Three words can make a reassuring promise: “You are covered.”
On an insurance website, they might describe protection under a policy. In a product warranty, they could refer to repairs. In a message between colleagues, they might simply mean that someone has agreed to take a shift.
The wording is identical. The meaning depends on everything around it.
For a translator, that surrounding information helps determine which words to use. For a business using AI assisted translation, it raises an equally practical question: has the system been given enough information to make the right choice?
As translation becomes faster and easier to access, the quality of that information deserves more attention.
The conversation is moving beyond the model
Recent developments show how much attention translation technology providers are giving to context.
In February 2026, Google announced features that offer alternative translations with explanations of when to use them, alongside ways to ask about a particular situation, country or dialect. In September, DeepL announced connections to AI assistants that allow businesses to apply their company glossaries and style rules within those workflows. Google’s announcement and DeepL’s announcement describe these developments.
These capabilities are useful. They also point towards a broader business responsibility: deciding which terminology, tone and background information should guide the translation.
A tool may be capable of following a style guide. Someone still needs to ensure that the guide reflects how the organisation wants to communicate.
Context extends beyond the surrounding sentence
In translation, context includes the subject being discussed, the people being addressed and the action they are expected to take. It can also include information that never appears in the text itself.
Consider a software button labelled “Apply”. A screenshot may reveal whether it submits a job application, activates a filter or saves a setting. A list of isolated interface labels may leave that distinction unclear.
A product description presents a different challenge. The word “light” could refer to weight, colour or illumination. The product image and specifications may settle the question immediately.
Even when the basic meaning is clear, the audience changes the wording. Instructions for an experienced engineer may use terminology that would need explanation in a guide for a first-time user. Both versions need to be accurate, but they serve different readers.
The full document can resolve some of these questions. Screenshots, reference materials and a short conversation with the client can resolve others.
Better wording does not settle every question
AI can make a passage sound smoother without resolving an underlying misunderstanding.
A study published at ACL in July 2026 examined how large language models revise literary translations across seven language pairs. The researchers found that improvements came mainly from fluency, with limited improvement in how faithfully the translations conveyed the source. That finding concerns the literary tasks studied; it is not a verdict on every AI translation workflow. It does, however, illustrate why polished language and preserved meaning need separate attention. Read the research.
For a business, this distinction matters when reviewing a translation. A passage may be easy to read while using the wrong sense of a technical term, addressing the wrong audience or making a promise stronger than intended.
Asking a system to “make it more natural” may improve the writing. Resolving uncertainty about a product feature or an organisation’s policy requires the relevant information.
Cultural choices need a clear purpose
Context also shapes how a message is received.
A playful phrase may suit a consumer campaign but feel out of place in a service disruption notice. A familiar cultural reference may need explanation or adaptation for another audience. The appropriate level of formality can depend on the market, the relationship and the channel.
There is rarely one instruction that settles all of these choices. “Make it friendly” leaves considerable room for interpretation. A more useful brief explains who the reader is, how the organisation usually addresses them and what the communication needs to achieve.
Professional judgement includes knowing when to adapt and when to preserve. A campaign may allow considerable freedom. An approved product claim may require close adherence to the original wording. That boundary should be clear before either a linguist or an AI system starts making changes.
A useful brief can be short
Businesses do not need to write pages of instructions for every translation. A small amount of relevant information can remove a great deal of uncertainty.
Before a project begins, share:
Audience and market: who will read the content, where they are based and how much they already know.
Purpose: what readers should understand or do after reading it.
Placement: where the text will appear, with screenshots or layouts where useful.
Approved language: terminology, product names, style guidance and relevant previous translations.
Boundaries: wording that must remain precise, content that may be adapted and questions that need client approval.
These details help people and technology work from the same understanding. Reference material should also be current: an outdated glossary or superseded product description can introduce confusion of its own.
Knowing when to ask
One of the most useful contributions a language professional can make is recognising that a question needs answering.
If a phrase could describe two different processes, the next step may be to contact the author. If a screenshot contradicts an instruction, someone needs to establish which is current. If a marketing adaptation changes a claim, the client may need to approve it.
These decisions require an understanding of the project and responsibility for the finished communication. They belong throughout the workflow, from the initial brief to review in the final document or interface.
AI gives businesses more ways to produce multilingual content. Making that content useful still depends on knowing what the message means, who it is for and where it will be used.
For your next translation project, share the background that makes the wording clear to your own team. That knowledge is part of what needs to travel into the next language.
Planning a multilingual project? Talk to Simpson Soft UK about your content, audience and the best approach for your translation.