Translation Is Cheap. Mistranslation Is Expensive. Here's How to Tell.

Some of the most famous translation mishaps in business history:

• HSBC spent $10 million rebranding after its 2009 "Assume Nothing" campaign translated in several markets as "Do Nothing"
• Electrolux marketing its vacuums in the US with the tagline "Nothing sucks like an Electrolux" — technically correct, catastrophically tone-deaf
• Pepsi's early 1960s entrance into China with "Pepsi brings you back to life" rendered as "Pepsi brings your ancestors back from the grave"

These are the horror stories everyone tells. The ones that don't get told are the hundreds of modern companies shipping machine-translated product descriptions, support articles, legal notices, and email subject lines — and losing customers one mildly-wrong sentence at a time.

The honest truth about modern AI translation

It's good. It's really good for a wide band of use cases. Modern neural MT and LLM-based translation handle grammar, idiom, tense, and most cultural references at a quality that would've been impossible five years ago.

For a blog post explaining how to configure your Gmail settings, it's indistinguishable from a professional translator. For the copy on your About page, it's probably better than the translator would've been because the original English copy isn't exactly Hemingway either.

For 95% of a typical website, modern AI translation is the right tool.

Here's where it isn't.

The four places where machine translation breaks down

1. Brand-critical copywriting

Your tagline. Your hero headline. The 6-word promise on your pricing page. Anything where the specific word order, rhythm, cultural connotation, and brand voice are the point. AI can produce "correct" translations of these that are also flat, generic, and instantly forgettable. A good copywriter adapts the idea into something that lands in-culture — sometimes replacing the original entirely.

This is called transcreation and it's worth paying for.

2. Legal and regulatory language

Privacy policies, terms of service, refund disclosures, medical claims, financial disclaimers. A mistranslation here isn't just awkward — it can be actionable. In the EU, a regulator taking a strict reading of your German privacy policy could fine you based on what the text actually says, not what you meant. Use a human lawyer-translator or a legal-grade translation service for these. Specifically.

3. Named-entity trip hazards

Your product is called "Spark." Your CTA button says "Spark it up." Machine translation will happily translate "Spark" as the common noun in every language — and it'll render your product name as completely different words on every page. Most translation widgets now support a do-not-translate list for exactly this reason. Use it religiously for brand names, proper nouns, and product-specific jargon.

4. Humor, irony, and wordplay

Puns don't survive translation. Irony depends on cultural context that doesn't port. Humor is dense with shared reference, and a shared-reference joke in one language is just a confusing sentence in another. If your brand voice is dry-witty-clever (think a lot of modern SaaS marketing), expect that flavor to flatten in machine translation. It's not wrong, it's just beige.

The framework: what to translate, how

The way to read this: start with AI translation across the whole site. Then do a one-time pass on the top ~20 strings (hero, nav, primary CTAs, key taglines) and either hand-edit them or commission transcreation. For legal, get a specialist regardless.

How SiteDialect handles this

Three relevant mechanics:

1. Do-not-translate list — your brand names, product names, proper nouns are served as-is in every language
2. Manual override — for any specific string, you can write the human-translated version directly in the dashboard and we'll serve that instead of the AI version
3. Glossary — you can specify how a term should be translated (e.g., "dashboard" should always render as "panel de control" in Spanish) and we'll apply it consistently

The workflow we recommend: launch with AI translation across all pages, monitor the highest-traffic translated pages for the first month, flag any awkward or brand-sensitive strings, and override those specific strings in the dashboard. You get the 95% coverage of AI at close to zero cost, and the 5% that matters most gets human attention.

The real answer

Translation is cheap when you're shipping a lot of content that mostly needs to be understood. Transcreation is what you pay for when you need copy that persuades. Most websites need both in roughly the ratio the table above suggests.

Don't go 100% on either side. A site that's 100% AI-translated flattens its brand voice in every non-English market. A site that's 100% hand-translated took nine months to ship, cost $200k, and will still look 80% like the AI version a customer pastes into DeepL.