Translation Quality

Translation Quality is the third layer of your Cassian Score™, weighted at 20%. It is only active when your store has translations enabled. If your store is monolingual, this layer is excluded entirely and the other four layers are reweighted to total 100%.

Cassian AI™ compares your translated content against the source language and evaluates it for completeness, accuracy, and naturalness. A bad translation doesn't just hurt your score — it loses sales in markets you've already paid to reach.

---

What Cassian AI evaluates

Machine translation artifacts

Automated translation tools have become remarkably capable, but they still produce characteristic errors that a native speaker will notice immediately. Cassian AI identifies phrases and sentence structures that appear to be unedited machine translation output.

Common machine translation artifacts Cassian looks for:

• Word-for-word literal translations that produce grammatically odd sentences
• Idiomatic expressions in the source language rendered literally rather than translated culturally
• Terminology that doesn't match industry norms in the target language (a word that's technically correct but not what that market actually uses)
• Inconsistent tone or register — content that alternates between formal and informal forms of address
• Brand names or product names that were left untranslated when a localised version exists

This check is about the quality of the translation as a reading experience, not just whether words are present. Content that passes a grammar check can still read as obviously machine-translated.

Missing translations

Cassian identifies pages, product descriptions, and metafields where translated content is entirely absent. A customer browsing in French who lands on an English-only product page — or worse, a page that's partially translated — is very likely to leave.

What's checked for completeness:

• Product titles and descriptions per language
• Collection page titles and descriptions
• Page content (FAQ pages, About pages, shipping and returns information)
• Navigation items and footer links
• Metafields used for additional product information
• Alt text on product images

Variant translation completeness

Product variants — size, colour, material, style — are one of the most commonly missed translation targets in Shopify stores. Customers selecting a variant see the variant name and often the variant-specific description. If those aren't translated, the checkout experience breaks down in the target language.

Cassian checks:

• Variant option names (e.g., "Size", "Colour") translated in each active language
• Variant option values (e.g., "Small", "Medium", "Red", "Blue") in each language
• Variant-specific descriptions or metafields that differ from the main product description
• Variant image alt text in translated languages

This is typically the largest category of translation issues in stores that use Shopify's translation tools, because variant translations are managed separately from product-level translations and are easy to overlook.

RTL rendering issues

For right-to-left languages — including Arabic, Hebrew, Persian (Farsi), and Urdu — Cassian checks that the page layout actually renders correctly in RTL context. This includes:

• Content alignment (text should align right)
• Layout mirroring (navigation, product images, and content blocks should mirror horizontally)
• Punctuation placement (quotation marks, parentheses)
• Number and date formatting

RTL rendering is typically a theme-level configuration. Shopify's Dawn theme and most major marketplace themes support RTL, but custom themes may need explicit RTL CSS.

Metafield translation accuracy

Shopify uses metafields to store structured product data — ingredients, care instructions, technical specifications, certifications, and more. These metafields are increasingly used in product pages and are increasingly important for AI-powered product searches.

Cassian compares translated metafields against the source language content:

• Are all metafields present in each translated language?
• Do the translated values correspond to the source content (or are they blank, incomplete, or pointing to source-language content)?
• Are translated metafields complete in character count relative to the source (very short translated versions of long source metafields are flagged)?

---

Setting up translations for Cassian to evaluate

For Cassian to evaluate your translations, you need to tell it that your store is multilingual.

If you're connected via Shopify: language detection is automatic. Cassian reads your Shopify store's active markets and locale configuration. You can still override the detected languages in Site Settings.

If you use a third-party translation app (Shopify Translate & Adapt, Weglot, Langify, etc.): Cassian detects translations from the rendered page content and hreflang tags. No special configuration is needed — the standard language settings are sufficient.

---

Supported languages

Cassian supports 26+ languages for translation evaluation. Coverage varies:

• Full analysis (prose quality, machine translation artifact detection, completeness): Major languages including English, Spanish, French, German, Italian, Portuguese, Dutch, Swedish, Norwegian, Danish, Finnish, Polish, Czech, Romanian, Hungarian, Greek.
• RTL language support: Arabic, Hebrew, Persian, Urdu.
• Structural analysis (completeness, missing translations, variant coverage): All configured languages, including less common ones where full prose analysis isn't available.

If your store uses a language not in the list above, completeness checks will still run — Cassian will flag missing content — but prose quality analysis may be limited.

---

Monolingual stores

If your store operates in a single language with no translations, Translation Quality is excluded from your Cassian Score calculation entirely. The remaining four layers are automatically reweighted to total 100%:

---

Where to see Translation Quality results

• Issues page → filter by "Translation" category. All translation issues listed by severity with affected language identified.
• Scans page → click any scan → Translation Quality section shows your layer score and all issues found.
• Dashboard overview → the Translation Quality card shows your current layer score. If translations are disabled, this card shows the layer as inactive.

---

Common questions

My store is in English only — why is Translation Quality showing in my score? Check Site Settings → Language. If translations are accidentally enabled, disable them. Cassian will exclude the layer from your next scan's score calculation.

Does Cassian support right-to-left languages? Yes — Arabic, Hebrew, Persian, and Urdu are supported for both translation quality analysis and RTL rendering checks.

How does Cassian know which language is the source? From your primary language setting in Site Settings. If you've connected Shopify, it's also read from your Shopify store's default market language. You can always override this manually in Site Settings.

How many languages can Cassian check simultaneously? All languages configured in your store's translation setup. There's no per-scan language limit. If your store is in 8 languages, all 8 are evaluated in a single scan.

My translation app generates translations automatically — will Cassian flag all of them as machine-translated? Not necessarily. If the automatic translations read naturally in the target language, Cassian won't flag them. The check looks for patterns that read as unedited machine output — stilted phrasing, literal idiom translations, terminology mismatches. Well-tuned automatic translations that have been reviewed and edited typically pass.

Translation Quality is active but I have zero issues — is that correct? Yes, it's possible. If all your translated content is present, complete, and reads naturally, Cassian will find nothing to flag. A Translation Quality score of 100 means no issues were found in that layer during the scan.

Cassian flagged a translation as a machine translation artifact but we wrote it manually. Machine translation artifact checks look for patterns that correlate with automatic translation. Occasionally, manually written translations that use literal constructions from the source language can trigger the same patterns. If the flagged content reads correctly to a native speaker, dismiss the issue from the Issues page.