AI Translations

How Docsbook automatically translates your documentation using Claude AI.

How It Works#

You enable a language in Translation Settings.
Docsbook fetches your markdown files from GitHub.
Each file is sent to Claude (by Anthropic) for translation.
Translated pages are cached for fast delivery.
When you push updates to GitHub, only changed pages are re-translated.

No manual work. No translation files to maintain. No YAML keys.

What Gets Translated#

Translated	Not translated
Body text	Code blocks
Headings	Inline code (`like this`)
Tables	URLs and links
Image alt text	File paths
Sidebar navigation labels	Technical identifiers
Page titles and meta descriptions	—

Code is intentionally left in the original language — it should stay consistent regardless of the reader's locale.

Translation Cache#

Translations are cached per page per language, keyed to the file's content hash on GitHub.

First visit after a content change triggers re-translation (a few seconds).
All subsequent visits serve the cached version instantly.
Cached translations persist even if you temporarily disable a language.

This means re-enabling a language that was previously active is instant — no re-translation needed.

Translation Quality#

Claude produces high-quality technical documentation translations — the same model that powers Claude.ai, trained on a large corpus of technical writing in dozens of languages.

Why Claude Outperforms Generic Translation#

Generic machine translation (Google Translate, DeepL) is built for general-purpose text. Developer documentation has a different structure: imperative commands, technical terms, code-adjacent prose, and precision-critical instructions where a mistranslation causes a broken setup.

Claude understands context. It knows that "run the command" means ejecuta el comando, not corre el comando, and that a parameter name should stay in English even when the surrounding sentence is in French. It preserves meaning across sentence restructuring — not just word-for-word substitution.

Works best for:

Step-by-step guides
API documentation
FAQs
Configuration references

May need manual review for:

Brand-specific terminology
Cultural references or idioms
Highly domain-specific jargon

Quality at Scale#

Human translators cost $0.10–$0.30 per word. A 50-page documentation site is typically 25,000–40,000 words. That's $2,500–$12,000 per language — before you account for updates.

With Docsbook, translation is included in your plan. Every update to your docs is re-translated automatically for only the changed pages. No invoices. No turnaround time. No coordination overhead.

SEO Impact of Translations#

Every translated language version is a fully separate set of URLs, indexed independently by Google.

docsbook.io/{user}/{repo}        → ranks for English queries
docsbook.io/{user}/{repo}/es     → ranks for Spanish queries
docsbook.io/{user}/{repo}/de     → ranks for German queries

Docsbook automatically adds hreflang tags to every page, which tells Google which language version to show to which audience. Without these tags, Google may show the wrong language to international visitors — or ignore translated pages entirely.

What this means in practice: enabling Spanish translation doesn't just serve your existing Spanish-speaking users better. It creates an entirely new set of pages that rank for Spanish-language searches your English docs never would have appeared in. You're multiplying your search surface area by the number of languages you publish.

A documentation site that was ranking for 500 English queries now ranks for 500 queries in each additional language — passively, without any extra content work.

Writing Tips for Better AI Translation#

Clear source content produces better translations. Follow these guidelines:

Use short, direct sentences. Long compound sentences lose nuance in translation.
Avoid idioms. "It's a piece of cake" doesn't translate literally. Say "It's simple."
Use ISO date format. Write 2024-03-25, not "March 25th" — dates are interpreted differently across locales.
Keep code comments in English. They're excluded from translation but should be readable by all developers.
Use consistent terminology. If you call something a "workspace" in one place, don't call it a "project" in another.

Go global without the translation overhead. Enable translations →