How to Spot a Bad Machine Translation Before You Send It

Free machine translation tools are good enough, often enough, that it's tempting to paste a paragraph in, copy the output, and move on. Most of the time nothing goes wrong. But "most of the time" is a dangerous standard when the text is going to a client, an immigration office, a business partner, or anyone forming a judgment about you from your words. The failures that slip through machine translation are rarely obvious typos — they're small, structural problems that a non-native reader (or the tool itself) won't flag, but a native reader will notice immediately.

Here is a practical checklist to run before you send anything translated, whether you did it with a free tool, a paid one, or a person in a hurry.

1. Check register, not just meaning

Machine translation is good at getting the literal content right and much weaker at getting the tone right. A business email translated into French or Japanese can come out technically correct but jarringly informal, or the reverse — stiffly formal when the original was warm and casual. Read the translated version and ask: does this sound like how a professional would actually address this specific person? If the original was a friendly note to a colleague and the output reads like a legal notice, something got lost.

2. Hunt for literally-translated idioms

Idioms are the single most common failure point. Phrases like "hit the ground running," "ballpark figure," or "circle back" get translated word-for-word by weaker engines, producing sentences that are grammatically fine but mean nothing in the target language. If your source text has any idiom, slang, or figure of speech, search specifically for that phrase in the output and ask whether it would make sense to someone who has never heard the English original.

3. Verify every number, date, and unit

This is the check people skip most often, and it's the one with the highest cost when it fails. Decimal separators flip between English (1,000.50) and many European conventions (1.000,50). Date formats swap day and month depending on locale. Measurement units (miles vs. kilometers, Fahrenheit vs. Celsius) are sometimes converted and sometimes just copied over unchanged. Go through the translated text and manually cross-check every number against the source — don't trust that the tool got the conversion right just because it looks like a number.

4. Watch for false friends

False friends are words that look similar across languages but mean different things — "actuellement" in French means "currently," not "actually"; "embarazada" in Spanish means "pregnant," not "embarrassed." Machine translation has mostly solved the famous examples, but it still trips on domain-specific false friends: technical terms, brand names used as generic nouns, or industry jargon that has a different accepted meaning in the target language's professional context. If the text is for a specific industry, it's worth having someone who works in that industry in the target language skim the key terms.

5. Re-read names, titles, and legal terms with extra suspicion

Proper nouns, job titles, and legal or contractual terms are where machine translation is most likely to either mistranslate or inconsistently translate the same term two different ways in the same document. A contract that calls the same party "the Client" in one paragraph and "the Customer" in another, because the source used one word but the tool rendered it two different ways, creates real ambiguity. Search the document for your key defined terms and confirm each one is translated the same way every time it appears.

6. Read it back in the original language (the "round-trip" test)

A quick, imperfect but useful check: take the translated output and run it back through a different translation engine into the original language. You're not checking for a perfect match — paraphrasing is normal — but if the round-trip version says something structurally different from what you started with (different subject, different claim, missing negation), that's a strong signal something broke in the first pass. A missing "not" is the single most dangerous failure mode here: it flips the meaning of a sentence entirely while looking completely unremarkable.

7. Check pronouns and ambiguous references

Languages differ enormously in how much they rely on context versus explicit pronouns. English leans on "he/she/it/they" heavily; many languages drop pronouns when the subject is implied by verb conjugation, and machine translation sometimes inserts the wrong one when converting between the two styles, or loses track of who "they" refers to across a long paragraph. This matters especially in narrative or explanatory text with multiple people or entities being discussed.

8. Have a native or bilingual reader skim it — even five minutes helps

None of the checks above fully substitute for a native-level read of the final text. You don't need a full professional review for routine, low-stakes material, but if the translation is going to a client, a government form, or anyone who will judge you by the result, a five-minute skim from someone fluent in the target language will catch tone and naturalness problems that no checklist can.

When it's worth paying for a human (or AI-assisted) pass

Free machine translation is genuinely fine for understanding a foreign-language email or getting the gist of an article. It's a worse bet for anything outbound and high-stakes: cover letters, business correspondence, formal requests, or any document where getting the tone and precision right affects how you're perceived. In those cases, a translation done with attention to register, idiom, and consistency — not just literal word-for-word conversion — is worth the cost difference.

If you have a document up to about 1,000 words that needs to read naturally and accurately in another language, Lucid Desk's Translation offer delivers a fluent, register-aware translation as plain text within minutes, preserving tone and meaning rather than converting word by word.