German to English Marketing Translations: How AI can help and hinder us

I’ve been translating for almost 15 years—*since before the days when AI was even a familiar term, much less a heavily hyped tool that we “have” to use in order to stay competitive. That means I’ve done the heavy creative work of juggling phrasing and nuance without having a sounding board at my fingertips. I’ve spent days pondering a particularly sticky phrase, only to have the perfect solution pop into my mind minutes after sending off the finished translation. So I appreciate AI. It helps me work more efficiently and generate ideas that I might not have come up with on my own.

However, I also know its limits. And I know that relying on it too heavily can hinder my creativity and my ability to add the special touch that my clients pay me for. In this article, I’ll discuss how I use AI in my marketing translations—and how I don’t use it. Then we’ll take a closer look at AI translations created by both Claude and ChatGPT, each using two different prompts. Finally, we’ll compare these to my own non-AI translation of the text to see what I did differently than the bots.

What are marketing translations?

First of all, what do I mean when I say marketing translations? And how do they differ from other kinds of translations? There’s some debate about this, but I’ll define “marketing translations” as the translation of marketing materials, advertising copy, blog content or corporate communications that are shared with prospective customers or other important stakeholders.

To be effective, marketing translations need to sound natural in the target language and also incorporate good copywriting techniques. For that reason, marketing translators will often adapt the writing style and/or the content to make it more appealing to the new target audience. When the adaptations are more extensive, this is called transcreation—a mix of translation and creative copywriting. But even with only minor adaptations, the writing style and ease of reading is of greater importance than with more technical translations.

How can AI help with marketing translations?

As you might have heard by now, AI can help you translate more efficiently. But don’t worry—I’m not here to sing the praises of MTPE. Quite the opposite.

I’ve found that I do my best work by translating manually and occasionally checking in with an AI tool to get ideas when I’m stuck on a tricky sentence, word, or phrase. But I never just toss my source text into the tool and post-edit. If I did that, I would surely miss some of the nuance and potential turns of phrase that come from reading the source text and translating word by word.

This is how I like to use AI in my work:

1. Phrasing suggestions: When I know that a sentence sounds clunky, but can’t quite find a better way to say it, I’ll copy-paste the source sentence into an AI tool (removing any identifying information if necessary) and ask it to give me five possible translations. I’ll then take elements of my version and the AI versions and puzzle them together until I’m satisfied with it.
2. Adapting tone and register: In marketing translations, the real challenge is often changing the tone and register for a new audience. If you adhere too closely to the source text, your marketing text won’t do its job, which is to appeal to an audience and get them to buy.
   Before AI, I used to let my translations sit and “marinate” for at least a day or two before refining the tone and style. And that’s ideal. But on a tight deadline, an AI tool can help you get that much-needed distance from the text without the wait.
3. Researching idioms and puns: Marketing materials are often filled with idioms, colloquial language, or puns that you may not understand immediately. (Even after living in Germany for 20 years, I still get blindsided occasionally!) AI can explain these to you in just a few seconds. That said, it’s always important to confirm AI’s explanation. As we all know, it can make mistakes.

How well can AI translate a paragraph of marketing text?

In the following example, I’ve slightly altered a paragraph of text from one of my actual translation projects and fed it to Claude and ChatGPT, using two different prompts. Let’s see how they each performed:

Claude – First attempt with a basic prompt

ChatGPT – First attempt with a basic prompt

Claude – Second attempt with a more sophisticated prompt

ChatGPT – Second attempt with a more sophisticated prompt

My translation (created in the pre-ChatGPT days):

Since 2016, we’ve been changing things up in the IT refurbishment industry. Our mission: Combining top tech with maximum sustainability. And we work hard to make it happen. From our humble beginnings as a box mover, we worked our way up to become an integral part of the European IT refurbishment industry. In 2016, we started out with a team of just four employees. Now, we have more than 90 people working here – and we’re also the proud recipients of the German Environmental Award 2024.

So, as you can see, the prompts you use do make a difference. With the simple “translate this” prompt, I find that ChatGPT produces a far more natural translation than Claude, although it is far from perfect. With the more sophisticated prompt, I prefer several elements of Claude’s translation—particularly the phrase “grown into a recognized name” as compared to ChatGPT’s “grown into an established player”.

However, for me, the crux of this paragraph is the last sentence. If translated directly, the information does not flow logically in English. It bounces from the number of employees, over to the environmental award, and back to the number of employees. In my translation, I restructured it (and split it into two sentences) to fix this issue. And in our AI test, only ChatGPT with the more sophisticated prompt managed to do something similar.

So, as you can see, AI can’t be relied on blindly. The particular tool and the prompts you use will make a difference. However, it is a useful tool to help you brainstorm, rephrase things, and research terms and idioms you might not be familiar with. And it might just help you find that phase you were looking for—before you hit send.

[*All em-dashes are my own]