AI has already replaced a large share of routine translation – bulk content, internal documents, and first drafts where “good enough” is genuinely good enough. What it has not replaced is work where being wrong is expensive or the wording carries weight: legal, medical, and regulatory translation, marketing that must persuade in a new culture, literary work, and live interpreting where tone and context matter. So the honest answer to will AI replace translators is that the job is shifting rather than vanishing: less typing from scratch, more reviewing machine output, localising, and certifying accuracy. Rates for commodity work are falling. Specialists and certified translators are holding up better. Here’s which work is going, which is safe, and how to move.
Start with the task, not the job title
Most articles ask whether translators will be replaced. That’s the wrong unit of measurement. A translator’s week is a stack of very different tasks, and AI has eaten through them at wildly different speeds. Someone translating internal HR memos and someone certifying birth certificates for immigration cases are both “translators,” and their risk profiles have almost nothing in common.
| Translation work type | What AI handles now | What still needs a human | Risk level |
|---|---|---|---|
| Internal docs, emails, support tickets | Full first pass, frequently shipped raw | Occasional spot-checks on sensitive material | High |
| Technical manuals, help centres | Near-final drafts in major language pairs | Terminology consistency, safety warnings | High |
| E-commerce listings, user reviews | Bulk output at scale, no human in the loop | Brand-critical category pages | High |
| Marketing and ad copy | Literal, grammatical renderings | Persuasion, humour, cultural rewrite | Medium-low |
| Contracts, patents, court filings | Drafts and terminology suggestions | Certification, liability, legal sign-off | Low |
| Medical, pharma, regulatory | Drafts, glossary work | Clinical accuracy, regulatory approval | Low |
| Literary, games, creative subtitling | Rough drafts | Voice, rhythm, register, jokes that land | Medium–low |
| Live interpreting (court, medical, diplomatic) | Prep, glossaries, captions, transcripts | The actual job | Low |
Find your own row before you panic. Or before you relax.
What machine translation genuinely does well now
Quietly, in a lot of organisations, translation stopped being a purchased service and became a button.
The clearest public example came in January 2026, when IMF Managing Director Kristalina Georgieva told an audience at Davos that the Fund had gone from 200 translators and interpreters down to just 50. Not a budget cut. A technology decision. And the US Bureau of Labor Statistics said the same thing in its own dry language in its 2024–34 projections write-up: interpreters and translators have become more productive as AI translation improves, which reduces employment demand for them. World Economic Forum Bureau of Labor Statistics
Where does output actually ship without a human touching it? Repetitive, high-volume, low-consequence text. Internal documentation. Support tickets. Product descriptions in a catalogue of 40,000 SKUs. Subtitles on a webinar nobody will litigate over. The economics are brutal and simple: if the cost of a small error is close to zero, the market will not pay a human to prevent it.
This is the same pattern playing out in other creative fields – the same logic behind the debate over whether AI will replace graphic designers. The bottom of the market goes first.
Where it still breaks
Now the other half, which rarely makes headlines because nothing exploded.
- Legal liability. In 2018 a US federal judge in Kansas threw out evidence in United States v. Cruz-Zamora because a trooper had used Google Translate to ask for consent to search a car. The court found the app produced “literal but nonsensical” output and ruled it unreasonable to rely on it for consent. Courts and regulators want a named human who signs. Martin Law Firm
- Medical consequences. A 2019 JAMA Internal Medicine study by Khoong and colleagues checked machine-translated emergency discharge instructions and found that 2% of Spanish and 8% of Chinese sentences carried potential for significant or life-threatening harm. Highly usable. Not unsupervised. ResearchGate
- Persuasion. Grammatically perfect copy that doesn’t sell is a failure. Transcreation – rewriting a campaign so it works on a different audience – is a marketing job wearing a linguist’s coat.
- Voice. Literary and creative work still reads flat. A February 2026 study by the data company Appen tested seven major AI models across 20 languages and found they handled general cultural content relatively well but struggled badly with idioms and puns, often leaving idioms untranslated. Slator
- Invention. Researchers at Alibaba, in an October 2025 study of 17 large language models across 11 English-to-X pairs, reported that translation hallucination remains a significant problem. A model that fabricates a dosage or a clause is worse than no translation at all. Slator
- Live, high-stakes interpreting. Which brings us to the question underneath the question.
So will AI replace interpreters? Here’s a useful bit of history. In June 2018, the then Lord Chief Justice of England and Wales predicted publicly that high-quality simultaneous translation would arrive within a few years and courtroom interpreters would be finished. Eight years on, courts still book humans. AI has arrived in interpreting, but sideways: the 2025 Slator Linguist Survey found roughly 55% of professional interpreters using AI tools somewhere in their work – mostly terminology lookup, key-term extraction and automated transcripts, not automated courtroom or medical interpreting. Andy Benzo, president of the American Translators Association, told CNN in January 2026 that the risks of AI in high-stakes settings are “humongous”. InterpretertrainCNN
One caveat that ruins most quality claims
Almost every confident statement about AI quality is really a statement about English–Spanish.
A 2021 study in the Journal of General Internal Medicine by Taira and colleagues tested the same emergency-department instructions across seven languages. Spanish came out at 94% accuracy, Tagalog 90%, Korean 82.5%, Chinese 81.7%, Farsi 67.5% – and Armenian 55%. The authors concluded machine translation was too inconsistent between languages to be relied on for patient instructions. nih
That spread is the whole story for anyone working outside the top handful of language pairs. Benchmarks are improving – Microsoft’s LINGUA programme funded 11 projects building datasets for low-resource European languages in 2026, and France’s INRIA has launched similar work – but the gap is real today. Don’t let a headline about AI vs human translation written by someone working in English–German tell you what your Amharic or Khmer market looks like. Slator
What’s actually happening to the market
Two things at once, which is why the conversation feels so confused.
