AI for translation — how to reach new-language customers without an agency

You want to reach customers in another language, but a translation agency costs more than the job is worth. AI has made good translation cheap enough for the smallest businesses — as long as you know where the line sits.

How good it actually is

Modern AI translation is well past the stiff machine translations of the past. It understands context, catches the tone reasonably well, and produces text that reads naturally in most major languages.

But it is not infallible. It stumbles on jargon, wordplay and cultural references, and it does not know when a phrasing sounds odd to a native speaker. It gives you a very good draft — not a finished publication you skip reading.

Three levels, three demands on quality

It helps to think in levels, because not everything needs the same precision:

• Internal and informal — an email to a supplier, a quick message. Raw AI translation is fine here.
• Outward-facing but low stakes — a short customer message, a standard reply. Read through, fix the obvious, send.
• Public and lasting — websites, contracts, marketing copy. Here someone who knows the language should proofread before it goes live.

The more visible and lasting the text, the more a human read-through is worth.

Translate meaning, not just words

The most common trap is to ask for a word-for-word translation. The result is technically correct and still wrong, because what works in one language does not necessarily work in another.

A slogan, a saying or a lightly ironic tone often falls flat when translated directly. Ask instead for the meaning and the effect to be preserved, not the words. For marketing copy this counts double: the point is that the new reader should feel the same thing, not that the sentence should mirror the original.

Mind jargon and names

The most common mistake is AI translating things that should not be translated: product names, company names, industry terms people know in the original language.

Give it a short list of words to leave untouched, and of how your fixed terms should be rendered in the new language. That keeps the same service from suddenly having three different names in English.

Keep the tone consistent

Text translated piece by piece, at different times, soon drifts in style. One page is formal, the next casual, and the whole thing feels patchy.

Decide on one tone in advance — for example "direct and informal, the way we speak to customers" — and ask for it every time. Consistent tone is what separates a translated website from one that feels written in the language from the start.

Quality checks that actually work

You do not need to understand the language yourself to weed out the worst errors:

• Translate it back. Ask AI to translate the text back to your own language, and see whether the meaning survived the trip.
• Check numbers, dates and names. This is where errors do the most damage.
• Get a native eye on anything public and lasting. One read-through catches what the machine cannot hear.

These steps take minutes, and they are the difference between a translation you dare to publish and one you simply hope for the best with.

Remember that language is not the whole job

A translated website is more than words. Date formats, currency, phone numbers and examples may all need adapting, not just translating.

A customer in another market expects a price in their currency and an example they recognise, not a local reference translated word for word. AI can help you catch these, but you have to ask. Ask what in the text should be adapted culturally, not just linguistically — that is often where a translation goes from understandable to credible.

If you would like to work out which parts of your communication are worth translating, you are welcome to book a quick call.