On this page
Where AI helps and where the risk is
AI translates fast and, given enough context, keeps a consistent tone across long texts. For a product page, an application note or a newsletter, it can produce a first draft in minutes. The risk sits in technical terms, units, decimals and safety warnings, which is exactly where a mistake has consequences.
Take one example. If the glossary does not say how to translate working volume, the output can confuse a bioreactor's working volume with its total volume. For a process engineer comparing 50 and 200-litre systems, that difference changes the purchase. That is why the order is glossary, draft and review, never the other way round.
Glossary first, translation second
The glossary is a table listing each technical term in the reference language with its approved translation in every target language. Engineering should sign it off and, ideally, so should the local distributor, who knows how customers actually talk. It then goes into the AI tool with every job, so the same terms are used every time.
- Process terms such as working volume, kLa, diafiltration or bioburden, with the approved translation or a decision to keep them in English.
- Product names, models and brands, which are never translated.
- Units and number formats, such as litres, rpm or each country's decimal separator.
- Warning and safety phrases, with fixed wording that is never rewritten.
Which AI tool is best for translating technical documents?
Whichever lets you work with a glossary, give context and keep the document's formatting. The workflow matters more than the tool. Run a test with one product page and one application note, have a native engineer review both and compare the errors. Confidential material such as customer URS documents should not go into tools that do not guarantee confidentiality in their contract.
How much control each type of content needs
Not all content carries the same risk. A newsletter with an awkward phrase can be fixed in the next issue, but a badly translated safety instruction can hurt someone. The table sorts the usual content of a bioprocess equipment manufacturer by the level of control it needs before publication.
| Content | Risk | How to translate it |
|---|---|---|
| Newsletters and social posts | Low | AI with glossary and a native read-through |
| Product and application web pages | Medium | AI with glossary and review by a native speaker from the industry |
| Application notes and white papers | Medium to high | AI with glossary and review by a native process engineer |
| Instruction manuals and safety information | High | Specialist technical translator, engineering review and version control |
| Qualification documents and declarations | High | Inside the quality system, with whoever is accountable |
Want us to look at which technical content you could translate with AI and which needs tighter control? Leave your details and we will tell you where to start.
Review my contentWhat a native engineer's review should cover
The reviewer needs to be a native speaker who understands the process. A good general translator can miss that diafiltration and dialysis are different things, or that a pH range has been written wrongly. The review compares source and target sentence by sentence, checks figures, units and warnings, and feeds corrections back into the glossary.
It is work that needs someone who can bridge departments. David Monje, founder of crida studio, led marketing at TECNIC Bioprocess Solutions for four and a half years, bridging engineering, product and sales to open markets such as Germany, Scandinavia and Asia. We set up the glossary and the workflow with your team, and the final review is done by a native-speaking engineer from your company or your distributor.
Can we translate the instruction manual with AI?
The regulation does not say which tool to use. It says the result must be in the language set by each Member State and be clear, understandable and legible, and the manufacturer is accountable for it. If AI produces the draft, the review must be complete and done by someone who masters both the language and the equipment.
Why manuals fall under the Machinery Regulation
For equipment that counts as machinery, Regulation (EU) 2023/1230 applies from 20 January 2027. Article 10(7) requires instructions for use, safety information and the information in Annex III to be in a language easily understood by users, as determined by the Member State, and to be clear, understandable and legible. Article 15 requires distributors to check this before they sell.
The same article allows instructions in digital form. In that case they must be printable and downloadable, stay online for the expected lifetime of the equipment and for at least ten years after it is placed on the market, and if the customer asks for paper at the time of purchase, it must be provided free of charge within one month.
Section 1.7.4 of Annex III makes one exception. Maintenance instructions for specialised personnel authorised by the manufacturer may be supplied in just one official EU language that those staff understand. The rest has to be in the language set by each Member State where the equipment is sold. So it makes sense to keep the glossary and translations under version control, just like the equipment itself.
Do datasheets need translating too?
If they are part of how you sell in that market, yes. Engineers compare specifications and want them in their own language or, at the very least, in English. AI saves a lot of time here because the text is short and repetitive, but every figure, every unit and the decimal separator need checking, since they vary from country to country.
What Google makes of an AI-translated website
Google looks at the value of a page more than the tool used to make it. Its policy on scaled content abuse targets pages created mainly to manipulate rankings that offer little or no value to users, no matter how they were created. Its examples include using generative AI to produce many pages without adding value, and translating scraped content from other sites automatically.
In practice, a site translated with AI and properly reviewed, with one page per language and correct hreflang, is just a normal multilingual website. A site with hundreds of pages machine-translated in one go and never checked is much closer to what Google describes as abuse, and a native-speaking customer will notice by the first sentence.
