How to use AI to translate an Umbraco website

Can AI translate an Umbraco website?

Yes. Modern language models can produce fluent, natural translations that are ready for everyday website use with little or no editing, depending on the language pair and source material. Translating a CMS also means handling more than a paragraph of text: an Umbraco page may contain rich text, nested Block List or Block Grid items, links, media selections, settings and identifiers. A safe tool must translate the human-readable text while leaving the protected structure intact.

For general business content, you can use AI to do the bulk of the translation work and bring new languages to your site much faster. Omni keeps those translations inside Umbraco's normal variant, review and publishing process, so editors stay in control.

Why use an LLM instead of Google Translate or Microsoft Translator?

A modern large language model (LLM) can interpret the meaning, tone and intent of the text it receives and follow instructions about how to translate it. You can explain the audience, ask for a particular style, and specify how brand names or industry terms should be handled. That flexibility is especially useful for website copy, where a phrase needs to sound natural to a customer as well as convey the original meaning.

For example, “Get up and running” on a software page is about getting started. An LLM can use the supplied context and instructions to choose an appropriate expression in the target language. This makes it a strong fit for service pages, product descriptions and marketing content where tone and intent matter.

Google Translate and Microsoft Translator are useful translation services, and their advanced offerings also support context and customisation. The advantage of an LLM-based Umbraco workflow is the flexibility to guide the translation through a chat profile and glossary, then apply it directly to supported pages and blocks. With Omni, you can translate a whole section, preserve the CMS structure and review the resulting language variants in the backoffice, without copying text between tools.

Three ways to use AI translation

What should an Umbraco AI translation tool handle?

• Language variants: create or update the correct culture without overwriting the source.
• Structured content: extract text from supported blocks and rich text, then rebuild the value without changing keys, layout or settings.
• Selection: let editors see and choose the pages and properties included before spending provider tokens.
• Drafts: leave publication to an authorised editor.
• Terminology: apply glossary rules for product names and preferred phrases.
• Governance: restrict which provider profiles and prompts editors can use.
• Operations: show progress, individual failures and a durable job history.
• Maintenance: identify source text that has changed since the last translation.

Which AI provider should you use?

The right provider depends on language quality, regional availability, data-processing requirements, model cost and organisational policy. Umbraco.AI provides a common integration layer for supported chat providers, while profiles control the model and instructions used for a task.

Test representative content rather than choosing from a generic benchmark. Include short labels, long-form copy, rich text, brand terminology, ambiguous phrases and the languages that matter to the real project.

A safe AI translation workflow

1. Configure languages and variant properties in Umbraco.
2. Install and test an approved AI provider and translation profile.
3. Preview the text that will be sent and exclude unsuitable fields.
4. Start with one representative page before translating a section.
5. Save generated values as a draft.
6. Give everyday content an editorial check for meaning, tone and presentation; use specialist review for legal, medical or regulated claims.
7. Publish through the normal Umbraco workflow.
8. Track later source changes and retranslate only what needs attention.

What should not be translated automatically?

Identifiers, URLs, keys, UDIs, dates, numbers, colours, picker selections and arbitrary configuration should normally remain unchanged. Sensitive personal data, confidential material and regulated claims may also require a different provider policy or an entirely human workflow.

Is AI translation good enough to publish?

For most everyday website use cases, modern AI translation can be good enough to publish with little or no editing when you choose a capable model for your languages. Service pages, general product information, blog posts and FAQs are all strong candidates. The benefit is substantial: AI handles the translation work, leaving your team to focus on the content and markets that matter.

A human review is still preferred, just as it is for content written in your original language. For routine pages, that can be a light editorial check rather than a full translation exercise. Content with specialised legal terminology, medical advice or regulated claims deserves a qualified reviewer. Omni saves translations as drafts so you can apply the level of review that suits each page and publish when you are ready.