Website Localisation: Why Translation Alone Does Not Deliver Results

The situation companies face today

As growth in the domestic market slows, more companies are looking to expand their sales channels abroad. The first move is usually website localisation: translating the existing site into English and adding a language switcher.

Yet open the metrics a few months later and a familiar pattern appears. Overseas visitors have increased but enquiries have not. Time on site is short, and most visitors leave from the first screen. The result is generally put down to insufficient brand recognition, which leads to adding one more language the following year.

In many cases the problem is not recognition. Translation is only one stage of localisation, and that stage was addressed while the rest was left untouched. Overseas customers must pass through three stages before they reach the point of making an enquiry, and drop-off occurs at each stage for a different reason.

What is lost at each of the three stages

Stage 1 — The site is not found in search

If translated pages do not appear in search results, the later stages do not exist. The common failures here are technical.

No distinct address per language. An approach that swaps only the on-screen text with JavaScript is recognised by search engines as a single page. The English page is not included in the search index at all. Each language needs its own address.

Language relationships are not declared. The Korean and English versions of the same content must state that they are alternative languages of one another. This information is conveyed with the hreflang attribute, and every page must declare cross-references to all language versions, including itself. Without this declaration, search engines either treat the two pages as duplicate content or serve visitors the wrong language.

Metadata is left untranslated. A surprising number of sites translate the body copy but leave the page title and description in Korean. What appears in search results is not the body but those two fields, so if they are in Korean, no click occurs.

Search terms are not translated. This is the most frequently missed point. Translating domestic terminology literally produces wording that diverges from what local customers actually search for. The translation is accurate, but nothing matches in search. This is why local search term research must precede translation.

Stage 2 — Visitors arrive but do not trust the site

What an overseas visitor who reaches the page judges within a few seconds is not the quality of the product but whether the company is real and whether they can do business with it. What obstructs that judgement is usually not translation quality but the absence of information.

• No contact method by local convention (telephone numbers without a country code, only domestic messaging apps)
• No information confirming the company exists (year of founding, location, size)
• Trading terms not stated (minimum order quantity, settlement currency, shipping destinations)
• No indication of response hours in local time

The stilted phrasing characteristic of machine translation also erodes trust, but the absence of the information above is more damaging. If the trading terms are clear, enquiries come even when the phrasing is slightly awkward. The reverse does not hold.

Stage 3 — The enquiry is made but the response breaks down

Even where a first enquiry arrives, one stage remains: time zones and language.

Overseas enquiries arrive outside domestic working hours. Reply the next morning and the sender is already talking to another supplier. And when the reply is written in a domestic employee's second language, each round trip adds another day. Most initial enquiries disappear during this process.

What is needed here is not a perfect response but an immediate one. Drop-off falls substantially even if only an acknowledgement of receipt and an expected reply time are sent automatically. Prepare automatic answers to frequently asked questions and the sender already has half the information they need before anyone comes into the office.

Where response is automated, always restrict the basis of answers to internal documents and provide a path that hands low-confidence enquiries to a person. The loss created by one incorrect answer is greater than the loss from responding late.

Keeping brand terminology consistent

The problem that arises most often in localisation and is found latest is inconsistent terminology: a product name translated differently from page to page, or the same technical term expressed differently in each language. Visitors read this not as "the translation is awkward" but as "this company is not well organised".

Raising translation quality does not solve the problem, because the judgement differs at each point of translation. The solution is to fix the judgement in advance, that is, to build a glossary. A glossary contains three kinds of entry.

Words not to translate. Proper nouns that must retain their original form, such as product names, service names, patented technology names and brand slogans. Without this list, product names are rendered by meaning in each language, and overseas customers can no longer find them in search.

Standard translations per language. For industry and technical terminology, fix a single official expression in each language. If one concept appears as a different word on each page, search visibility is diluted and trust suffers.

Expressions not to use. Manage as a prohibited terms list any words that do not fit the brand image, expressions with negative connotations locally, and terminology that evokes a competitor.

Maintaining the glossary matters more than creating it. Without a procedure for registering entries each time a new product launches, the glossary will be out of step with reality within six months. Assign an owner for registration and build term registration into the new product launch process.

What to check before adding languages

Before adding a new language, confirm the following for the languages already in operation. If any item is blocked here, adding a language will produce drop-off at the same point in the same way.

The last item is particularly important. Without separating metrics by language, there is no way to know at which stage the drop-off occurs. Looking only at total enquiry volume makes it easy to reach the mistaken conclusion that localisation had no effect.

