OnlineDocTranslatorGuides

Running your practice

Website Localization: What Translators and Agencies Actually Do on the Project

Website localization is the end-to-end adaptation of a site's content, UI, formats, and SEO for a target locale, performed by translators using CAT tools, TMs

Website localization sits inside the broader discipline covered in our guide to How App Localization Works and What Clients Are Buying, but on the production floor it looks different: the translator opens a resource bundle, not a design file, and ships back strings a developer can merge without reformatting a single line. This page is written from that chair — the translator's — not the buyer's.

What is website localization from the translator's side of the desk?

Website localization is the adaptation of a site's strings, UI, media, number and date formats, and SEO metadata for a specific locale. Translators receive extracted resource files, work in CAT tools against translation memory, then return locale-stamped files for the dev team to merge. The source HTML stays in the repo.

The producer's view breaks the work into five artifacts:

  • Resource bundle — the extracted strings the developer hands over.
  • Translation memory (TM) — prior approved segments reused across projects.
  • Term base — the brand's locked terminology, per locale.
  • Style guide — tone, register, and forbidden phrases.
  • Staging URL — the in-context build the reviewer opens.

How does website localization differ from website translation?

Website translation converts source strings into target-language strings. Website localization also adapts dates, currencies, measurement units, imagery, legal copy, color meaning, address fields, and SEO metadata so the page reads as native rather than translated.

DimensionTranslationLocalization
ScopeSource text → target textText, UI, formats, SEO, legal, media
DeliverableTranslated stringsLocale-stamped resource files
QALinguistic reviewLinguistic plus locale and in-context UI
Dev dependencyLowHigh — requires i18n-ready source
Billed byWordWeighted word plus engineering time

Why is it abbreviated L10n and paired with i18n?

Localization becomes L10n because 10 letters sit between the L and the n; internationalization becomes i18n for the same reason. L10n is the translator's work on each locale; i18n is the engineering work that makes those locales possible.

The i18n vs l10n split matters for scoping:

  • i18n is a one-time engineering task: externalize strings, isolate locale-sensitive logic, wire up ICU MessageFormat, add hreflang scaffolding.
  • L10n repeats per locale: translate, review, QA, sign off.
  • The handoff artifact between them is the resource bundle — JSON, XLIFF, PO, YAML, .strings, or .resx.

What does a translator receive at the start of a website localization project?

A translator receives extracted resource files plus a style guide, term base, translation memory, screenshots or a staging URL, and a locale brief. The six resource formats seen most often are JSON, XLIFF, gettext PO, YAML, .strings, and .resx. The source HTML itself rarely crosses the handoff.

Which resource file formats appear most often on web projects?

JSON and XLIFF 2.1 dominate modern web stacks, followed by gettext PO on PHP sites, YAML on Rails, .strings on iOS web-wrapper apps, and .resx on .NET. Each format carries different pluralization and placeholder conventions.

FormatTypical stackCAT tools that open it cleanlyPitfall for the translator
JSON (flat or nested)React, Vue, Next.js — see React i18n: How Translators Fit Into a React Localization PipelineTrados Studio, memoQ, Phrase, SmartcatICU placeholders break if quotes are escaped wrong
XLIFF 2.1Any TMS-exported bilingualTrados, memoQ, OmegaT, PhraseInline tags must be preserved byte-for-byte
gettext POWordPress, Drupal, DjangoPoedit, OmegaT, memoQmsgid_plural needs separate msgstr[0..n] per language
YAMLRuby on RailsPhrase, Lokalise, SmartcatIndentation errors kill the whole file
.stringsiOS web-wrapper, macOSXcode, memoQ, PhraseUnescaped percent signs crash the format call
.resxASP.NET, .NET MAUITrados, memoQ, Multilingual App ToolkitComment field holds context that must not be translated

Some stacks hide the file layer entirely. Our guide to How Webflow Localization Works and What Translators Need to Know covers that contrast: no file ever reaches the translator — the TMS calls the site's API instead.

What does a usable localization brief include?

