How to Translate E-Learning Courses: AI vs Agency (2026)

Translating an e-learning course through a traditional agency involves exporting text files, briefing a translation team, waiting two to six weeks, receiving translated files, and then rebuilding the course in your authoring tool - often manually, section by section. The cost per language is typically £500 to £3,000 depending on course length and language pair. AI course translation produces a complete, independent, publish-ready course in a new language in minutes, from within the editor, without exporting any files or using any external tool. CourseAgent supports 19 languages and translates every learner-facing element - content, quiz questions, answer options, feedback text, and course metadata - in a single action.

Why e-learning translation is harder than document translation

Document translation is a solved problem. Give a translator a Word file, get a translated Word file back. E-learning translation is significantly more complex because the content is not a linear document - it is structured into topics, pages, sections, and section types, each with its own field structure. A slideshow has multiple slides, each with a title and body. A flip card has a front and a back. An accordion has panels. A quiz question has a stem, multiple answer options, correct answer mappings, and individual feedback strings for each option.

Traditional translation workflows handle this by exporting to XLIFF or a custom spreadsheet format, sending to the translator, and then importing back and manually checking that every string is in the right place and that quiz logic still works. Even with good tooling, this process is error-prone, time-consuming, and requires the authoring tool to be reopened, edited, and re-published for every language.

The result is that most organisations either do not translate their e-learning at all (defaulting to English even for non-English-speaking audiences), translate only the highest-volume courses (leaving the rest in the source language), or commission agency translations that take months and cost more than the original course build.

The translation that never gets commissioned is the training gap nobody talks about. English-only content in a multilingual workforce is not a cost saving - it is a liability.

What AI course translation actually covers

The distinction between interface localisation and content translation matters here. Interface localisation - translating the buttons, labels, and navigation of the learning platform - is relatively straightforward and most LMS platforms offer it. Content translation - translating the actual course material - is the hard part, and most platforms do not do it at all.

CourseAgent's AI translation covers every learner-facing element of course content:

• All section text - body content, bullets, image captions
• All interactive section content - slideshow cards, accordion panels, flip card fronts and backs, tab panels, timeline cards and node labels, quote text and attributions, hotspot labels and descriptions, progressive card content
• All quiz content - question stems, every answer option, correct and incorrect feedback text, with correct answer mappings preserved
• All course metadata - title, description, introduction, topic titles, and topic descriptions

What is preserved unchanged: course structure, section types, page layouts, images, colour schemes, and all formatting. Only the text changes - so the translated course looks and feels identical to the original, with exactly the same interactive elements in exactly the same places.

The 19 supported languages

• Common: English (UK), English (US)
• Western European: Spanish, French, German, Italian, Portuguese, Dutch
• Nordic: Swedish, Norwegian Bokmål, Danish, Finnish
• Central & Eastern European: Polish, Czech, Hungarian, Romanian
• Other: Hindi, Chinese Simplified, Welsh

Each translated course is a fully independent course with its own version history, its own analytics, and its own publication workflow. It can be deployed to a dedicated academy portal, published as a URL, or exported as a SCORM package - all independently of the source course.

Which of the 19 languages should you prioritise?

Picking the first few languages to translate is not about counting native speakers globally - it is about where corporate e-learning demand is concentrated and where your workforce or buyers actually sit. Based on the markets where AI course authoring searches consistently grow outside English-speaking countries, this is a reasonable starting order for most UK and US-based teams:

The remaining supported languages - Czech, Hungarian, Romanian, Welsh, English US/UK variants - are typically translated in response to specific contracts rather than as a default rollout. If you do not know which to pick first: Spanish, French and German cover roughly 80% of multilingual e-learning demand for European-headquartered organisations. For US-headquartered firms, lead with Spanish (LATAM and domestic Hispanic workforce), then Portuguese (Brazil), then French (Canada plus EU).

Traditional agency translation vs AI translation

SCORM translation and the live-update caveat

For teams pushing courses into an LMS, translation lives or dies on SCORM. Static SCORM export is the same problem multiplied by the number of languages: every language becomes its own zip file, uploaded separately, versioned separately, and updated separately. Any content correction has to be applied to the source, retranslated, re-exported, and re-uploaded for every language - and in most LMSs, replacing a SCORM package resets learner progress on that course.

Dynamic SCORM changes the shape of the problem. Because the SCORM package points to a hosted course rather than embedding the content, a translated version updates in place on the LMS: fix a typo in the source, retranslate the affected sections, and the LMS reflects the change without a fresh upload. The trade-off is worth being honest about - live in-progress learners on any course (any dynamic SCORM implementation, ours or a competitor's) can have their progress affected if a substantive structural change is pushed mid-course. For copy corrections and additions this is a non-issue; for restructuring the course you should sequence updates the way you would sequence any live release.

