It depends on the document type and how it will be used. For some applications, such as understanding the general content of a foreign-language document or early-stage drafting for internal use, AI translation tools can be genuinely useful. For regulatory submissions, Instructions for Use (IFU), labelling, clinical documentation, and anything reviewed by a notified body or competent authority, AI translation alone is not sufficient.

The core issue is not that AI translation is always wrong. It is that it is inconsistently right. In medical device translation, a single mistranslated term in a dosage instruction, a safety warning, or a contraindication can have serious consequences. The fluency of AI output makes errors harder to spot, not easier.

For full details of what our IFU translation services cover, and how we manage accuracy in regulated documentation, see our IVDR and MDR translation services page.

Yes, and we have done so for many years, but in a fundamentally different way from pasting text into a free online tool. We combine two technologies: Translation Memory Software (TMS) and Neural Machine Translation (NMT), with specialist human oversight at every stage. Our NMT is professional-grade, currently among the most capable available for medical and regulated content, providing intelligent translation suggestions that our specialist translators review, refine, and approve. Nothing leaves our process without expert human sign-off.

This combination of the best available AI technology with genuine domain expertise is what delivers the speed, consistency, and accuracy that regulated industries require. It is also what makes our approach categorically different from unmanaged machine translation.

Yes, this is possible, and we can accommodate it. However, the cost saving is likely to be considerably less than expected, and in some cases the approach creates more work rather than less.

The saving from submitting an AI-translated draft for professional post-editing, rather than commissioning a full translation, is roughly in the range of 30 to 50 per cent. That range should be treated as a guide only, not a guarantee. The upper end requires a high-quality AI translation as a starting point, ideally from professional-grade tools, covering content that is not highly complex or terminology-heavy. For medical device documentation, the saving is more often towards the lower end of that range. Several factors determine what, if anything, you save:

• The quality of the AI translation submitted. If the source content is complex, highly technical, or contains specialised regulatory terminology, AI tools frequently mistranslate in ways that are not obvious. A translator reviewing a poor AI draft may need to retranslate large sections, reducing the saving substantially.
• The nature of the errors. AI translation errors in medical content are often subtle: plausible-sounding but technically incorrect. Identifying and correcting them can take as long as translating from scratch, because the reviewer must verify meaning at every step rather than working from the source text directly.
• The loss of your terminology database. Working outside our TMS process means the translation does not contribute to your client database, and you do not benefit from the accumulated savings that build with each project we handle for you.

Our honest view: the managed combination of NMT and TMS that we already apply to every project is likely to deliver better value than the AI-plus-proofread route, with greater consistency and without the compliance risk.

The following types of errors occur regularly in AI-translated medical device content:

• Terminology inconsistency. AI tools select from multiple plausible translations for a single technical term, often varying the choice across a document. In an IFU, using three different terms for the same component creates ambiguity that can confuse users and raise questions at regulatory review.
• False cognates and near-miss translations. Words that look similar across languages but carry different or subtly different regulatory meanings are a consistent source of error. AI tools produce translations that are grammatically correct but technically wrong.
• Loss of regulatory register. MDR and IVDR documentation requires specific regulatory language that reflects the legal and compliance context. Generic AI tools are not trained on this register and frequently produce translations that read naturally but fail to meet the specific language requirements of the regulation.
• Context-free translation of isolated strings. For device software, UI text strings are often translated without surrounding context. A term such as ‘Run’ could be a button label, a diagnostic command, or a noun, with very different appropriate translations. Without context, AI tools guess.
• Mistranslation of dosage, measurement, and safety-critical content. Numerical values, units, and safety warnings are the highest-risk content in medical translation. Errors here can have direct patient safety implications.
• Post-market surveillance and labelling inconsistency. Inconsistencies between translated versions of labelling, IFU, and post-market surveillance reports are increasingly scrutinised by notified bodies. AI translation, without a managed terminology database, makes consistency across documents very difficult to maintain.

Our downloadable IVDR Translation Guide covers the specific language requirements for IVD manufacturers in detail, including common translation pitfalls at the regulatory submission stage.

These questions will usually clarify the right approach:

• Is the document regulatory, patient-facing, or safety-critical? If yes, specialist professional translation with human oversight is required.
• Will it be submitted to a notified body, competent authority, or regulatory agency? If yes, the translation process needs to be evidenced and defensible.
• Does the content contain specialised medical or device-specific terminology? If yes, a translator with domain knowledge is essential.
• Is it an internal document, a gisting exercise, or low-stakes communication? AI tools may be appropriate in this case.

If you are uncertain, speaking to us before making a decision is both straightforward and cost-free. We can advise on the right process for your specific documents, regulatory context, and timeline.