Machine translation post-editing, or MTPE, is when an AI engine translates a text and a human translator then reviews, corrects and adapts it. It isn’t traditional human translation, nor is it letting ChatGPT do the job and sending the output straight to the client. It’s designed for high-volume content, tight deadlines and technical texts where speed matters. It doesn’t work for sworn translation, critical medical documentation or sensitive legal texts. In this article I explain when it makes sense, when it doesn’t, and how it differs from using ChatGPT directly.
What exactly is MTPE?
MTPE stands for Machine Translation Post-Editing. It’s a two-step process: first, a machine translation engine (DeepL, Google, proprietary models) translates the document; then a native human translator reviews the output, corrects it and adapts it to the sector and the end reader.
The international standard governing this process is UNE-ISO 18587:2020 (the Spanish edition of ISO 18587), which defines how full post-editing must be carried out so that the result is on a par with professional human translation (UNE-ISO 18587:2020).
Put simply: the AI proposes, a human decides.
The two types: light post-editing and full post-editing
Not all post-editing is the same. There are two levels, and choosing the wrong one costs you either money or quality.
The question you need to ask yourself as a buyer is very specific: will someone outside my company read this text and judge the company by how it’s written? If the answer is yes, you need full post-editing. If you just need to understand what it says, light post-editing will do.
When does MTPE make sense, and when doesn’t it?
This is the nuance almost nobody in the industry tells you about. MTPE is a very good tool, but it isn’t the answer to everything.
When it makes sense:
- Repetitive technical documentation: operating manuals, technical data sheets, safety data sheets.
- Extensive product catalogues in several languages.
- E-commerce content with thousands of product descriptions.
- Multilingual internal documentation (policies, training, support).
- Subtitling and transcription of corporate videos, always with review.
When it does NOT make sense:
- Sworn translation. By law, it is signed by a sworn translator appointed by the Spanish Ministry of Foreign Affairs, and the liability is personal. There’s no place for AI here.
- Critical medical documentation: patient information leaflets, clinical trials, regulatory labelling. A single error can have regulatory or health consequences.
- Contracts and sensitive legal documents. Interpreting contractual nuances isn’t something you delegate.
- Creative marketing and transcreation. When the value lies in the tone rather than the literal meaning, AI falls short.
- Crisis communications and press releases where every word counts.
My advice is simple: if an inaccuracy could cause you a regulatory, commercial or reputational problem, post-editing isn’t the way to go. That calls for human translation.
The difference between what ChatGPT does and what we do at a translation agency
This is the question I’ve been asked most this year. “Rubén, why do I need you if I’ve already got ChatGPT?” And it’s a fair question.
ChatGPT and DeepL have taken a huge leap forward. For short, general texts and in languages with a large training corpus (English, Spanish, French, German), the output quality is good. Sometimes surprisingly good.
What AI can’t do on its own – and why MTPE exists as a service:
- Guarantee your corporate terminology. If your company has called something a “pumping station” for 10 years and the AI translates it as “plant pumping”, it shows. An agency maintains your glossary and your translation memory.
- Take responsibility for the result. If the AI makes up a figure or a proper name (and it does), there’s nobody to hold accountable. At a professional agency the work is on record: an identified post-editor, translation memory and version control. The standard for translation services is ISO 17100:2015, which Overseas has held since 2017; the specific standard for post-editing is ISO 18587:2020. Both guarantee quality traceability and accountability for the final result.
- Adapt to your sector. A heavy engineering text, a cosmetics text and a banking text all use different jargon. General-purpose AI mixes registers. A sector-specialist translator doesn’t.
- Meet the standard. For public tenders, certifications, audits and certain corporate clients, the translation has to come from a certified provider. ChatGPT isn’t one.
- Genuine confidentiality. Whatever you put into a public chatbot goes into its systems. An agency signs an NDA and works with closed workflows.
In a nutshell: ChatGPT translates. An agency offering MTPE delivers translations that someone puts their name to.
What the industry is saying in 2026
This isn’t just my opinion: it’s what the public data from the industry’s research firms shows.
- In Slator’s 2025 survey of the world’s leading agencies, 84% of providers said their clients had explicitly asked for human review of AI translations in the past year (Slator AI Translation Resources).
- In the same survey, between 90% and 98% of machine translation users apply some level of post-editing. Hardly anyone publishes raw AI output.
- CSA Research, in its report Navigating the Post-Localization Era, surveyed 152 B2B companies in 2026: nearly 95% already use AI or machine translation, but they treat it as a managed process, not a button to press (CSA Research).
- Nimdzi Insights estimates that the global language industry is worth around 73.4 billion dollars in 2026, with AI + human review services being the fastest-growing segment.
In other words, the market isn’t replacing agencies with AI. It’s replacing “all-human” with “AI + professional human review”. That’s exactly what MTPE is.
How we do it at Overseas Translations
Here’s the real process, no filters. At Overseas we’ve been working with MTPE for several years and have built it into our certified workflow.
- We analyse your document with our CAT tool (the software used to manage translation projects, memories and glossaries) and decide whether it’s suitable for MTPE or needs direct human translation.
- We run the text through the DeepL API or whichever engine best suits your language pair, always on a closed business account to guarantee confidentiality.
- We assign the post-editing to a native human translator who specialises in your sector (engineering, marketing, medical, legal, technical).
- We apply your translation memory and glossary, if you have them. If not, we build them with you from the very first project.
- We deliver the final file in whatever format you need (Word, InDesign, XLIFF, HTML – you name it).
The whole process runs under our ISO 9001:2015 and ISO 17100:2015 certifications, held since 2017. ISO 17100 requires us to work with qualified professional translators and to review every project. In our MTPE projects we also apply the criteria of ISO 18587:2020, the specific post-editing standard, as industry best practice. If you’d like to find out more, I explain it here: AI translation services and MTPE in detail.
If you’d like to know how much a specific project would cost, or whether full or light post-editing is right for you, we can look into it in 10 minutes by phone or email. We’re here to help.
Does your company translate with AI but want to make the process more professional? Email us at [email protected] or visit ovstranslations.com/traduccion-ia-mtpe and we’ll assess your specific case.
