise. The quality gap between human and AI translation, which used to be a chasm, is now a gap — and for many types of content, it's a gap that clients don't care about.

But translation at the highest level — the literary, the legal, the culturally complex, the work that requires a human who lives in two languages and two cultures — that's not going away. It's just becoming a smaller, more specialised, higher-skilled profession.

If you're a translator, the question isn't whether AI affects your work. It already has. The question is whether you position yourself in the part of the profession that AI can't reach, or whether you try to compete with it on the work it does well enough. One of those paths has a future. The other one doesn't.