See Dynamic SCORM explained for a full breakdown of how the mechanic works. The short version: dynamic SCORM plus AI translation collapses the "publish once in 19 languages, keep them all current" problem from a multi-week agency project to a single editor action.

When AI translation is appropriate - and when it isn't

AI translation produces high-quality output for the vast majority of corporate e-learning content: policy and procedure courses, compliance training, onboarding programmes, skills development, and product knowledge. The output is fluent, grammatically correct, and contextually appropriate - substantially better than machine translation tools like DeepL or Google Translate because it receives the full structural context of the course alongside the text.

There are scenarios where human review or human translation is still warranted:

• Regulated content with precise legal terminology. A data protection course translated into French for a Belgian financial services firm should be reviewed by someone who knows GDPR terminology in French regulatory contexts - AI will produce accurate general French, but may not use the precise terminology a regulator would use.
• Content with organisation-specific terminology or product names. Brand names, product terms, and internal processes that have established translations in the target organisation should be verified rather than accepted as-is.
• Languages with significant regional variation. Spanish covers Castilian, Latin American, and regional variants. Chinese Simplified is appropriate for mainland China audiences but not for Taiwan. For audiences where regional variation matters, a native reviewer adds value.
• High-stakes safety or medical content. Where mistranslation creates physical risk, human expert review of AI output is appropriate regardless of AI quality.

For most corporate e-learning content in standard European languages, AI translation produces output that is ready to publish with a quick proofread by a bilingual team member - not a full human translation review.

An alternative to translation: native-language generation

AI translation is not the only approach to multilingual content in CourseAgent. If you know in advance that a course needs to be available in multiple languages, you can generate each language version natively - using the AI course authoring feature with the target language selected from the start. Native generation tends to produce more idiomatic output than translation, because the AI writes directly in the target language rather than converting from English.

The practical question is which approach fits your workflow. If you have an existing English course that needs to be available in additional languages, translate it. If you are starting from a brief and know the languages in advance, generate each version natively. Both approaches produce fully independent, publish-ready courses with the same structural fidelity.

A realistic cost comparison

Assume a typical 45-minute compliance course with 40 sections and a 15-question assessment. Agency translation into six languages (Spanish, French, German, Italian, Portuguese, Dutch) at a mid-market rate averages £8,000 to £14,000 and takes eight to twelve weeks end-to-end once you include briefing, translation, review, and rebuild in the authoring tool. The same six-language rollout in CourseAgent completes in the same session as the source course build and costs a small fraction of the agency figure - with dynamic SCORM meaning that any subsequent correction propagates to all six versions without re-upload.

The economic implication is straightforward: teams that could previously afford to translate one or two flagship courses can now translate the whole library at similar total cost, and keep every version current instead of letting non-English versions drift out of date.

Frequently asked questions

Do you support right-to-left languages like Arabic or Hebrew? Not currently. The 19 supported languages are all left-to-right. Arabic and Hebrew are on the roadmap but require full RTL layout support in the learner runtime, which is a distinct piece of work from translation itself.

Can I edit the AI translation before publishing? Yes. Every translated section is fully editable in the same editor as the source. A common workflow is: translate the whole course in one action, then have a bilingual reviewer walk through the terminology-sensitive sections and adjust as needed.

Does the assessment still grade correctly in the translated version? Yes. Correct-answer mapping is preserved as part of the translation, so a learner selecting the translated equivalent of the correct option receives the correct outcome. Quiz feedback text is translated per option, so incorrect-answer explanations remain specific.

Can I translate a course that was uploaded from PowerPoint or a document? Yes. Once the course exists in the CourseAgent editor - whether generated by AI, built manually, or imported from PowerPoint, Word, or PDF - translation works the same way.

What happens to updates after translation? Updates to the source course do not automatically flow to translations - you retranslate the changed sections. In practice this is a single action per language and takes seconds, because you are retranslating only what changed rather than the whole course.

The short version

E-learning translation used to be a slow, expensive, error-prone process that most organisations avoided or rationed. AI translation in CourseAgent collapses it into a single editor action per language, preserves every interactive element and quiz mapping, supports 19 languages, and integrates with dynamic SCORM so updates propagate without re-upload. For most corporate e-learning content in standard European languages, the AI output is publish-ready with a light proofread. For regulated, high-stakes, or brand-sensitive content, treat AI translation as a strong first draft that a bilingual reviewer refines.

Ready to see it? Start a free trial or read how CourseAgent builds courses end-to-end.