A usable brief lists target locale, tone register, forbidden terms, brand voice reference, staging URL, character limits per key, and milestone deadlines — not just an EN-to-ES language pair.

  • Locale code, not just language — es-MX differs from es-ES in currency, phrasing, and formality.
  • Character limits per key, especially for buttons, nav, and tooltips.
  • Forbidden terms, including competitor names and legally risky phrasing.
  • Reference URLs for existing localized pages the client treats as benchmark.
  • Milestones for first draft, in-context review, and final sign-off.

How it works

What are the key steps a translator performs during website localization?

1

How does translation memory leverage change the pricing math?

Translation memory leverage drops effective rates on repeat and high-fuzzy segments to roughly 25%–75% of the full word rate, so a 10,000-word site with 40% TM leverage bills closer to 7,000 weighted words at the full rate.

Match bandTypical rate (% of full)
Repetitions25%
100% match25%–30%
95%–99% fuzzy50%
85%–94% fuzzy60%–70%
75%–84% fuzzy75%
No match100%

2

What does in-context review catch that offline translation misses?

In-context review catches truncated buttons, overlapped RTL icons, misplaced decimal separators, concatenation bugs, and legal copy that overflows a modal — issues invisible when strings are translated row-by-row in a spreadsheet.

  • German "Benachrichtigungen" blowing out a 14-character nav slot.
  • Arabic layout flipping icons but leaving a chevron pointing the wrong way.
  • A dynamic sentence like "You have {count} {item}" producing ungrammatical output in Russian.
  • A currency symbol glued to the wrong side of the number for fr-FR (should be "19,99 €", not "€19.99").

What locale rules does the translator have to apply beyond language?

The translator applies locale rules for date order, decimal and thousands separators, currency placement, measurement units, phone and postal formats, address field order, and 24-hour versus 12-hour clock display — all distinct from the language choice itself.

Fielden-US (origin)de-DEfr-FRja-JPar-SA
DateMM/DD/YYYYDD.MM.YYYYDD/MM/YYYYYYYY/MM/DDDD/MM/YYYY
Decimal separator1,234.561.234,561 234,561,234.56١٬٢٣٤٫٥٦
Currency$19.9919,99 €19,99 €¥1,99919.99 ر.س
Clock12h24h24h24h12h
Address orderStreet → City → State → ZIPStreet → ZIP → CityStreet → ZIP → CityZIP → Prefecture → City → StreetStreet → District → City → ZIP
Phone+1 (555) 123-4567+49 30 1234567+33 1 23 45 67 89+81 3-1234-5678+966 50 123 4567
Unitsmiles, °F, lbkm, °C, kgkm, °C, kgkm, °C, kgkm, °C, kg

RTL languages — Arabic, Hebrew, Persian, Urdu — add a layout mirror on top of the locale fields: directional icons flip, number sequences reverse, and progress indicators run right to left. The translator flags each one for engineering.

How are plurals handled with ICU MessageFormat?

ICU MessageFormat expresses plurals as zero, one, two, few, many, other categories. Russian uses one, few, many; Arabic uses all six; English uses only one and other. The translator fills the categories the target language requires, not the ones the English source had.

A single English segment like "{count, plural, one {# item} other {# items}}" expands to three Russian cases and six Arabic cases. Skipping a category breaks the render at runtime.

How much does translated text expand or contract?

Translated text expansion ranges from +15% for Spanish and French to +30% for German and roughly +40% for Finnish; Chinese and Japanese typically contract by 20%–50%. UI character limits set against English copy routinely break for Germanic and Finno-Ugric targets.

Target languageExpansion vs English
Spanish, French, Italian, Portuguese+15% to +20%
German, Dutch+30%
Finnish, Hungarian+40%
Russian, Polish+10% to +20%
Arabic+25% (plus RTL mirror)
Chinese (Simplified), Japanese−20% to −50%

Before real translation starts, pseudo-localization replaces source characters with accented variants and padding — "Săvé Čhángēs!!" — surfacing UI breakage, hardcoded strings, and font-coverage gaps while the English copy is still legible to the developer.

