Post-editing to ISO 18587 Your AI translation, checked by a human.
You have run a text through DeepL, Google Translate or a language model such as ChatGPT. A native specialist checks the result to ISO 18587 for fidelity to the original, terminology, hallucinations and style. Light for internal texts, full for anything ready to publish. Send us your machine output.
Four ISO certificationsTranslation partner for companies with high standards
- What it is: The human rework of machine and AI translations to ISO 18587, by a native specialist.
- Who it's for: Companies that use DeepL, Google Translate or language models and need the result to be dependable.
- What it delivers: Fidelity to the original, consistent terminology and the right style, from €0.05/word depending on the level.
- The difference: Post-editing works on texts that came out of a machine. Texts written by people are handled by editing & proofreading.
What the machine phrases fluently but never verifies
Four weak spots that recur in AI translations
Modern machine translation sounds convincing even when it is wrong. That is exactly the danger: a fluent sentence looks right until somebody with subject knowledge reads it against the original. Post-editing closes that gap.
Language models sometimes add content that never appears in the original. In fluent prose nobody notices. In a contract or a patient information leaflet, they do.
The same component is called one thing here and another there. The machine doesn't know your agreed terms; a specialist enforces your glossary throughout.
Machines rarely hit your brand voice. A sharp campaign turns into a correct but colourless text. In marketing, that costs you impact.
One word with two meanings, a reference spanning several sentences: where context decides, the machine tends to miss it. The specialist reads against the original.
Light, full or with a four-eyes check
The level depends on where the text is going
ISO 18587 defines two levels, and a second pass can be added on request. The choice follows the destination of the text, meaning whether it stays in-house or goes out into the world, rather than the budget. Both levels are handled by a native specialist.
Corrects what makes the text unclear or wrong: accuracy, grammar, terminology in broad strokes, hallucinations removed. For internal texts and knowledge bases. From €0.05/word.
Revises the text so that it is hard to tell apart from a human translation: style, terminology in detail, tone pitched at the target market. For texts that go out into the world. From €0.08/word.
On request, a second specialist reviews the result, as known from ISO 17100. For publication-ready or regulated content with an audit trail.
If the text stays in-house, light is enough. If it goes to customers, the public or the authorities, take full. If in doubt, we advise you before the order on which level suits your text. You set the level once per document type, and every order then runs at that level automatically.
What post-editing costs
A per-word rate with a worked example
We bill post-editing by the word, because the effort grows with the volume and the per-word rate stays comparable across languages. We name the binding fixed price once we have looked at your file.
| Service | Price from | Suitable for |
|---|---|---|
| Light post-editing (ISO 18587) | €0.05/word | Internal texts, knowledge bases, employee communications |
| Full post-editing (ISO 18587) | €0.08/word | Websites, marketing, annual reports from DeepL or AI output |
| Four-eyes check (on request) | on request | Publication-ready texts, regulated content, audit trail |
| Express (24 hrs) | +30% to 50% | Deadline-critical deliveries, campaign launches |
A worked example: an annual report of 15,000 words whose English DeepL translation is to be revised as full post-editing comes to about €1,200. As light post-editing for the internal version, from about €750. The pace runs at roughly 4,000 to 5,000 words per working day. These are guide figures and depend on the level, the language pair and the quality of the source output. For recurring projects we agree framework terms with tiered volume discounts. More on per-word rates on the prices page.
When post-editing pays off most
Wherever volume meets the need for reliability
The larger the volume of text and the higher the demand for accuracy, the more the human check repays itself. Typical cases:
Thousands of items, machine translated, that still have to be right everywhere in e-commerce. Full for the sales pages, light for internal fields.
Manuals from the editorial system, where wrong terminology gets expensive. Post-editing secures the technical translation that came out of the machine.
Help articles and FAQs in many languages that need to be quickly understandable without being print-ready. The classic case for light post-editing.
Content that goes out into the world and has to land. Full post-editing is worth it here, so the text sounds like your brand rather than like a machine.
