--- title: "Is Google Translate accurate enough for a website?" description: "A 2021 study found Google Translate kept the meaning of 55% to 94% of sentences, by language. What that means for a website, and which pages need review." url: "https://www.langrelay.com/blog/is-google-translate-accurate" --- 1. [Home](https://www.langrelay.com/) 2. [Blog](https://www.langrelay.com/blog) 3. Is Google Translate accurate enough for a website? ArticleTranslation quality # Is Google Translate accurate enough for a website? A 2021 study found Google Translate kept the meaning of 55% to 94% of sentences, by language. What that means for a website, and which pages need review. [Chris Jayden](https://www.langrelay.com/about) Founder Published 6 October 2026 Reading time 7 min read ![Two people hand a basket of fruit between boats at a night river town, under lanterns patterned with green and orange leaf tiles.](https://www.langrelay.com/_emdash/api/media/file/01M4ET68Q6GKT31WM5RA995SHK.webp) On this page 1. What the published studies measured 2. What these numbers can and cannot tell a site owner 3. Where website text trips up machine translation 4. Labels without context 5. Names that are also ordinary words 6. Numbers, prices and dates 7. Form of address 8. Legal and regulated text 9. Does Google penalize machine-translated pages? 10. Which pages can go out as machine output 11. How LangRelay keeps names and register consistent 12. Common questions 13. Is Google Translate reliable for business use? 14. Which languages does Google Translate handle best? 15. Is DeepL or ChatGPT more accurate than Google Translate? Google Translate gets the meaning across most of the time in widely spoken languages, and noticeably less often in others. A 2021 study found it kept the meaning of 55% to 94% of sentences depending on the target language, and an older one saw an even wider spread. On a website, the mistakes that hurt most are in short, high-stakes text: a price, a product name, a button label, a line in your terms. Machine output is often good enough for some of your pages and not for others, and you need to know which is which before anything goes live. ## What the published studies measured Accuracy is hard to state as one number, because it depends on the language pair, the kind of text and who judges it. The most useful public evidence comes from medicine, where researchers had a reason to test short, consequential sentences with native speakers. In 2014, Sumant Patil and Patrick Davies took ten common statements a doctor might say to a family and ran them through Google Translate into 26 languages. Native speakers translated each result back into English. Of the 260 translations, 150 (57.7%) were correct. Portuguese did best at 90% and Swahili worst at 10%. One of their examples shows how a fluent sentence can still be wrong: "Your wife is stable" came back as "Your wife cannot fall over". Google's engine has changed since then, and a later team said so. In 2021 Breena Taira and colleagues noted that the 2014 study came before an improvement to Google Translate's algorithm, then ran their own test. They gave 20 emergency department discharge instructions to native speakers of seven languages. The meaning survived in 330 of 400 translations, or 82.5%. By language, Spanish reached 94%, Tagalog 90%, Korean 82.5%, Chinese 81.7%, Vietnamese 77.5%, Farsi 67.5% and Armenian 55%. Their conclusion was that Google Translate "is inconsistent between languages and should not be relied on for patient instructions." A 2024 study by Pavithra Rao, Lauren McGee and Casey Seideman compared Google Translate with ChatGPT on post-operative instructions, judged by professional translators. Google Translate got 18.1% of Spanish sentences wrong, 10.6% of Vietnamese sentences and 41.6% of Russian ones. The spread between languages was again larger than any single average suggests. (The ChatGPT side of that study is in [our article on ChatGPT for website translation](https://www.langrelay.com/blog/chatgpt-website-translation).) ### What these numbers can and cannot tell a site owner All three studies used medical instructions and small samples in a handful of languages. None of them tested a product page or a checkout. They do agree on two things that carry over to websites. The target language changes the result a great deal, and a sentence can read smoothly while saying something different from the source. If your site goes into Spanish, your odds look better than if it goes into Armenian, though in both cases someone still has to check what the page now says. ## Where website text trips up machine translation Most of a website is easier to translate than a medical instruction. Some of it is harder, because it is short or tied to your business in ways a general engine cannot know. ### Labels without context A button that says "Book" could be a verb (reserve a table) or a noun. In Spanish that is the difference between "Reservar" and "Libro". A menu item or a form label gives the engine one or two words and no sentence around them, so it guesses. ### Names that are also ordinary words If your plan is called "Growth" or your product line is "Harbor", a translation engine has no reason to leave those words alone. It may translate them on one page and keep them on another, depending on the sentence. ### Numbers, prices and dates A changed figure, a lost negation ("no setup fee" becoming "setup fee") or a date written in the wrong order reads naturally to anyone who cannot compare it to the source, and it can cost you money. ### Form of address German, Spanish, French, Italian, Dutch and Turkish all distinguish a formal and an informal "you". A shop that says "du" on its home page and "Sie" in its checkout reads as two different companies. A general engine can switch between them unless something tells it which to use. ### Legal and regulated text Terms of service, privacy notices and anything a regulator reads need a qualified translator. Machine output can tell you roughly what a clause says, but it should not be what you publish. ## Does Google penalize machine-translated pages? Google's spam policies were last updated on 2026-08-28. Under scaled content abuse they list "Scraping feeds, search results, or other content to generate many pages (including through automated transformations like synonymizing, translating, or other obfuscation techniques), where little value is provided to users." The concern there is mass-produced pages with little value. The policy does not call a translated page spam. It names translation as one of the methods used to produce such pages at scale, which is a reason to have a person look at what you publish. Google's guidance on multilingual sites adds two practical points. It recommends that you "use different URLs for each language version of a page rather