A foreign-language site with Google Translate: when is it enough?

A foreign-language version produced with Google Translate is enough to convey the rough meaning of a text, but it does not produce the words a buyer searching in that language uses, the form of address they expect or the language of trust. Google’s translation plugin for websites is also no longer supported from 1 October 2026.

The search “Google Translate website” asks about two different things. The first is the plugin added to a page that translates the text instantly when the visitor chooses a language. The second is preparing text with machine translation and publishing it as separate language pages. The two lead to different results; for an exporter planning a B2B website, the answer to both is below. Let us start with the plugin, because for it the date is now fixed.

Can the Google Translate plugin still be used?

In 2020 Google reopened the plugin called Website Translator only for public, non-profit and non-commercial sites focused on COVID-19, and recommended the Cloud Translation API for all other sites. According to a note added to the same announcement on 1 September 2026, the plugin will no longer be supported from 1 October 2026; site owners who want continuous automatic translation are directed to the Google Cloud API. For an exporter still offering a language option on its site through this plugin, this is a decision date.

The plugin already had a weakness on the search side: the translation is produced in the visitor’s browser after the page opens, and the translated text has no address of its own. Google, by contrast, recommends giving each language version a separate address and not switching the content’s language based on cookies or browser settings. A German text without its own address cannot appear as a result in a German search.

Is publishing separate language pages with machine translation a problem?

Technically it is possible, but there are two risks. The first is Google’s spam policy: the “scaled content abuse” heading cites, as an example, producing many pages that add little value for users through automated transformations such as synonymizing or translating content gathered from elsewhere. Translating your own text is not exactly this description; but hundreds of machine-translated pages published without review open the door to the same problem — little value.

The second, and the more expensive one for an exporter, is the reader’s trust. A faulty German product page tells the buyer how much care the company takes over detail. For a purchaser looking for quality when choosing a supplier, language is also a quality signal.

Why does translation miss the keyword?

Translation preserves the meaning of the sentence, not the query. The search the buyer types in their own language may not be the dictionary equivalent of the Turkish term: a product’s official name and its market name can differ, and something said in two words in Turkish may be searched for as a single compound word in German. Machine translation does not know which one is established in that market. We explained this difference in detail in the guides keyword research by language and translation or localization.

How were this site’s three languages written?

This site is published in Turkish, English and German, and the English and German texts are written not as translations of Turkish sentences but with the search patterns of each language. The address segments are also separate in the three languages: the Turkish, English and German addresses of the service pages carry the term searched for in that language. Because the German text is written for a corporate buyer, it is built throughout on the formal “Sie”; a German B2B page written with “du” speaks in the wrong tone even if its grammar is correct.

A test protects this too. The site’s test suite measures whether the FAQ headings on the English and German service pages contain that language’s own purchasing pattern — “how much” or “cost” in English, “was kostet” in German — on at least two pages. Translated strings are also marked in the file with a translation note and sent to the site owner for approval: at any moment it is known which text has been verified by a human.

German also has a visual cost: German text is noticeably longer than Turkish and English, and headings can overflow. On this site the width axis of the heading typeface is narrowed for German, so that the same heading stays on the same number of lines; details in the guide German text expansion.

When is Google Translate enough?

There are places where it is enough, and hiding them would not be honest:

• Internal use — to understand an incoming foreign-language email or a technical data sheet.
• First draft — as a starting text to be corrected by a translator or by an employee who speaks the language natively.
• Market testing — to see whether a few pages in a language attract interest before investing heavily in it; provided they are reviewed and published at their own address.
• Language-independent data — content whose translation causes no debate, such as units, standard numbers and measurement tables.

In which details does machine translation go wrong?

Formal errors come before errors of meaning, and these are the most glaring on a B2B page:

• Form of address — in German a corporate buyer is addressed with “Sie”; machine translation may mix both forms of address on the same page.
• Product names — a product name known in the industry by its own name becomes unrecognizable when translated into its dictionary equivalent; product names should be marked “do not translate” in the term list.
• Numbers and units — the decimal separator is a comma in Turkish and German and a point in English; in a weight or tolerance value this is not a spelling error but a value error.
• Dates and delivery terms — the order of day and month varies from language to language; misreading a delivery date is a direct commercial dispute.
• Interface text — buttons, error messages and form labels are short, so they are translated without context, and this is where wrong meanings appear most often.

Where it is not enough is everywhere the buyer makes a decision: product descriptions, certificates and quality declarations, delivery terms, the quote form and error messages. In these texts a translation error is not an error of meaning but a commercial error.

What is the order for using machine translation safely?

• Research in that language first — find out which word the buyer searches with before translating.
• Build a term list — approved equivalents for product names, technical terms and the company’s fixed phrases.
• Use machine translation as a draft — the text goes to correction, not to publication.
• Have a native speaker check it — especially the form of address and product names.
• Give each language its own address and declare the language set completely — hreflang errors arise most often here.
• Set up a maintenance routine — when the Turkish page changes, how will the other languages find out?

If a right-to-left language such as Arabic is being opened, the list grows by one item: the direction of the page changes too. One of the seven languages of Karcem, the knitted fabric site we manage, is Arabic. When we opened the Arabic version of the fabric page on 23 September 2026, the page was at its own address, the last segment of the address was in Arabic and the document declared a right-to-left direction. A plugin’s instant translation does not produce all three of these. The full skeleton of a foreign-language site is in the guide what an exporter’s website should look like.

For a site that presents your products to foreign buyers in their own language, B2B website builds and measures the path to product, certificate and quote in the target market’s language.