Translation and localisation that keeps register, idiom and intent. Below are 7 copy-ready prompts. Fill in the [BRACKETS], copy, and paste into ChatGPT, Claude, Gemini or any capable assistant.
Machine translation solved the words a while ago. What is still hard is register, idiom, and the things that do not transfer — and that is where prompting makes the difference.
The 7 prompts
Translate with context, not word by word
Get a translation that works for its actual purpose.
Translate this from [SOURCE LANGUAGE] to [TARGET LANGUAGE]. TEXT: """ [PASTE] """ WHAT IT IS: [CONTRACT / MARKETING / TECHNICAL DOC / EMAIL / UI TEXT / LITERARY] AUDIENCE: [WHO READS THE TRANSLATION, WHERE] REGISTER: [formal / neutral / casual] TARGET VARIANT: [e.g. Latin American Spanish, Brazilian Portuguese, Simplified Chinese] PRIORITY: [literal accuracy / natural reading / persuasive effect] Produce: 1. THE TRANSLATION. 2. TRANSLATION NOTES for every non-obvious choice: - Where a word has no direct equivalent, what you chose and what is lost - Idioms: what you did instead of translating literally - Register decisions, including formal/informal address where the target language distinguishes them (tú/usted, tu/vous, du/Sie). State which you used and why - this is often the most consequential choice in the whole translation. - Where the source is ambiguous and the target forces a choice (gender, number, definiteness, tense) 3. WHAT DOES NOT TRANSLATE - wordplay, cultural references, humour, connotation. For each: what it did in the source, and the options - substitute a local equivalent, explain it, or drop it. Say what you did. 4. FORMAT AND CONVENTION CHANGES - dates, numbers, currency, units, addresses, names, punctuation conventions (quotation marks and spacing differ by language). 5. TEXT EXPANSION - if this is for a UI or a fixed layout, note where the target text is significantly longer or shorter than the source. Translated text commonly runs 20-30% longer in some language pairs and breaks layouts. 6. WHAT NEEDS A HUMAN - anything where a mistranslation would carry real consequences: legal terms, medical or safety instructions, anything binding. Flag these plainly rather than translating them confidently. 7. BACK-TRANSLATION of any sentence where accuracy is critical, so I can check the meaning survived. Do not produce a translation that is technically correct and reads as translated. If the priority is natural reading, prioritise it and say what you changed to get there.
What you get: A translation with decision notes, untranslatable elements handled explicitly, expansion warnings and back-translation of critical lines.
Tip: The formal/informal address decision in point 2 is the one that most often gets a translation wrong in a way monolingual reviewers cannot see.
Localise content rather than translating it
Adapt material so it works in another market.
Localise this for [TARGET MARKET]. CONTENT: """ [PASTE] """ SOURCE MARKET: [WHERE IT WAS WRITTEN FOR] TARGET MARKET: [COUNTRY/REGION AND LANGUAGE] WHAT IT IS: [marketing / product / support / legal / educational] GOAL IN THE NEW MARKET: [WHAT IT SHOULD ACHIEVE] Localisation is not translation. Assess and adapt: 1. EXAMPLES AND REFERENCES - brands, places, sports, holidays, public figures, historical events, TV and film references. Which do not land? What replaces each? 2. CULTURAL ASSUMPTIONS - assumptions about how people work, shop, pay, travel, or organise family life. Directness of communication style varies enormously and is the most common localisation failure: copy that reads as confident in one market reads as arrogant in another. 3. PRACTICAL CONVENTIONS - currency and price formatting, dates, measurement units, phone and address formats, paper sizes, name order, time formats, the working week. 4. LEGAL AND REGULATORY - claims that are permissible in one market and not another: health claims, comparative advertising, guarantees, data and privacy statements, required disclosures. Flag anything needing local legal review rather than adapting it yourself. 5. PAYMENT AND LOGISTICS - if relevant: payment methods people actually use there, delivery expectations, returns norms. 6. TONE - how formal, how direct, how much humour, how much self-promotion is acceptable. 7. VISUAL AND LAYOUT NOTES - text expansion, reading direction, colour associations, imagery that may not suit. 8. THE LOCALISED VERSION. 9. WHAT I CHANGED AND WHY - so a native speaker can review the decisions. 10. WHAT NEEDS A LOCAL PERSON - be explicit. There are things about a market that cannot be reasoned out from outside it, and a confident wrong answer is worse than a flag. Do not simply translate. If the content fundamentally does not work in the target market, say so.