than using cookies or browser settings to adjust the content language on the page." And it warns that if you dynamically change content based on language settings, "Google might not find and crawl all your variations." A script that rewrites the page in the visitor's browser is one way to end up in that second case. Our [multilingual SEO page](https://www.langrelay.com/multilingual-seo) covers the URL side in more detail. ## Which pages can go out as machine output Sort your pages by what an error on them would cost you. An old blog post with a stiff sentence costs almost nothing, so machine output with a quick read for meaning is usually enough there. A pricing page with a wrong number costs real money, so every figure, plan name and button gets checked on the page itself. Terms and privacy pages sit at the far end and go to a qualified translator. How much reading the middle group takes depends on your volume, and that reading is post-editing. Our guide to [machine translation post-editing](https://www.langrelay.com/blog/machine-translation-post-editing) goes through what to check, and [machine vs human translation](https://www.langrelay.com/blog/machine-vs-human-translation) covers when to pay for a translator from the start. ## How LangRelay keeps names and register consistent LangRelay uses AI translation and gives it two kinds of instruction that a general-purpose tool does not have for your site. The first is the Formality register. For each target language that has one (German, Spanish, French, Italian, Dutch and Turkish), you choose Language default, Formal or Informal in the Site's language settings. Every machine translation opened after that is produced in that register. Text translated earlier keeps its register until you ask for a new machine translation of it, so a change never rewrites approved pages behind your back. The second is the [Glossary](https://www.langrelay.com/blog/translation-glossary), on paid plans. For each term you set a required translation per language, or Keep as source for names like "Harbor" that must stay as written, and you choose whether case matters. Machine translation follows those terms from then on. Neither of those replaces reading the result. By default a page goes live as soon as its translation is ready, and free checks hold a page with untranslated text, the wrong language or a changed price; to approve every page yourself, choose Ask me first. You can read and edit the translation on the rendered page in the visual editor, at desktop, tablet and mobile widths, which is where a mistranslated button shows up. The [translation quality page](https://www.langrelay.com/translation-quality) describes the controls together. If you want to see them on your own site, you can [book a demo](https://www.langrelay.com/book-a-demo). ## Common questions ### Is Google Translate reliable for business use? For understanding a message, often yes. For publishing on a company website, treat it as a first draft. The 2021 study found accuracy ranging from 55% to 94% across seven languages on short, clear sentences, and business pages carry names and prices that need a person to confirm them. ### Which languages does Google Translate handle best? In the studies above, Western European languages scored highest: Portuguese at 90% in the 2014 study and Spanish at 94% in the 2021 study. Swahili (10%) and Armenian (55%) scored lowest. Those are small samples of medical text, so use them as a sign of the spread rather than a ranking. ### Is DeepL or ChatGPT more accurate than Google Translate? It depends on the language. The 2024 comparison above tested ChatGPT 3.5: it was well ahead in Spanish, Google Translate was ahead in Vietnamese, and the authors judged both weak in Russian. That study did not include DeepL, and we have not run a comparison of our own, so we make no claim about which engine is best. ## Sources The pages this one cites, with the day each was read. 1. 01[Patil and Davies, Use of Google Translate in medical communication: evaluation of accuracy (BMJ, 2014)](https://europepmc.org/article/PMC/PMC4266233)europepmc.org, checked 6 October 2026 2. 02[Taira, Kreger, Orue and Diamond, A Pragmatic Assessment of Google Translate for Emergency Department Instructions (Journal of General Internal Medicine, 2021)](https://escholarship.org/uc/item/520761c6)escholarship.org, checked 6 October 2026 3. 03[Rao, McGee and Seideman, A comparative assessment of ChatGPT vs. Google Translate for the translation of patient instructions (Journal of Medical Artificial Intelligence, 2024)](https://jmai.amegroups.org/article/view/9019/html)jmai.amegroups.org, checked 6 October 2026 4. 04[Google Search Central, Spam policies for Google web search (last updated 2026-08-28)](https://developers.google.com/search/docs/essentials/spam-policies)developers.google.com, checked 6 October 2026 5. 05[Google Search Central, Managing multi-regional and multilingual sites (last updated 2025-12-10)](https://developers.google.com/search/docs/specialty/international/managing-multi-regional-sites)developers.google.com, checked 6 October 2026 [Chris Jayden](https://www.langrelay.com/about) Founder, LangRelay Chris Jayden is the founder of LangRelay. Chris has worked on websites of every size, from large e-commerce stores to small sites with a handful of pages, and started LangRelay after years of translating them the hard way. On this page 1. What the published studies measured 2. What these numbers can and cannot tell a site owner 3. Where website text trips up machine translation 4. Labels without context 5. Names that are also ordinary words 6. Numbers, prices and dates 7. Form of address 8. Legal and regulated text 9. Does Google penalize machine-translated pages? 10. Which pages can go out as machine output 11. How LangRelay keeps names and register consistent 12. Common questions 13. Is Google Translate reliable for business use? 14. Which languages does Google Translate handle best? 15. Is DeepL or ChatGPT more accurate than Google Translate? Share - [Share on LinkedIn (opens in a new tab)](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.langrelay.com%2Fblog%2Fis-google-translate-accurate) - [Share on X (opens in a new tab)](https://x.com/intent/post?url=https%3A%2F%2Fwww.langrelay.com%2Fblog%2Fis-google-translate-accurate&text=Is%20Google%20Translate%20accurate%20enough%20for%20a%20website%3F) See it on your site A 30-minute walkthrough on one of your pages. [Book a demo](https://www.langrelay.com/book-a-demo) - Google Translate - machine translation - translation accuracy Book a demo ## See LangRelay on your own site Book 30 minutes with the team. We translate one of your pages live and walk through review, publishing and multilingual SEO. [Book a demo](https://www.langrelay.com/book-a-demo)[See pricing](https://www.langrelay.com/pricing)