The translation market's response to generative AI

AI changes the work clients commission, the prices they expect and the skills translators need. The transition includes post-editing, pressure on rates, niche specialisation and new review roles, while the scale of job displacement remains uncertain.

Limits of market figures

A freelance translator who loses a client cannot readily determine whether the client found another translator, had less work or adopted AI. ¹ Estimates of lost jobs, savings and profit gains would depend on self-reporting; in May 2024 no such estimate was available to one translator discussing the market. ^{2, 3} A claim that one translator had lost 90% of her work in the preceding year was a social-media anecdote, not a measured industry share. ¹ CSA Research was identified as an organisation following industry trends where future figures might become available. ²

Cost savings and profit gains depend on whose costs are counted, including hidden costs of operating AI and environmental costs that concerned some in the translation industry. ⁴ Language pairs also differ: English–Spanish has many competing translators, whereas a pair served by few translators can command higher rates. ^{4, 3}

Changing commissions

By September 2023 translation workflows had changed substantially, with professionals often asked to post-edit machine output rather than translate from scratch. ⁵ Translators were integrating translation memories and terminology-management systems to improve productivity, specialising in medical niches or moving towards creative, nuanced work such as medical marketing and health campaigns. ⁵

In May 2024 professional translators expressed both concern about job security and optimism about better tools. ^{6, 7} Language service providers were adopting AI rapidly to remain competitive, sometimes without understanding its repercussions or proper use. ^{7, 8} Some clients who had switched to AI returned to human translators, while others had not returned. ⁸

Hybrid commissions included thorough post-editing of openly AI-pretranslated texts whose clients doubted their quality. ⁸ Audiovisual work included checking whether AI-generated dubbing sounded natural in corporate videos whose scripts had been translated by people. ^{8, 9} Companies were combining AI and human expertise to improve their services. ¹⁰

End users appreciated greater accessibility and efficiency but also needed to understand where human expertise remained necessary. ^{9, 11} One interpreting client accepted that technology could convey 80% of his message but would not risk the other 20% being misunderstood or omitted. ¹¹ One translator saw a larger effect from neural machine translation's arrival in 2016 than from generative AI by May 2024, when clients remained wary of output problems. ^{1, 4}

Rates and client relationships

In March 2024 falling or sharply falling rates were a perceived threat among translators as neural machine translation and generative AI spread. ¹² Under client pressure, many agencies had shifted to machine-translation post-editing across subjects and difficulty levels, including medicine, leaving freelancers to accept a fraction of their usual rate or lose work. ¹² Top-tier medical translators increasingly sought direct clients paying premium rates for accurate human work, while less lucrative medical post-editing often went to less experienced or qualified translators, with potentially harmful consequences. ¹²

Direct clients also asked for quick, discounted revision of their own machine translations, such as a DeepL-translated handbook for patients preparing for open-heart surgery. ¹³ Translators then had to explain why medical translation was a high-stakes specialism deserving appropriate pay; the explanatory emails and calls were almost never billed. ^{13, 14}

In late 2024 an agency owner whose business did not use machine translation or post-editing reported colleagues' experience of reduced rates, which did not necessarily mean reduced hourly earnings. ¹⁵ The effect depended on per-word, hourly or project charging and the client type; reductions were considered more likely with agencies, which used machine translation more often, than with direct clients. ¹⁵

Adaptation and the pace of change

Medical confidentiality was viewed in late 2024 as a restraint on AI adoption because breaches could have serious consequences and clients were less likely to risk confidential work. ¹⁶ A contemporaneous forecast expected gradual development rather than an immediate pharmaceutical switch to machine translation or cloud engines, because the sector was comparatively conservative about fundamental workflow changes. ¹⁷

Adaptation advice in May 2024 covered five directions: ^{9, 10}

1. Specialise in niches where translators remain needed, accepting that not every sentence requires a human translator.
2. Improve post-editing skills as these commissions become part of the available work.
3. Reskill and upskill to make AI useful and remain competitive.
4. Diversify into other kinds of work.
5. Partner with medical writers who value human involvement, even where a human does not perform every task.

^{9, 10}

Participation in AI development

Refusing AI-review work can affect the quality of systems without stopping their development. ^{18, 19} In a May 2024 account, some expert Spanish speakers rejected text-to-speech review commissions because they did not want to help AI. ^{18, 19} A Spanish-speaking translator judged OpenAI's Spanish voices unnatural and difficult to listen to, likening them to a third-generation US Spanish speaker, and attributed this to inadequate expert review. ¹⁹ She expected similar problems in text and favoured contributing expertise to research rather than boycotting it. ¹⁹