bengarves.com/work/localization-and-translation
Localization and Translation
Project One-Pager: Language Subdirectory Technical SEO Updates
From Ben Garves’s work at Udemy. Re-created from the original document, with people’s names, internal links and non-public numbers removed or generalized.
Project brief
- Company
- Udemy
- Format
- Project one-pager
- Product
- Ben Garves
- Teams
- Product, Engineering, SEO/Marketing
- Prioritization
- RICE
- Scope
- 18 language subdirectories
Proposes connecting Udemy’s translated and localized pages across 18 language subdirectories to their default pages with hreflang clusters, so search engines send international learners to the page in their own language without hurting existing rankings.
- Localization
- hreflang
- International SEO
- Technical SEO
Background
Udemy serves learners globally across 18 language subdirectories (ar, de, ro, ja, id, th, pl, es, ru, ko, it, pt, tr, fr, nl, vi, zh-tw, zh-cn), providing localized versions of our course catalog and category pages. Currently, we have significant international content that is invisible to search engines:
- Noindexed “good” localized pages: Tens of thousands of high-value localized pages across 18 languages are currently marked with
noindex,followdirectives. These are being indexed as a part of the Crawl Efficiency Cleanup work. - Noindexed “poor” localized paginated pages: More than a hundred thousand paginated versions of language subdirectory pages are marked
noindex,follow. These are being indexed as a part of the Crawl Efficiency Cleanup work.
Hreflang usage
To follow best practices and benefit from language clusters, pages that are translated or localized versions of default pages should be grouped using hreflang references in their <head>.
Who is the user?
- Primary users
- International learners searching for online courses in their native language (Portuguese, Spanish, Japanese, Korean, etc.)
- Local market users who prefer or require content in their language for discovery and decision-making
- Search engine crawlers attempting to discover and index international content
- Users with characteristics
- Language preference: Users searching in non-English languages on Google, Bing, Yandex, Baidu, etc.
- Geographic location: Users in Brazil, Portugal, Spain, Mexico, Japan, South Korea, Germany, France, and other international markets
- Search intent: Looking for courses, skills, or learning paths in their native language
Why is solving these problems big enough to matter?
Udemy currently has tens of thousands of translated pages, spanning 18 languages and thousands of root “default” pages. By connecting these with proper hreflang use, we’ll create thousands of distinct clusters that allow crawlers to understand the depth and quality of our translated content and localized courses, resulting in greater referral rates from organic traffic sources directly to a user’s preferred language page, where they’re also more likely to convert.
What is the hypothesis?
By clustering localized pages with their default, we will increase organic visibility, traffic, and conversions from international search markets without negatively impacting site quality, crawl budget, or existing rankings.
What are the success criteria?
- Increase in total number of indexed language pages
- Increase in organic sessions to language version pages
- Increase in conversion rates on default pages
- Increase in conversion rates on language version pages
- Net-neutral organic session count on default pages
What are we building?
The root page, or “default,” should refer to itself as the “x-default.” For example:
html
<link rel="alternate" href="http://example.com/" hreflang="x-default" />The root page should refer to all language versions using hreflang and each version page’s language. For example:
html
<link rel="alternate" href="http://example.com/es" hreflang="es" />The language versions of pages should refer to their full cluster of the x-default and other language versions. For example:
html
<link rel="alternate" href="http://example.com/" hreflang="x-default" />
<link rel="alternate" href="http://example.com/es" hreflang="es" />
<link rel="alternate" href="http://example.com/pt" hreflang="pt" />Related links
What are the risks?
| Risk | Severity | Likelihood | Mitigation |
|---|---|---|---|
| Cannibalization of English page rankings | High | Low | Monitor English page performance throughout test; proper hreflang should prevent this; Search Console will show if Google confuses language versions; roll back immediately if detected. |
| Poor translation quality affects user experience | Medium | Medium | Conduct content quality audit before launch; have localization team review flagged pages; monitor engagement metrics and bounce rates; iterate on poor content post-launch. |
| Technical implementation errors | High | Low | Thorough QA in staging; code review; sample testing in production; phased rollout allows catching errors early; feature flags for easy rollback. |
| Hreflang conflicts or errors | Medium | Low | Validate hreflang implementation before launch; use hreflang testing tools; Search Console reports on hreflang errors; fix immediately if detected. |
| User confusion from search results | Low | Low | Hreflang should serve correct language to users; Google’s language detection should work properly; monitor bounce rates from organic for quality signals. |
Cannibalization of English page rankings
- Severity
- High
- Likelihood
- Low
- Mitigation
- Monitor English page performance throughout test; proper hreflang should prevent this; Search Console will show if Google confuses language versions; roll back immediately if detected.
Poor translation quality affects user experience
- Severity
- Medium
- Likelihood
- Medium
- Mitigation
- Conduct content quality audit before launch; have localization team review flagged pages; monitor engagement metrics and bounce rates; iterate on poor content post-launch.
Technical implementation errors
- Severity
- High
- Likelihood
- Low
- Mitigation
- Thorough QA in staging; code review; sample testing in production; phased rollout allows catching errors early; feature flags for easy rollback.
Hreflang conflicts or errors
- Severity
- Medium
- Likelihood
- Low
- Mitigation
- Validate hreflang implementation before launch; use hreflang testing tools; Search Console reports on hreflang errors; fix immediately if detected.
User confusion from search results
- Severity
- Low
- Likelihood
- Low
- Mitigation
- Hreflang should serve correct language to users; Google’s language detection should work properly; monitor bounce rates from organic for quality signals.
How will we go to market with this launch?
Because language clusters are only powerful in their aggregate, this work will be completed and rolled out in full or at the speed and scope recommended by our engineering partners based on the changes needed to such a large volume of pages.