Polish Translators' Earnings Drop 45 Percent as AI Takes Over the Market

Back in 2023, a technical translator who spoke to the media under the name Radosław recorded the best earnings of his nineteen-year career. A year later his income had dropped by 45 percent. The reason is no secret: translation agencies switched en masse to the DeepL and ChatGPT language models, leaving humans to fix machine translations for a fraction of the old rate.

From record year to collapse in twelve months

Radosław started earning from translation while still a student, nearly two decades ago, and eventually set up his own business specializing in technical texts. That segment was long considered resistant to automation, since it demands precise terminology and familiarity with the client's industry. Around 2013 his monthly revenue reached 13,000 zlotys, which he admits was an impressive result at the time.

The first wave of change came with CAT tools, computer-assisted translation software that stores previously translated sentences in a translator's own database. They were meant to speed up the work, but agencies quickly found a way to use them to cut costs: for sentences 80 percent similar to ones already translated, they began paying only 40 percent of the full rate, regardless of the fact that the database was the translator's own property and life's work.

DeepL and ChatGPT rewrite the rules

The real turning point came with the rise of large language models. Radosław initially used DeepL and ChatGPT himself, which let him speed up his work and post the best earnings of his career in 2023. A year later the roles reversed: agencies started sending him texts already translated by machine, expecting only proofreading and editing for a rate lower than the previous full translation price.

The market collapsed because agencies quickly figured out how to exploit automated translation - Radosław, technical translator

The billing mechanism changed the nature of the work itself. Instead of translating a text from scratch using his own experience and terminology, the translator increasingly acts as an editor correcting an algorithm's output. The MTPE model, machine translation post-editing, is becoming the industry standard rather than the exception.

Not just one case

Radosław's story, first reported by Onet, spread to other outlets in recent days, including National Geographic Polska, showing the issue resonates well beyond a single individual case. In discussions on industry forums, other users confirm the scale of the shift, describing for example using GPT to translate an entire online store's catalog, more than three thousand product descriptions from English into Polish, for the cost of just over a dozen dollars in API tokens.

Industry critics point out, however, that the quality of machine translation can be uneven and works best for simple advertising or catalog texts, while it struggles more with the nuances of specialized, legal, or literary texts. Even so, companies looking to cut costs are increasingly less likely to ask about quality differences, since they can pay for the end result alone rather than the process.

What it means for Poland

The phenomenon fits into a broader trend of falling rates in language freelancing, which Polish labor market reports have been noting for months. The translation profession, alongside graphic designers and copywriters, is among those where automation through generative tools is advancing fastest and translating most directly into workers' incomes, not just into faster turnaround.

For business clients, this means cheaper and faster translations, but also raises the question of who is responsible for errors in machine-translated technical documents, contracts, or safety instructions. For translators themselves, the outlook is tougher: either retrain toward editing and quality control of AI translations, or specialize in niches where precision and legal liability still require a human.

Despite the drop in income, Radosław plans to stay in the profession, shifting his services toward verifying and editing machine-generated text. Industry observers say this is the scenario awaiting most translators currently working for agencies in the coming years.