Documents where an error of meaning is a risk. Full post-editing with a four-eyes check, handled confidentially to ISO 27001.
You already translate with DeepL or a language model. Post-editing is the human control layer on top, connected directly to your systems through tolingo connect where the need is ongoing.
Post-editing assumes a usable raw output. Where the machine translation is too poor, for instance with rare language pairs or highly specialised content, the rework becomes more effort than a clean new translation. In that case we tell you so before the order and recommend specialist translation to ISO 17100. We check the quality of the source output before we propose a level.
Post-editing in the translation process
What comes before it, what sits beside it
Post-editing assumes a machine translation. Depending on what you already have and where the text is going, a different service fits:
No translation yet? The AI translates on EU servers and a specialist edits straight afterwards to ISO 18587. Translating and checking in one order.
Your text was written by people rather than machine translated? Then editing is the right route, at an hourly rate on request.
Where no machine should be involved at all: a native specialist translates to ISO 17100 and a second revises. The gold standard for demanding texts.
In post-editing, hits from your store apply first and new segments go into the level you chose. Costs fall and terms stay consistent.
Common questions about post-editing
Levels, languages, distinctions, security
What's the difference between light and full post-editing?
Light post-editing corrects only what makes the text unclear or wrong, and language and style stay largely as the machine produced them. Suitable for internal documents, from €0.05/word. Full post-editing revises the text to the point where it is hard to tell apart from a purely human translation. Suitable for websites, marketing and annual reports, from €0.08/word. Both to ISO 18587.
Can you check DeepL or ChatGPT translations too?
Yes. For any machine or AI output, the service runs as post-editing to ISO 18587, whether the result came from DeepL, Google Translate or a language model such as ChatGPT or Claude. We check fidelity to the original, terminology, hallucinations and style.
How does post-editing differ from editing & proofreading?
Post-editing works on machine translated texts to ISO 18587. Editing & proofreading works on texts written by people, such as a German or English original, at an hourly rate. In short: if the text came out of a machine, post-editing is right. If it was written by hand, editing is.
Which languages do you offer?
Over 220 languages, in both directions. Every language variant is checked by a native specialist, so British against US English, for example, or continental against Brazilian Portuguese. Common language pairs such as German and English have their own service pages.
How fast is delivery?
We reckon on roughly 4,000 to 5,000 words per working day, depending on the level. An annual report of 15,000 words is realistic as full post-editing in three to four working days. Express within 24 hours is possible, with a surcharge of 30% to 50%.
How do you protect confidential documents?
tolingo is certified to ISO 27001. Files are transferred encrypted and processed exclusively on European servers, and our specialists sign NDAs. For annual reports and contracts we set up dedicated project environments with restricted access on request.
Text services
Looking for the right text service?
Post-editing, editing & proofreading, copywriting and transcription sit side by side. The overview shows which service suits which text.
Give your AI translation the human check.
Upload your machine output, ideally along with the original text. We'll recommend the right level and name the fixed price and delivery date.
Reply within ~10 minutes (within business hours) · ISO 18587 · NDA on requestGlossary: terms around post-editing
- Post-editing (MTPE)
- The human rework of a machine translation. The abbreviation MTPE stands for machine translation post-editing. A native specialist corrects the machine output to ISO 18587 for fidelity to the original, terminology and style.
- Light and full post-editing
- The two levels defined by ISO 18587. Light removes only what distorts the meaning and suits internal texts. Full works the text up close to a human translation and suits content that is ready to publish. Which level fits is set out in the section on light or full.
- ISO 18587
- The international standard for the post-editing of machine translations. It defines the requirements for the process and for the competence of post-editors, along with the distinction between light and full. It complements ISO 17100 for purely human translation.
- Hallucination
- A failure mode of machine translation in which the system produces content that does not appear in the source text. The sentence reads fluently but conveys something that was never there. Post-editing catches such places by reading against the original.