What you get: A localised version with adapted references and conventions, flagged legal risks, and an explicit list of what needs a local reviewer.
Tip: Point 10 is the honest part. Localisation done from outside a market gets the units right and the tone wrong, and only a local reader can catch that.
Review a translation someone else produced
Quality-check translated content.
Review this translation. SOURCE ([SOURCE LANGUAGE]): """ [PASTE] """ TRANSLATION ([TARGET LANGUAGE]): """ [PASTE] """ PURPOSE: [WHAT IT IS FOR] AUDIENCE: [WHO READS IT] Review and report as a table: Source text | Translation | Issue | Severity | Suggested fix. Check for: 1. ACCURACY - meaning changed, added, or omitted. Severity critical for anything that alters an instruction, a number, a legal term, or a safety warning. 2. OMISSIONS - content present in the source and missing in the translation. Check systematically; these are easy to miss. 3. ADDITIONS - content in the translation that is not in the source. 4. FALSE FRIENDS - words that look like a cognate and mean something different. 5. LITERAL TRANSLATION - correct word-for-word, wrong in the target language. Idioms translated literally, source-language sentence structure preserved where the target would not use it. 6. REGISTER - too formal or too casual for the context. Check formal/informal address consistency throughout; inconsistency here is jarring and common. 7. CONSISTENCY - the same source term translated differently in different places. For technical content this is a serious problem. 8. NUMBERS, DATES, UNITS - transcription errors, unconverted units, date format ambiguity, decimal separator differences. 9. NAMES AND PROPER NOUNS - correctly handled, consistently. 10. NATURALNESS - does it read as originally written in the target language, or as a translation? Then: - CRITICAL ISSUES - anything that must be fixed before publication - THE OVERALL VERDICT - publishable / needs revision / needs retranslation - PATTERNS - recurring issues suggesting a systematic problem rather than isolated errors Be specific. Quote both versions for every issue.
What you get: A severity-rated issue table with both versions quoted, critical issues separated, and recurring patterns identified.
Tip: Point 7's consistency check catches the errors that damage technical documentation most. One term translated three ways makes a manual unusable.
Translate UI strings for a product
Handle interface text, which has rules of its own.
Translate these UI strings from [SOURCE] to [TARGET]. STRINGS: """ [PASTE - INCLUDE THE KEY/ID AND ANY CHARACTER LIMITS] """ PRODUCT: [WHAT IT IS] USERS: [WHO] PLATFORM: [web / iOS / Android / desktop] TONE: [DESCRIPTION] UI translation has constraints ordinary translation does not. For each string produce: key | source | translation | character count | notes. Apply these rules: 1. CONTEXT IS MISSING - for every string where you cannot tell from the text alone what it does, flag [CONTEXT NEEDED: your question]. 'Open' as a button and 'Open' as a status need different translations in most languages, and guessing produces embarrassing errors. 2. LENGTH - flag any translation exceeding a stated character limit, and give a shorter alternative. Note where the target language is systematically longer. 3. CONSISTENCY - the same source term must translate the same way everywhere. Produce a glossary of recurring terms and their chosen translation. 4. PLACEHOLDERS - preserve every variable exactly. Note where the target language's word order requires the placeholder to move, and whether the string format supports that. 5. PLURALS - flag every string with a count. Languages have different plural rules, and a string built by concatenation will be wrong. Say which plural forms the target language needs. 6. GRAMMATICAL GENDER AND AGREEMENT - where a string inserts a variable whose gender affects surrounding words, flag it. This cannot be solved by translation alone and needs a code change. 7. CONCATENATION - flag any string that appears to be a sentence fragment assembled with others. These almost never translate correctly and should be restructured as full strings. 8. TERMINOLOGY - platform conventions matter. Use the standard target-language term for standard UI concepts rather than a literal translation. 9. TONE - button labels and error messages have conventions that differ by language. Note where a direct translation would be rude or oddly formal. End with: THE STRINGS THAT NEED DEVELOPER CHANGES, not just translation.
