Translation, localisation, transcreation: the spectrum
The three words are often used loosely, which leads to the wrong work being commissioned and the wrong expectations being set. They describe a spectrum of how far the target text may depart from the source.
| Approach | What the target text keeps | What may change | Typical assets | Who does it |
|---|---|---|---|---|
| Translation | Meaning, structure and terminology | Grammar and word order only | Specifications, terms and conditions, help articles, product data | Professional translator, often with translation memory |
| Localisation | Meaning and purpose | Formats, units, examples, references, idiom | Web pages, emails, onboarding flows, sales collateral | Translator with marketing experience and a style guide |
| Transcreation | Intent and effect | Words, imagery, structure and sometimes the idea itself | Taglines, headlines, ad copy, campaign concepts, video scripts | Copywriter who is a native speaker of the target language, working from a brief |
The further right an asset sits, the more it costs, the longer it takes and the more it depends on the brief rather than the source text. Transcreation is closer to copywriting than to translation, and it is commissioned and reviewed as copywriting: against the objective, the audience and the brand, with options and a rationale, not with a word-for-word check against the original.
A useful habit is to label every asset in the localisation request with its place on the spectrum. The label tells the supplier what is being bought, tells the reviewer what to check against, and stops a translator being blamed for a flat headline they were never asked to rewrite. When a market reports that the copy is accurate but flat or odd, the diagnosis of translated marketing copy that reads wrong starts from that label and traces the fault to the brief, the termbase, the style guide, the supplier, the machine draft or the review.
What to transcreate and what to translate
The deciding question is what the copy is for. Copy that informs, instructs or binds needs to say exactly what the source says, so it is translated, and accuracy matters more than flair. Copy that persuades, entertains or needs to be remembered has to produce the same response in the reader, which often means saying something different, so it is transcreated.
Wordplay, rhyme, idiom, humour and cultural references almost never survive translation. A pun on a product name, a reference to a national sporting moment or a joke about the weather in one country will at best be explained in the target language, which kills it, and at worst be misread. The transcreation brief should state what the line is meant to make the reader think or feel, so the writer can find a local route to the same place. Press releases split the same way: facts and boilerplate are translated, while the headline, lead and quotes are transcreated and approved by the person quoted, as the guide to running PR in several countries explains.
Claims are a special case. A claim that is substantiated and acceptable in one market may need evidence, qualification or removal in another, because advertising rules differ. The Advertising Standards Authority sets out the UK's rules for advertising, and other markets have their own regulators and codes. Treat every comparative, environmental, health or price claim as something to be checked locally rather than translated, and route it through whoever owns compliance in that market. This is general direction, not legal advice.
In between sits the bulk of web, email and sales copy, which is localised: the meaning is kept but examples, formats, references and idiom are adapted. Most of the decisions there are covered by a good style guide, which is why the style guide is worth building before the first large batch of content goes out. For email, the guide to running email across countries and languages applies the same split to subject lines, body copy and footers, and stores each subscriber's language so the right edition reaches them.
Termbases, style guides and the brief
Three documents keep a brand consistent across languages. They are cheap to create compared with the cost of fixing inconsistent copy across a whole site later.
The termbase is a list of terms with their approved translation in each language, plus terms that must not be translated at all (brand names, product names, some feature names). It includes a short definition and context for each term so translators choose correctly, and it records rejected alternatives so a term does not drift back. Without a termbase, the same feature ends up with several names in the same language, and customers and support teams cannot tell whether they are the same thing. Because a new name will not be translated, it has to be checked with native speakers in every market before it is chosen, which the tutorial on naming a product or brand builds into its screens.
The style guide covers voice and conventions per language: formal or informal address, how directly to speak, sentence length, punctuation and capitalisation conventions, how to write dates, numbers, currency and units, and how to handle gender-inclusive language where the language marks gender. It also records local sensitivities the market team knows about. The general principles of keeping content consistent over time are covered in the content governance guide; the localisation style guide is the language-specific layer of that governance. The tutorial on writing a localisation style guide builds one for a target language with the in-country reviewer, from the address decision to reviewed examples.
The brief travels with every job. For translation it can be short: audience, context (where the text appears), any length limits and the label on the spectrum. For transcreation it is a real creative brief: the objective, the audience in that market, the single thought the reader should take away, mandatory elements, things to avoid and how the work will be judged. A transcreation request that consists only of the source text is a translation request in disguise, and it will come back as one. The localisation brief template gives these fields as a table, from the approach for each asset to the reviewers who sign off.
Keyword research in the target language
Search terms do not translate. People in another market may use a different word for the category, a borrowed English term, a brand name used generically, or a phrase that reflects a different stage of awareness. A page optimised for a translated keyword can be perfectly written and still miss most of the search demand, which is one of the commonest reasons localised pages attract little organic traffic.