What to measure at each stage

Pinpointing the drop-off requires a different metric at each stage. Record the values below separately for each language.

If an earlier stage is blocked, the figures for the later stages carry no meaning. Attempting to improve bounce rate while search impressions are near zero is the wrong order. Work through the stages in sequence, from the first.

Which language to add first

It is safer to base language selection on current evidence rather than market size.

• Check the country and language distribution of existing visitors. Language regions already arriving are the first priority.
• Check the countries that actual enquiries have come from. This is a stronger signal than traffic.
• Check the languages competitors support. This is evidence that a market has already formed.

Choose a language on market size alone and you end up paying maintenance costs to run a language version that generates neither traffic nor enquiries. Bear in mind that once a language is added, every subsequent content change costs more.

Not confusing language with country also matters. Countries that share a language still differ in currency, shipping terms and regulation. When covering several countries with a single language version, state at least the trading terms separately by country.

Reducing the operational burden

The real cost of a multilingual site arises not in the initial translation but in operation, because every language must be updated whenever a product is added or a price changes. Four principles reduce that burden.

Manage source and translation separately. If translated text is embedded in the interface code, development is required for every change. Manage the copy outside the code and translation updates no longer involve development.

Divide content by frequency of change. Content that barely changes, such as the company overview, is best translated carefully and fixed, while information that changes often, such as prices and stock, is more realistically handled by combining automatic translation with human review. Demand the same quality of all content and updates fall behind.

Put in place a way to detect missed updates. Unless the state in which the source has changed but the translation has not is flagged automatically, the language versions drift apart little by little. This inconsistency directly damages trust.

Connect the glossary to the translation process. A glossary that exists only as a document and is not applied at the point of translation has no effect. Whether translation is automated or outsourced, build a structure in which registered terms are applied automatically.

Apply different standards by content type

Translate all content the same way and the cost is incurred in full while updates fall behind. Divide content by type and set different standards for each.

In particular, terms of service and the privacy policy are subjects for localisation, not translation. Carry across a document drafted against domestic law and the sentences may read naturally while failing to meet local requirements. Once the target countries are settled, have these documents in particular reviewed separately.

Regulatory matters to confirm in advance

Once you actually begin receiving overseas visitors, requirements arise that a domestic site never faced. Responding after an incident costs far more, so confirm them in advance.

Consent for cookies and tracking tools. For visitors in Europe, tracking for analytics or advertising purposes cannot run without prior consent. Confirm that the consent banner genuinely blocks script execution before consent is given. Implementations that display the banner while the scripts already run are common.

Notice of personal data collection. If an enquiry form collects names and contact details, the purpose of collection, retention period and method of destruction must be stated in the local language.

Where the data is stored. Establish which country's servers hold visitor information and enquiry content. Depending on the region, transfer abroad requires separate notice or a legal basis.

Disclosure that responses are automated. If responses are automated, make it possible for the other party to tell that they are not dealing with a person. Regulations requiring this disclosure are increasing, and the disclosure itself does not reduce trust.

A 90-day sequence

Attempting to change everything at once generally means the work never starts. Proceed stage by stage, in order.

Days 1–30 — measurement and diagnosis. First reach a state in which traffic, drop-off and enquiries can be viewed separately by language. Then use the checklist above to assess the current language versions and pinpoint the stage that is blocked. Nothing new is built at this stage.

Days 31–60 — put the search stage in order. Sort out per-language addresses, hreflang declarations, and translation of titles and descriptions. In parallel, research local search terms for the main products and services and apply them to the copy. Search visibility takes time to appear after changes are made, so it is best to begin here.

Days 61–90 — the trust and response stages. Bring trading terms, contact methods and response hours into line with local convention. Prepare automatic acknowledgement of enquiries and answers to frequently asked questions. In the same period, draft the glossary and define the registration procedure.

The essential point of this sequence is to establish measurement first. Start with improvements and no measurement, and 90 days later there will still be no way to know what worked.

Summary

The results of localisation are determined by how many stages have been addressed rather than by translation quality.

• A distinct address per language and hreflang declarations are a precondition, not an option.
• Page titles and descriptions must be translated as well as body copy for clicks to occur in search results.
• Literal translation without local search term research does not match in search.
• The absence of trading terms and contact methods is more damaging than awkward phrasing.
• Fix proper nouns and technical terms in a glossary so that expressions do not vary by language.
• What closes the time difference is an immediate response, not a perfect one.
• Separate the stage metrics by language and work through the stages in sequence, from the first.

Before deciding to add one more language, establish which stage is blocked in the languages you already run.