Demand for language services hasn’t collapsed. Slator’s 2026 market report values the global language solutions and AI market at USD 30.85 billion for 2025 and projects USD 36.10 billion by 2031 – a compound annual growth rate of 2.65%. Growth, but slow growth, and much of the spend is moving from human hours to software. Slator
Individual earnings are a harder story. The UK Society of Authors surveyed its members in January 2024 and found that 36% of translators said they had already lost work to generative AI, 43% said their income had fallen because of it, and 77% expected it to hurt their future earnings. There’s also econometric evidence: Carl Frey and Pedro Llanos-Paredes at Oxford analysed US data from 2010–23 and found that regions with heavier Google Translate use saw slower growth in translator jobs. The Next Web CNN
Official employment numbers sit somewhere in between. The BLS projects 2% growth for interpreters and translators from 2024 to 2034 – slower than average – with about 6,900 openings a year, most of them replacing people who leave. The median wage was $59,440 in May 2024. bls
And here’s the part that gets skipped. In Slator’s 2025 survey of language service integrators, 84% said clients had specifically asked them for human editing of AI translation output in the previous year. Commodity translation is compressing. Review is growing. If you want the wider view of which occupations are being reshaped this way, we covered it in what jobs AI will replace by 2030. Slator
From translating to reviewing: what MTPE really involves
Machine translation post-editing means the machine drafts and you fix it. In practice that ranges from a light pass – catch the howlers, ship it – to full post-editing, where you check terminology, restructure clumsy sentences, and take responsibility for the result.
Does it pay? Depends entirely on which end you’re at. Light post-editing priced per word, at volume, for an agency you’ve never spoken to, is a race you cannot win. Full post-editing with quality assurance in a regulated domain, priced by the hour and backed by your credentials, is a different business.
The Slator linguist survey of 260 professionals found something telling: beyond post-editing itself, linguists were being hired for AI prompting, terminology management, data management, and labelling data to train large language models. The job description is expanding, not just shrinking. Slator
The IMF makes the point neatly. Months after cutting its language staff, it advertised in August 2026 for a Chief Translator – a role leading the adoption of AI tools, terminology and project management, and editorial and quality assurance work. Fewer translators. A more senior translator. That’s the shape of the future of translation jobs in one hiring decision. Slator
Who is most exposed
Run yourself through this. Each “yes” is a point.
- Most of your income comes from one or two very common language pairs (EN↔ES, EN↔FR, EN↔DE, EN↔PT).
- You translate general content with no subject specialism – web copy, blogs, generic business documents.
- You have no certification, accreditation, or sworn status.
- Your clients are agencies rather than end clients, and you rarely speak to the person who needs the translation.
- You compete mainly on price and turnaround.
- You do not currently use CAT tools, terminology management, or AI-assisted workflows.
- You have never done live interpreting.
Four or more, and the pressure you’re feeling isn’t imagination. It’s structural, and it will keep building. Our guide on how to not get replaced by AI at work is a decent companion read.
Who is safest
Nobody is guaranteed. But some positions have moats.
Certified and sworn translators, because their signature carries legal weight a model cannot. Note that requirements differ enormously by country – some have state-appointed sworn translators, others rely on professional associations. Check your own national body; in the US that’s the American Translators Association, in the UK the Chartered Institute of Linguists.
Then: regulated domains where an error triggers liability – pharma, medical devices, patents, financial filings. Interpreters in court, healthcare, and diplomacy. Localisation and transcreation specialists who are half marketer. And translators in rare or low-resource pairs, where the training data simply isn’t there.
The common thread is the same one running through most AI-proof careers: accountability, context, and consequences.
How to reposition: a 90-day plan
Days 1–30: pick a lane
Choose one domain you can defend – not “legal and medical and marketing.” One. Preferably one you already touch. Audit your last year of invoices: which work paid best per hour, not per word? Start rebuilding around that.
Days 31–60: get fluent in the workflow
Learn to run the machine rather than race it. That means real post-editing discipline, terminology and glossary management, prompt design for translation and QA, and honest evaluation of output quality. If you can explain to a client why a model failed on their text, you’re no longer interchangeable. This is exactly the kind of thing worth putting on paper – see AI skills to add to a resume.
Days 61–90: fix your client mix
Chase credentials in your chosen domain, and start replacing agency volume with direct clients. Direct clients buy outcomes and accountability. Agencies buy words. One good direct relationship can be worth ten agency accounts. Longer-term thinking on this lives in how to stay relevant in the age of AI.
Is it still worth becoming a translator?
Straight answer: yes, but only on conditions – and the conditions are non-negotiable now in a way they weren’t ten years ago.
Studying languages purely to convert text for money is a weak plan. Studying languages plus law, plus medicine, plus games, plus finance – and learning to direct AI systems rather than compete with them – is still a strong one. When Slator asked linguists in 2025 whether they’d recommend the career, plenty said yes with caveats, and the caveats clustered around two things: specialise deeply, and learn to use AI properly. Slator
If you’re mainly interested in languages for yourself rather than professionally, that’s a different road – start with how to use AI to learn a new language instead.
The translators doing well right now aren’t the fastest typists. They’re the ones who decide what ships, fix what’s broken, and put their name to it. That’s a different skill set, and it’s learnable.
If you want to make that move deliberately – from producing translations to directing and certifying them – the AI Certificate Program is built for exactly that shift: working with AI systems, evaluating their output, and building workflows clients will pay for. It comes with a certificate of completion (no employment or income guarantees, obviously). If your work leans toward marketing and transcreation, ChatGPT for Content Creation is the more direct fit.