On this page
Wire handles multi-language sites as a first-class feature. Not a plugin. Not a third-party translation layer. Built into the content pipeline, the build system, and the quality gates. Explore Wire's content capabilities. Write in your primary language. Wire translates, validates, and refuses to build if the translation quality drops.
The Translation Paradox
Multi-language websites face a contradiction. Weglot's study of 1.3 million AI Overview citations across 236 websites found that translated sites gain 327% more visibility in Google AI Overviews compared to untranslated sites, with a 24% increase in total citations per prompt. Translation also acts as an authority signal: 33% more English citations and 16% more Spanish citations per query, even in the site's original language. The evidence is clear that translation pays off.
But translation alone is no longer enough. Search Engine Land reports that AI systems now perform "semantic collapse," representing content as numerical vectors that encode meaning rather than language-specific text. Two pages with identical information in different languages normalize into near-identical representations. The most confident version wins globally, regardless of language. Translated replicas without genuine local differentiation are increasingly treated as redundant.
The resolution: translation remains a necessary foundation, but 2026 multi-language hubs must go beyond word-for-word conversion. German users search "Projektmanagement-Software," not literal translations of English terms. Market-specific differentiation (local pricing, regulatory content, region-specific examples, localized keyword targeting) separates hubs that rank from translated clones that get collapsed.
The Cost of Not Translating
Sites that skip translation do not just miss out on visibility. They actively cede traffic to competitors they cannot control. Google's translate.google.com proxy pulls in 377 million monthly visits from untranslated websites (Ahrefs, June 2025). A major Spanish book retailer selling English titles saw 64% less visibility in AI results. When it did appear, 36% of citations pointed to Google Translate's proxy page instead of the retailer's own domain.
As Weglot CMO Eugene Ernoult noted: "The problem grows because untranslated sites never build authority in unserved languages. As a result, they have little user engagement, which reduces trust signals that Google values."
Translation management systems like Lokalise, Phrase, and Crowdin cost $25-300/month and solve the wrong problem. They manage translation strings, not content quality. Your translated pages can be grammatically perfect and still say nothing useful because the source content was thin. The same content depth problem affects any company homepage or product directory that adds languages without adding substance.
Why Build-Time Enforcement Matters
Crowdin's 2026 enterprise survey of 152 B2B professionals found that 95% already use AI or machine translation. Yet 20.4% report quality incidents since adoption, 79.6% consider glossary enforcement mandatory, and 68.4% require automated QA checks. Multi-provider setups (47.4%) now outnumber single-provider (32.2%), and 74.3% require localization team approval before AI translations reach production. The survey captured a shift in industry framing: "Enterprise teams aren't looking for the best AI model. They're looking for a way to govern multiple models, control what data gets exposed, and ensure consistent output across every language and every team. That's a platform problem, not a model problem."
Most tools respond to these quality concerns with warnings. Wire responds with BUILD REFUSED. The difference matters. A 2023 Intento survey found that raw machine translation output still requires significant human post-editing for publication quality. One anonymous enterprise respondent in the Crowdin survey put it bluntly: "AI struggles in remembering even short chapters and starts hallucinating quite quickly... results in language models for not-widely-spoken languages still require a lot of corrections." When your build system warns but does not block, those warnings get ignored. When it refuses to produce output, quality problems get fixed before publication.
How Wire Fits
Wire's multi-language pipeline addresses both the translation paradox and the governance gap.
The translate command translates page bodies using Claude, not Google Translate. Per-language styleguides control voice and terminology. This produces differentiated content, not semantic duplicates that AI systems collapse. Language asymmetry detection enforces parity: if a translation has less than 20% of the primary language's page count, BUILD REFUSED. No ghost-town translations. This also prevents "freshness drift," where one language updates faster than others and AI systems silently prefer that version globally.
Stopword detection (Gate 14) checks that pages in /de/ actually contain German text and pages in /en/ contain English text. Wrong language in wrong directory triggers BUILD REFUSED. Per-language styleguides go further: each language gets its own _styleguide.md for voice calibration. German technical writing differs from English technical writing in sentence structure, formality level, and compound noun conventions. Wire encodes that difference into the prompt system.
A unified GSC database (single .wire/gsc.db with a lang column) tracks search performance per language, so each market competes on its own terms. Wire validates reciprocal hreflang links from alternate frontmatter at build time. 75% of international sites have hreflang errors: missing self-references, asymmetric annotations, invalid language codes, redirecting URLs in clusters, and canonical conflicts. Wire catches all of these before deployment.
What to Customize
wire.yml
languages:
- code: en
name: English
docs_dir: docs/en
- code: de
name: Deutsch
docs_dir: docs/de
Per-Language Styleguides
Create docs/en/_styleguide.md and docs/de/_styleguide.md with language-specific editorial guidance. German readers expect different sentence structure, formality level, and technical terminology than English readers. Wire's prompt system injects the right styleguide per language automatically.
Content Commands
All commands require --lang on multi-language sites:
python -m wire.chief refine vendors --lang de
python -m wire.chief reword vendors --lang en
Exception: data runs all languages automatically.
Scope
Wire generates per-language sitemaps, hreflang tags from alternate frontmatter, language-specific RSS feeds, a unified search index, automatic language prefix in URLs (/en/, /de/), and cross-language link validation (RULE-69). These outputs require no configuration beyond the languages block in wire.yml.
Wire translates markdown content. It does not localize images, videos, or embedded media. Currency, date formats, and number formatting are not handled by Wire. These are CSS/JavaScript concerns for the template layer.
To get started, run python -m wire.migrate-lang to convert a single-language site to multi-language format. Configure languages in wire.yml with language codes and docs directories, then run python -m wire.chief translate --lang de to translate content. Run python -m wire.build to build all languages and verify quality. Check for RULE-69 warnings on cross-language link issues. For detailed architecture and configuration, see the multi-language guide. See all Wire use cases for more patterns.