Multilingual Copywriting

Ask any large marketing team in 2026 what changed in translation, and they will not mention a single vendor. Machine translation is now good enough for a first pass in most language pairs, and a large language model will draft a passable version of almost anything in seconds. That part of the job is largely solved, and pretending otherwise does not serve a client well.

So the interesting question for a Geneva agency working across three languages a day is not "can a machine translate this for you." It can. The interesting question is what still needs a person once it has.

What machine translation is actually good at

Give a modern engine a technical document, a set of product specifications or an internal policy, and it will usually produce something accurate and usable. Consistency across a long document, terminology recall, turnaround on volume — these are places where the tools have genuinely closed the gap on a human translator, and it would be dishonest to argue otherwise to a client who is paying for the difference.

Where it still falls over, reliably, is register. A model will translate the words correctly and still get the temperature wrong: too formal where the brand is warm, too breezy where the audience expects gravity, confidently unaware that a phrase which reads as friendly in Paris reads as presumptuous in Geneva. It has no way to know that, because register is not really a property of the sentence. It is a property of the relationship between the brand and the reader, and that relationship differs by market even when the language does not.

The part that does not automate

Three things still need a person who knows both the source and the target culture, not just the two languages.

• Cultural register. Swiss French is not Parisian French, and Swiss German is not spoken the way German is written in Berlin. A sustainability claim that reads as measured in Zurich can read as evasive in São Paulo, and vice versa. None of this shows up in a word-for-word check — it only shows up when someone who has actually worked both markets reads the copy and asks whether it sounds like it was written for this audience or merely delivered to it.
• Transcreation judgement. Slogans, headlines and anything built on wordplay rarely survive literal translation, and the fix is not a better translation — it is a different piece of copy that does the same job. That is a creative decision, made by someone weighing tone, rhythm and cultural reference against the brief, not a linguistic one a model can be prompted into reliably. We covered some well-known examples of what happens when nobody makes that call in our piece on ad campaigns that got lost in translation — Braniff's "Fly in leather" landing as "fly naked" in Mexican Spanish is the one people remember, and it is a translation failure a transcreation process exists specifically to catch.
• Accountability. When a claim goes out under a client's name — in a sustainability report, a regulatory-adjacent product description, an internal message from a chief executive — someone has to be able to say where it came from and stand behind it. A model cannot do that. It has no source, no memory of the brief, and no stake in getting it right. A named writer and a second human read do.

The Swiss case for still using people

Switzerland is the sharpest possible argument for this, because it collapses the whole problem into one small country. Three national languages, four if you count Italian, each carrying a genuinely different register and often a different audience expectation for the same brief. Geneva's international and NGO-facing tone is not Zurich's financial-sector tone, and neither is quite what a French or German copywriter working from Paris or Berlin would produce by default. Getting Swiss French and Swiss German right is not a matter of running the English draft through an engine twice — it takes people who work inside those registers routinely enough to hear when something is technically correct and still slightly off.

That is the specialisation we have built the agency around: English, French and German, inside one small market with unusually high expectations of precision, for clients who cannot afford a tone that reads as translated.

So what are you actually paying for

Not the conversion of words from one language to another — that part is close to commoditised, and a client who only needs that is right to look at the cheaper, faster option. What an agency still earns its fee for is the judgement layer on top: knowing which register a market expects, knowing when a literal translation will misfire and what to write instead, and being willing to put a name to the result. That is consistent with how we think about AI generally — we set it out in full in our position on AI-generated content — disclosure before you sign, a named writer, a second human read, every claim traceable to something you actually gave us.

None of that requires pretending the tools do not exist. It requires being honest about which part of the job they have taken over, and charging for the part they have not.

Working across English, French and German markets and want a second, cultural read on copy that already exists — or a full transcreation from scratch? Let's talk.