How it works

How does continuous localization change the translator's workflow?

1

Where does machine translation post-editing fit in?

Machine translation post-editing fits between TM pre-translate and human translation. The engine fills no-match segments, the translator edits the output to publishable quality, and the corrected segments feed back into the TM so leverage compounds over the project's life.

  • Light post-edit — fix grammar and critical mistranslations; accept engine style.
  • Full post-edit — bring output to the same quality bar as from-scratch human translation.
  • Rate math typically runs 50%–70% of the full word rate for full post-edit, 30%–40% for light.

2

How do AI-orchestrated localization platforms change what the translator actually does?

AI-orchestrated platforms now route, pre-translate, and quality-check strings automatically. Translators must know where human judgment overrides the pipeline — brand voice, cultural nuance, legal precision — and how to configure AI thresholds without eroding quality.

  • Set a confidence threshold below which segments route to a human regardless of match score.
  • Lock terminology at the term-base level so the engine cannot override brand terms.
  • Flag any segment carrying regulated copy (medical, financial, legal) for mandatory human review.

What does the translator hand back to the client?

The translator hands back localized resource files in the identical structure and key order as the source, an updated translation memory, a term base delta, a QA report flagging unresolved placeholders or UI overflow, and release notes identifying strings that still need engineering review.

A complete handback package:

  • Target-locale files, byte-identical in structure to source (no renamed keys, no reordered arrays).
  • Updated TMX export of the translation memory.
  • Term base delta — new entries added during the project.
  • QA report with every warning categorized: placeholder, length, format, terminology.
  • Release notes listing strings that need engineering follow-up before go-live.

Where does multilingual SEO sit in the translator's responsibilities?

Multilingual SEO sits at the metadata layer: the translator localizes title tags, meta descriptions, OG tags, image alt text, and URL slugs where the stack permits, then flags hreflang pairs and canonical mappings for the dev team to implement. Keyword research and URL architecture are not the translator's job.

The scope line is clean:

  • Translator owns: title tag, meta description, OG title and description, Twitter card copy, image alt text, slug translation.
  • Translator flags: hreflang pairs (en-US ↔ es-MX ↔ de-DE), canonical URL mapping, language-switcher labels.
  • Dev or SEO owns: URL structure (subdomain vs subdirectory vs ccTLD), server-side locale detection, keyword strategy, structured data markup.

Delivered this way, a website localization project leaves the translator's desk as a merge-ready set of files — strings resolved, locales respected, SEO metadata localized, QA documented — and lets the dev team ship the locale on the next build with no translation debt attached.

Frequently asked questions

What is the meaning of localization in translation?

Localization means adapting content for a specific target locale — not just translating words, but reworking cultural references, date formats, currencies, images, and tone so the result feels native to that market. Translation converts text from one language to another; localization goes further by aligning the entire user experience with local expectations. For a website, this can include rewriting idioms, resizing UI elements for text expansion (German and Finnish text can run 30–40% longer than English), and ensuring legal or compliance copy meets local standards.

What is the difference between localization and translation?

Translation converts meaning from one language to another, while localization adapts the full experience — cultural context, visuals, layout, and functionality — for a specific market. A translator working on a website localization project does more than render sentences: they flag culturally inappropriate imagery, adapt measurement units, restructure date formats (MM/DD/YYYY vs. DD/MM/YYYY), and adjust formality registers. Agencies typically layer translation, editing, and locale-specific review (TEP: translate, edit, proof) to deliver a fully localized product rather than a literal word-for-word conversion.

Can you give me an example of localization in translation?

A clear example is localizing a U.S. e-commerce site for the German market: the translator doesn't just convert English to German, but replaces dollar signs with euros, rewrites the returns policy to comply with EU consumer law, adjusts the formal 'Sie' register throughout, and flags any product imagery that may carry different cultural connotations. Another common example is adapting a software UI where button labels must stay within strict character limits — a phrase that fits in 12 characters in English may expand to 18 in Spanish, requiring both linguistic and UI adjustments.

