.webp)
Every freelance translator has felt it by now. A client sends a file that has already been run through an AI engine, and the brief no longer says "translate this," it says "post-edit this." Machine translation post-editing, usually shortened to MTPE, has quietly become one of the most common ways translation work reaches freelancers in 2026. The problem is that most translators were never taught how to do it well, how to price it fairly, or how to protect their own quality standards when a client assumes the hard part is already done. This guide walks through all three.
What MTPE Actually Is
Post-editing means taking raw machine-translated text and correcting it so it reads, and functions, the way a human translation would. It sits between two extremes. Raw machine translation is fast and free but unreliable for anything client-facing. Translating entirely from scratch is the most reliable option but the slowest and most expensive. MTPE exists because most content, most of the time, does not need either extreme. It needs a competent human checking, correcting, and taking responsibility for what an engine produced.
The industry generally recognizes two service levels, formalized in the ISO 18587 standard for post-editing of machine translation output. Understanding which one a client is actually asking for, before you quote or accept a project, avoids most of the friction that makes MTPE frustrating.
| Service level | What it means | Typical use case |
|---|---|---|
| Light post-editing | Fix only errors that change meaning, leave awkward but understandable phrasing alone | Internal documents, support tickets, user reviews, high-volume low-visibility content |
| Full post-editing | Correct meaning, grammar, style, terminology and tone until it reads as if written by a human translator | Marketing copy, legal and medical documents, UI text, anything published under the client's name |
If a client has not specified which one they want, ask before you start. Quoting and delivering full post-editing when they expected light, or the reverse, is the single most common source of disputes in MTPE work.
A Workflow That Actually Holds Up
Good post-editing is not skimming the machine output and fixing typos. It is a deliberate pass through the text, segment by segment, checking meaning first and polish second. A workflow that has held up well for translators handling MTPE at volume looks like this.
First, compare engines rather than trusting one blindly. Different AI models handle idiom, register, and terminology differently depending on the language pair and subject matter, and the model that produces the cleanest draft for a legal contract is rarely the same one that handles a marketing tagline well. WordBeam's multi-engine translation view shows GPT, Claude, Grok and DeepL output side by side for the same segment, so you pick the better starting draft instead of editing whichever one the client happened to attach.
Second, check the draft against your translation memory before you touch a single word. If a segment or a close match already exists from a previous project with the same client, the machine draft should defer to it, not the other way around. A neural translation memory that learns your own phrasing keeps terminology and tone consistent across a project instead of drifting sentence to sentence, which is one of the more common tells that a document was machine-edited rather than properly post-edited.
Third, run structured quality checks rather than relying on a final read-through alone. Numbers, dates, placeholders, tags and untranslated terms are exactly the kind of small errors a tired eye skips over on the fifth hour of editing but a client catches instantly. Automated QA checks and AI proofreading catch these before delivery, which matters more in MTPE than in from-scratch translation, since the source text was never written by a human who might have naturally avoided some of these traps.
Where Post-Editing Goes Wrong
Two failure modes account for most of the complaints clients have about post-edited work. The first is under-editing, where a translator accepts the machine draft's sentence structure even when it is unnatural, because rewriting the whole sentence feels slower than fixing a word here and there. Readers notice. Text that is grammatically correct but structurally foreign reads as machine-translated even after a human has touched it. The second is over-editing, where a translator effectively retranslates from scratch on a light post-editing brief, which takes full-translation time while getting paid the discounted MTPE rate. Knowing which failure mode you default to, and correcting for it deliberately, is worth more than any tool.
Pricing MTPE Fairly
MTPE is almost always priced at a discount to full translation, typically 30 to 50 percent off the per-word rate for light editing and 15 to 30 percent off for full editing, though this varies by language pair and how clean the source engine's output actually is. The discount should reflect real time saved, not an arbitrary industry figure a client quotes at you. If a source document is poorly formatted, technical, or in a language pair where machine engines still struggle, the time saved shrinks and the rate should reflect that honestly rather than absorbing the difference as unpaid labor. A fair rate protects the sustainability of taking MTPE work at all.
The Bottom Line
Machine translation post-editing is not a lesser form of translation, it is a different discipline with its own skill set, its own failure modes, and its own pricing logic. Translators who treat it as a quick cleanup job tend to burn out on the underpaid version of it. Translators who build a deliberate workflow around engine comparison, translation memory, and structured QA tend to find it is one of the more sustainable ways to grow a freelance practice as AI-assisted content keeps expanding across every industry.