On this page
The verdict, line by lineWorked examplesWhat research on creativity showsPost-editing is a service with a scopeFive questions to decide each lineCommon questionsAsk about machine translation vs human translation for Korean marketing and the honest answer is: it depends on the line. A product launch contains a spec table, a returns policy, a hero headline, three calls to action and a push notification. Those five pieces of text get five different verdicts.
The verdict, line by line
This is how we route a typical Korean launch kit. "Machine draft" means an engine or AI tool produces the first version and a Korean linguist edits it; "human" means a Korean writer works from the source and the brief.
| Copy element | Verdict | Why |
|---|---|---|
| Specification tables, dimensions, materials | Machine draft, then edit | Short factual strings; the risks are units, number formats and terminology, which a reviewer can check quickly. |
| Shipping, returns, account help | Machine draft, then edit | Informational; tone needs to be polite and consistent, which editing can fix. |
| Long product descriptions | Either, depending on volume and visibility | Mixed: facts plus persuasion. Top sellers deserve human copy; the long tail can start from a draft. |
| Headlines, slogans, taglines | Human (transcreation) | Wordplay, rhythm and implication are the content. A literal version is accurate and flat. |
| Calls to action and buttons | Human | Korean commands need the right politeness and length; tiny strings carry the most brand voice. |
| Push, text and messenger promotions | Human | Hard character limits, legal lines inside the copy, and one chance to sound like a person. |
| Claims, comparisons, health or performance wording | Human, with a flag to the reviewer | A stronger Korean claim is a claim you must prove. |
Worked examples
The pairs below are illustrative, written for this article. The engine-style cards describe the kind of output an unguided engine or a one-line AI prompt typically gives; the human cards describe what a Korean copywriter does instead.
Correct and lifeless
Big sound. Small footprint.The engine turns the two fragments into two flat, complete statements. Every word is there. "Footprint" comes through as a literal mark left by a foot or as an environmental term, and the two-beat rhythm is gone.Same idea, written in Korean
Big sound. Small footprint.The writer recasts the line as two parallel noun phrases: sound that fills the room, a body small enough for one hand. The contrast survives as a pair of parallel noun phrases, which Korean headlines use naturally. "Footprint" is replaced by what it meant: size.An order to a stranger
Grab yours before they're gone!The engine keeps the exclamation and uses a plain, blunt imperative. "Grab" becomes a physical verb, "yours" a pronoun Korean would drop, and the plain command sounds rude from a brand.Urgent but polite
Grab yours before they're gone!The writer states that stock is limited and closes with a polite request ending. The urgency moves into a fact (limited stock), and the verb becomes a polite suggestion that fits on a button.Notice what the engine-style versions get right: meaning, grammar, terminology. That is why they pass a quick check by someone who reads Korean as a second language, and why they still underperform. Marketing copy fails on voice, rhythm and politeness, which a word-level check does not test. We cover the grammar behind these patterns in why direct translation fails in the Korean market and the speech-level problem in Korean speech levels and where AI translation breaks.
The same engine does well on a line like "Dimensions: 120 x 80 x 45 mm. Weight: 1.2 kg." There is nothing to persuade, the Korean format is predictable, and a reviewer can confirm it in seconds. That is the whole argument for routing by line rather than by document.
What research on creativity shows
We know of no solid public benchmark for machine translation of Korean advertising copy, so be wary of anyone quoting precise success rates. A useful adjacent study is Guerberof Arenas and Toral (2022), which compared raw machine translation, post-edited machine translation and unaided human translation of a short story into Catalan and Dutch. Reviewers rated human translation highest for creativity, post-editing in the middle and raw machine output lowest, unanimously. The authors found the engine tended to produce literal solutions, and that post-editing constrained translators' creativity.
The machine draft does not just save time. It also sets the shape of the sentence the editor starts from.
The study is about literary text, not ads, and not Korean. But the mechanism it describes is the one we see in marketing review: an editor handed a fluent literal draft tends to fix errors rather than rewrite, and the copy keeps the source's shape.
Post-editing is a service with a scope
ISO 18587 sets requirements for full, human post-editing of machine translation output and for post-editors' competences. That is a real, professional service, and for informational text it is often the efficient choice. For persuasive copy, ask what the editor is being paid to do. If the brief is "fix errors", you get correct, flat Korean. If the brief is "make it work as Korean marketing", the editor will rewrite most short lines, and the cost approaches human translation while the literal draft still pulls on the result. Our post on combining AI and human translation sets out a risk-scoring method for routing content.
Five questions to decide each line
- Would a Korean customer notice if this line sounded translated? If yes, a human writes it.
- Does the line make a claim, comparison or number someone must prove? Flag it, whatever the route; see legal and regulatory checks for Korean marketing translation.
- Is there a hard length limit (button, push title, ad headline)? Machine drafts rarely respect it.
- Is the source line itself wordplay or rhythm? Then there is nothing literal to translate; brief it as transcreation.
- Will hundreds of similar strings follow? Then fix the voice in a style guide and glossary first, and let a machine draft the volume against it.
Common questions
Is machine translation ever good enough for Korean marketing?
For some parts of it, yes: specification tables, shipping and returns text, long-tail FAQs and internal drafts, provided a Korean reviewer checks terms, numbers and tone. For headlines, slogans, calls to action and short push copy, a machine draft rarely survives review.
Is post-editing cheaper than human translation for marketing?
For informational text, often. For persuasive copy, the editor frequently ends up rewriting the line, so the saving disappears and the draft can anchor the editor to a literal structure.
Can an AI tool be told to write in a friendly, polite Korean voice?
Instructions help with consistency, but someone still has to decide the voice, check that it holds across hundreds of strings, and catch claims that changed. Treat the tool as a drafting aid inside a reviewed workflow.
Not sure which lines need a human?
Send us your launch kit. We mark each element as machine-draft or human, and quote accordingly. Our services · Get in touch →