Localizing keywords: why translation is not enough

When a business expands into a new language, the usual first step is to translate the website, and the usual second step is to translate the keyword list. The result reads correctly and ranks poorly. Translation captures what words mean; it does not capture which words people in a market actually type, what they expect to find, or which competitors already own the results. Localisation means researching each market as if it were your first.

Why direct translation misses

Problem	Example
Different everyday term	"Sneakers" (US) vs "trainers" (UK); "mobile phone" vs "cell phone"; "Handy" in German for a mobile phone
Translated term nobody uses	A literal translation of "standing desk" may be grammatical but less searched than a borrowed English term or a different local phrase
English loanwords	In many markets, technical and marketing searches use English words ("SEO", "laptop", "coworking") rather than the local equivalent
Different modifiers	Price-focused modifiers in one market, quality or brand modifiers in another
Different intent	A query that shows guides in one country may show shops in another
Different seasons and events	Summer sales in June in one hemisphere, December in the other; holidays that do not exist elsewhere
Different units and formats	Sizes, measurements, date formats, currency

A hypothetical maker of merino clothing translating "base layer" into German, French and Dutch will find that each market has its own mix of native terms, English borrowings and brand names, and that the most searched phrase is not always the dictionary translation. Only data from each market shows which is which.

A process for each market

1. Start from the customer, locally. Ask local sales staff, distributors or customer service what customers call the product and what they ask. Read local forums, reviews and competitor sites in that language.
2. Seed with translations, then expand with local data. Use translated keywords only as seeds. Expand them with autocomplete, People Also Ask and keyword tools set to the target country and language.
3. Check the results pages from the target country. Search each important term with the right location and language settings. Note which page types rank and which competitors appear; they may be completely different from your home market. The search intent guide covers what to look for.
4. Group and prioritise per market, as you would at home; see the keyword research process. Volume, competition and business value can rank topics in a different order than at home.
5. Map keywords to pages per version. Each localised page gets its own primary keyword from local research, not a translated one.

A worked example

A hypothetical online shop selling standing desks in the UK plans a German version. The translated keyword list leads with the most literal German equivalent of "standing desk". Local research tells a more complicated story. Keyword tools set to Germany show that a compound native term and the phrase for a height-adjustable desk are both searched heavily, and that the English loanword appears too, mostly among office buyers. The results page for the native term shows shops and comparison articles; the results for the height-adjustable phrase lean towards shops with detailed specifications and delivery information. German competitors emphasise motor quality, noise levels and warranty length far more than the British ones, which lead with price and colour options.

The German category page therefore targets the native term as the primary keyword, covers the height-adjustable phrasing naturally in the copy, puts the specification table and warranty near the top, and lists delivery times to German postcodes. A German-speaking freelancer writes the buying guide from a brief rather than translating the English one, and swaps the British examples for ones that make sense to German buyers. None of that would have come out of a translation tool.

Common mistakes

• Choosing one term for all markets that share a language, such as the same Spanish keyword for Spain and Mexico.
• Translating the brand's own jargon, which nobody searches for in any language.
• Forgetting the results page: a keyword can be right and the page type wrong for that market.
• Localising the page but not the metadata, leaving English or literal titles in search results.
• Never revisiting: the first research for a new market is the least informed. Revisit after three to six months of real data.

Working with native speakers

Native speakers are the most valuable part of localisation, and they work best with a clear brief:

• The page's purpose and audience in that market.
• The primary keyword and a few related terms from local research, with the instruction to use them naturally.
• Permission to change examples, references and structure where the original does not fit.
• Local facts that must be checked: prices, regulations, delivery times, legal text.
• A request to flag anything that sounds foreign or unnatural.

Machine translation followed by native review can be efficient for long, factual content. For key landing pages, product categories and anything persuasive, writing or rewriting in the target language usually produces better results than editing a translation.

Adapting content, not just words

Localisation often changes more than vocabulary:

• Examples and references: local brands, places, holidays and units.
• Proof: reviews and case studies from customers in that market carry more weight than translated ones from elsewhere.
• Practical details: currency, taxes, delivery, returns, payment methods people expect, local customer service hours.
• Regulatory differences: product claims allowed in one country may not be in another.
• Tone: the right level of formality varies by language and audience.

Some pages may not need to exist in every market at all, and some markets need pages the original site never had, such as a guide to a local regulation or a comparison with a local competitor.

Titles, URLs and metadata

• Titles and meta descriptions written in the target language from local keywords, not translated from the original. Character counts differ by language; German titles, for example, often run longer. See the title guide.
• URL slugs in the local language (transliterated to plain characters if needed) help users recognise the page; keep them stable once chosen.
• Image alt text translated and adapted too.
• hreflang linking equivalent pages; see hreflang implementation.

Measuring each market separately

Set up a Search Console property (or a filtered view) for each language version, and compare markets on their own terms. A new language version typically takes months to build visibility, especially if it has few local links. Watch for:

• Queries the pages actually appear for, compared with the planned keywords. Unexpected queries often reveal the real local vocabulary.
• Pages with impressions but low CTR, which often signals titles that read as translations.
• Queries in the wrong language appearing for a version, a sign of hreflang or content problems.

Tracking average position per query in each market, which DeskSEO does once each property is connected, shows whether each version is building visibility or stalling.

Local AI assistants and search behaviour

People in different markets use search engines and AI assistants in different proportions, and they phrase questions to assistants in their own language with local context. A brand that is well described in English sources may be invisible when someone asks the same question in Spanish or Polish. Checking AI visibility per language, with questions written by native speakers as customers would ask them, is part of localisation now; see measuring AI visibility.

A minimum standard for launching a market

1. Local keyword research for the top 20–50 pages, with results pages checked from the target country.
2. Those pages written or reviewed by native speakers with a brief.
3. Local practical details: prices, delivery, contact, legal text.
4. Titles and descriptions written from local keywords.
5. hreflang and canonical tags correct for every version.
6. A measurement plan per market, reviewed monthly for the first six months.

Launching fewer pages that meet this standard beats launching a complete machine translation that does not.