What are the different types of localization?

The main types of localization are software/app localization, website localization, document/content localization, game localization, and multimedia localization (subtitling and dubbing). Each type carries different technical requirements: software localization involves handling string files and placeholders; website localization requires working with CMS platforms and considering SEO in the target language; game localization often demands creative adaptation of humor and culturally specific references. Agencies and translators may specialize in one or more of these types, and each demands a different combination of linguistic, technical, and cultural skills.

What is website localization?

Website localization is the process of adapting every layer of a website — text, images, navigation, SEO metadata, and functionality — so it feels purpose-built for a target locale rather than translated from somewhere else. For translators and agencies, this means working with exported content from a CMS or a CAT-tool-integrated connector, translating UI strings and marketing copy, adapting hreflang tags and meta descriptions for local search, and coordinating with developers on text expansion issues. A fully localized site typically requires multiple rounds of linguistic QA plus in-browser review to catch layout breaks.

What are the best practices for website localization?

Best practices for website localization include internationalizing the source content before translation begins, using a translation memory to ensure consistency across all pages, localizing SEO elements (title tags, meta descriptions, alt text) for each target language, and performing in-context linguistic review inside the live or staging site. Agencies should establish a style guide and glossary per language pair before the first segment is translated — this alone can cut revision cycles significantly. Text expansion allowances (typically 20–40% over English) should be built into UI design, and pseudo-localization testing helps catch hard-coded strings early.

What are the best localization tools for translation?

The most widely used localization tools among professional translators and agencies are CAT tools such as SDL Trados Studio, memoQ, and Phrase (formerly Memsource), combined with TMS platforms like Phrase TMS, Lokalise, or Crowdin for workflow management. These tools provide translation memory, terminology management, and CMS integrations that automate file export and import. For agencies producing certified translations at scale, adding an AI first-draft layer — then applying human post-editing and certification — can reduce per-word turnaround time substantially while keeping a licensed translator in the review chain.

Can AI do localization?

AI can produce strong first-draft translations for localization, but human translators and reviewers are still required to handle cultural adaptation, certified accuracy, and context-dependent decisions. Modern neural machine translation (NMT) engines perform well on high-resource language pairs and formulaic content like UI strings, but they routinely miss register nuance, locale-specific legal requirements, and culturally loaded imagery. For certified translations specifically, a qualified human translator must review, certify, and take professional responsibility for the output — AI-generated drafts accelerate that workflow but don't replace it.

What do translation agencies do?

Translation agencies manage the full workflow of a localization project: sourcing and vetting qualified translators, assigning language pairs, running TEP (translation, editing, proofreading) pipelines, handling file engineering, and delivering finished assets to the client. On a website localization project, an agency typically handles CMS integrations, coordinates with developers on text expansion and encoding issues, manages translation memories and glossaries, and performs final QA before handoff. Larger agencies also provide project management, DTP (desktop publishing), and certified translation services that require a credentialed translator's signature and statement of accuracy.

Is AI replacing translator jobs?

AI is reshaping translator roles rather than eliminating them — the dominant shift is from full translation to post-editing of machine translation (MTPE), which requires the same linguistic expertise but different workflow habits. Demand for certified translation, legal and medical translation, and culturally sensitive localization remains human-dependent because those outputs carry professional and legal accountability that AI cannot assume. Agencies increasingly use AI to generate first drafts and handle high-volume, repetitive content, which compresses per-word rates on commodity work but opens capacity for higher-value specialized projects.

Why is localization called L10n?

L10n is a numeronym — the 10 replaces the ten letters between the 'L' and the 'n' in 'localization,' following the same pattern as i18n for internationalization and a11y for accessibility. These abbreviations emerged in software development communities where long, frequently typed terms were shortened for efficiency in documentation, issue trackers, and commit messages. The convention is now standard across the translation and localization industry, and translators working on tech or software projects will encounter L10n, i18n, and g11n (globalization) routinely in client briefs and TMS platforms.

Related