Will AI Replace Translators?

Translators score 38% on Oxford Martin School's 2013 automation scale. Anthropic's 2026 Economic Index shows 43% observed AI exposure — meaning the theoretical risk and the observed reality are almost exactly aligned. Our combined score is 41% — in the grey zone, and unlike many professions, the two measurements tell the same story. Check your specific role here →

I am not going to write about this profession from a detached analytical distance. My sister was a freelance translator for fifteen years. She spoke multiple languages fluently, built long-term client relationships, delivered quality work that her clients valued. Then, in 2024, the work stopped coming. Not dramatically. Gradually, and then all at once. I wrote about what happened to her here.

The 38% score does not capture what she experienced.

What happened to translation

AI translation — DeepL, GPT-4, and their successors — reached a threshold of quality in the early 2020s that changed the economics of the profession fundamentally. Not perfect quality. Good enough quality. And in a market where clients are primarily measuring cost, good enough at near-zero cost is a different proposition than excellent at professional rates.

The impact fell hardest on freelance translators working in common language pairs — English to Spanish, French to English, German to Dutch. For these languages, AI translation quality is high and the tools are widely available. Clients who previously hired translators for routine business communication, marketing materials, and standard documentation began handling these in-house with AI tools.

Anthropic’s March 2026 research found that language and writing tasks have among the highest observed AI exposure of any category — not surprising, given that language is what large language models are fundamentally built to process.

Where human translators remain essential

Literary translation remains deeply human. The translation of a novel is not the transfer of information from one language to another. It is the recreation of a voice, a cultural moment, a specific aesthetic experience, in a different tongue. No current AI system can do this with the nuance that literary translation requires.

Specialist technical translation — in legal, medical, or highly specialised technical domains — also retains significant human value, because errors carry serious consequences and the contextual judgment required to avoid them remains beyond current AI capability.

Interpretation — live, spoken translation in diplomatic, legal, or healthcare contexts — is more resilient than written translation, because it involves real-time human interaction, cultural mediation, and the ability to navigate the unexpected.

The honest picture for translators in 2026

The 38% score is probably too low for freelance translators in common language pairs working on routine business content. That market has changed substantially and is not coming back.

The parts of translation that require cultural depth, literary sensitivity, specialist domain knowledge, or live human presence are more resilient — but they are a smaller market, and competition for that work has intensified as translators displaced from the commodity market move toward it.

My sister did not see it coming. She had done everything right — specialised, built relationships, delivered quality work consistently. What changed was not her. What changed was the cost of what she was competing against.

The most useful question for translators now is not whether AI can do translation. It can, adequately, for much of what clients were previously willing to pay for. The question is what human translation provides that AI translation cannot — and whether the market for that specific thing is large enough, and growing fast enough, to sustain a career.

Based on Oxford Martin School research (Frey & Osborne, 2013) and Anthropic Economic Index (March 2026).