Keyword research therefore happens in the target language, ideally led by a native speaker with access to the same research tools used at home. Start from the home keyword set as a list of needs, not of words; for each need, find the terms people in the market actually use, then check what the search results for those terms show. Classifying each term by intent, as described in the intent classification guide, matters more abroad, because the same need can sit at a different stage of the journey in a market where the category is newer. The tutorial on keyword research for a new market sets out that research step by step, from briefing the native-speaking researcher to deciding the pages.
The approved keyword list then goes into the termbase and the brief, so translators and transcreation writers use the searched-for terms in headings and copy rather than their own synonyms. Search engine optimisation (SEOImproving visibility in unpaid search results.) for the localised pages relies on this. So does paid search: ad copy written around the terms people use tends to earn a better click-through rate (CTRShare of impressions that result in a click.) than copy built on translated home-market keywords, and the landing page has to use the same language for the message to match.
Machine translation with human review
Machine translation has improved a great deal, and generative models can now produce fluent first drafts in many languages. Fluency is the trap: a fluent draft hides its errors, so they are found by customers rather than by reviewers. Errors cluster where marketing is most exposed: brand and product terms, claims, tone, idiom and anything that depends on context the model was not given.
The workable model is machine translation followed by human post-editing, applied by content type:
| Content type | Machine translation role | Human role |
|---|---|---|
| Product data, help articles, long informational content | First draft, constrained by the termbase | Full post-edit by a qualified linguist, then spot review in context |
| Web pages and emails | First draft or reference only | Edit to the style guide, check every claim and term, review in context |
| Headlines, ads, taglines, campaign concepts | Not recommended as a starting point | Transcreation from the brief by a native copywriter |
| Legal, regulatory and claims text | Not recommended | Professional translation and local compliance review |
The editorial controls that apply to any AI-assisted copy apply here as well. The AI-assisted content guide covers the review standard, disclosure considerations and the human accountability that has to sit behind published content. For localised content the reviewer must also be fluent in the target language; a reviewer who cannot read the output is checking formatting, not quality. Content that demonstrates experience and expertise, the qualities summarised as E-E-A-TExperience, Expertise, Authoritativeness, Trust., is especially hard to produce by machine in another language, because local examples, local references and local proof are exactly what a model translating from the source cannot supply.
In-country review and sign-off
In-country review is the step where a native speaker who knows the market reads the localised asset in its final context and confirms it works. It is distinct from the translator's own quality check: the translator checks the translation against the source, while the reviewer checks the result against the market.
Reviewers should be named per market and per content type, briefed on the brand, the termbase and the style guide, and given a turnaround time that the launch plan respects. Market colleagues in sales or support are often asked to review in their spare time, which is why review becomes the bottleneck; if internal reviewers are used, the time should be agreed with their managers and recorded as part of their role. Reviewers should correct the termbase when they change a term, not just the asset, or the same correction will be needed on every future job.
Sign-off has three parts, and they are often confused. Language sign-off confirms the copy is correct and natural. Commercial sign-off confirms the message, offer and pricing are right for the market, and belongs to the market ownerThe one named person accountable for a market's plan, channel mix, partners, results and commentary, and for commercial sign-off on its localised work.. Compliance sign-off confirms claims, disclosures and legal text are acceptable locally. Each should be explicit, recorded, and owned by one person.
The test of the whole workflow is performance. Compare the localised edition's conversion rate (CVRShare of visits or clicks that convert.) with the home edition on the same page type; a large, persistent gap is a signal to review the copy, the offer and the trust content together. The sibling guide on localising websites and campaigns covers the non-language reasons a page underperforms, and the localisation hub shows how the language and asset workflows fit together.
A fintech app splits its copy across the spectrum
Tallyway, a UK budgeting app, prepared its launch in Spain with about 18,000 words of copy. It labelled each asset before briefing suppliers: 12,000 words of help articles and in-app text went through machine translation with full post-editing against a 240-term termbase; 5,000 words of web pages and onboarding emails were localised by a translator with a fintech background; and 30 headlines, the app-store description and four paid social concepts were transcreated by a Spanish copywriter from a one-page brief. The copywriter dropped the UK tagline's pun and wrote three alternatives; the market owner chose one after a small preference test. Review took nine working days because the two in-country reviewers had blocked the time in advance.
Common mistakes
- Translating the taglineA short line that expresses the brand's position to customers and, used consistently for long enough, becomes one of the cues that identify the brand.. Persuasive lines sent for translation come back accurate and lifeless. Label them for transcreation and brief them as copy.
- No termbase before the first batch. Every translator picks their own terms, and the inconsistency costs more to fix than the termbase would have cost to build.
- Translating keywords. Localised pages built on translated home-market terms miss the words people actually search. Research terms in the target language.
- Treating fluent machine output as finished. Fluency hides errors in terms, claims and tone. Every machine draft needs a qualified human editor who reads the language.
- Unnamed reviewers. Review by whoever has time becomes review by nobody. Name reviewers per market and agree their time.
- Translating claims instead of checking them. A claim acceptable at home may not be acceptable in the target market. Route claims through local compliance.
- Corrections that never reach the termbase. Fix the source of the error, not only the asset, or it will recur on the next job.
