FAQ

What does this tool do?

The Website Localization Reviewer analyzes translated web pages and scores them across four dimensions: Linguistic Quality, Terminology Consistency, Cultural Adaptation, and Completeness. It generates an instant quality report including a prioritized list of specific flagged issues, without requiring the reviewer to read the target language. It can also analyze the same page across multiple languages simultaneously and export results as a text report or CSV.

Who is it for?

Content groups, localization managers, and QA reviewers who need to assess the quality of a translated website — as a post-publication audit or a pre-launch check.

How do I use it?

Paste the URL of a translated page and click Analyze. For a deeper review, also provide the source URL — the tool will compare them directly, catching omissions and mistranslations that are invisible from the target alone. Use the Multi-language tab to analyze the same page in several languages at once. Or click one of the demo buttons to see it in action immediately.

What are the three analysis modes?

• AI Judgment — Holistic scoring across all four dimensions, drawing on Claude's broad knowledge of the target language and localization conventions.
• MQM Framework — The industry-standard Multidimensional Quality Metrics framework. Errors are classified as Critical (−25), Major (−5), or Minor (−1), and the score is calculated from the total penalty.
• Comparison — Provide both a source and target URL. The tool compares them directly, flagging omissions, additions, and semantic shifts invisible from the target alone.

What is the Flagged Issues list?

Below the dimension score cards, the tool lists every specific problem it found across all four dimensions, ranked by severity — Critical first, then Major, then Minor. Each entry includes a quoted excerpt from the page and a one-sentence explanation of why it is an error. This list can be copied directly into a correction brief for a language vendor.

Can I analyze multiple languages at once?

Yes — use the Multi-language tab. Enter a URL and target language for each language version of the same page, then click Analyze All. Click 'Load EU Commission example' to see it pre-filled with five languages. Results appear in a comparison table as each language completes. Click any row to see its detailed flagged issues.

Can I export the results?

Yes. On the Analyzer tab, click '↓ Export report (.txt)' at the bottom of the results for a full text report including scores, explanations, and all flagged issues. On the Multi-language tab, use '↓ Export full report (.txt)' for a complete per-language breakdown, or '↓ Export scores (.csv)' for the scores table in spreadsheet format. Files are named with an ISO date and serial number for easy filing.

What sites work well?

Any site that serves text in the HTML response: government sites, news sites, corporate pages, documentation. The EU Commission (ec.europa.eu) is an ideal test case — 24 languages, clean HTML, and professional translations.

What sites won't work?

JavaScript-heavy sites that render content client-side (IKEA, Booking.com, many e-commerce platforms). The tool will return an error or very low score if it cannot extract enough text from the page.

Why does the score vary slightly between runs?

AI Judgment and Comparison modes involve qualitative assessment, which has inherent variability. MQM mode is more consistent because it follows a fixed error-classification and penalty formula.

Further development ideas

• API access — A REST API so the reviewer can be integrated into CI/CD pipelines, CMS workflows, or translation management systems for automated quality gates.
• MQM benchmark validation — Cross-reference AI-generated MQM scores against human-annotated gold standard datasets (such as the MQM Translation Quality dataset on HuggingFace) to measure and improve scoring accuracy.
• Rule-based grammar checking — LanguageTool integration for a fully deterministic Linguistic Quality score.
• Paste text mode — Analyze content that isn't publicly accessible — staging environments, protected pages, or raw translation output.
• Multi-page crawl — Analyze a full site section rather than a single page.
• Translation memory sync — Locate flagged strings in existing TM/TMX files for targeted correction.