What you get: Per-string translations with character counts, a term glossary, and flagged strings that need code changes rather than translation.
Tip: Point 7 is the one that saves projects. Concatenated sentence fragments cannot be translated correctly in any language with different word order, and the fix is in the code.
Write in a second language and get it corrected
Improve your own writing in a language you are still learning.
I wrote this in [LANGUAGE], which is not my first language. Correct and improve it. MY TEXT: """ [PASTE] """ MY FIRST LANGUAGE: [LANGUAGE] WHAT IT IS FOR: [PURPOSE AND AUDIENCE] REGISTER NEEDED: [formal / professional / casual] MY LEVEL: [beginner / intermediate / advanced] Produce: 1. THE CORRECTED VERSION - errors fixed, keeping my meaning and as much of my phrasing as is correct. 2. ERROR TABLE - Mine | Corrected | Type | Why. Types: grammar, word choice, word order, preposition, article, register, collocation, spelling. 3. FIRST-LANGUAGE INTERFERENCE - errors that come from my first language's structure. These are the ones I will keep making, so name the underlying pattern rather than just the instance. 4. WHAT WAS CORRECT BUT UNNATURAL - grammatically fine but not how a native speaker would say it. This is the hardest category to self-detect and the most valuable feedback at my level. Give the natural version. 5. COLLOCATIONS - where I used the wrong word pairing. These rarely appear in grammar books and are most of what makes writing sound foreign. 6. REGISTER CHECK - anything too formal or too casual for my stated purpose. Second-language writers commonly write too formally. 7. THE THREE PATTERNS to work on, ranked by how often they appeared. 8. THE POLISHED VERSION - if I want this to read as fluent rather than merely correct, the rewritten version. Kept separate from section 1 so I can see the difference between 'fixed' and 'native'. Explain in [MY FIRST LANGUAGE or English - state which you want]. Do not simply rewrite everything. I want to learn, so show me what was wrong and why.
What you get: A corrected version, an error table with first-language interference patterns named, and a separate fluent rewrite.
Tip: Section 4 is the one worth reading twice. Correct-but-unnatural is invisible to the writer and is what marks text as non-native.
Adapt tone across cultures for business communication
Write to someone in a different business culture without giving offence.
Help me write this message for a different business culture. WHAT I WANT TO SAY: [THE MESSAGE] MY BUSINESS CULTURE: [WHERE YOU ARE FROM / HOW YOU NORMALLY COMMUNICATE] THEIR BUSINESS CULTURE: [COUNTRY/REGION] RELATIONSHIP: [new contact / established / senior to me / junior / peer] CHANNEL: [email / message / call follow-up / formal letter] WHAT IS AT STAKE: [CONTEXT] MY DRAFT: [PASTE, or 'none'] Produce: 1. WHAT DIFFERS between the two communication cultures, relevant to this message. Consider: directness, how disagreement and refusal are expressed, how much relationship-building precedes business, formality and titles, hierarchy and who is addressed, how commitments are phrased, how 'no' is conveyed, and expected response times. 2. WHERE MY DRAFT OR INTENT WOULD MISFIRE - specifically. Directness read as rudeness, indirectness read as evasion, familiarity read as disrespect, or an assumption about decision-making authority. 3. THE ADAPTED MESSAGE. 4. WHAT I CHANGED AND WHY - so I learn the pattern rather than just getting one message written. 5. HOW TO READ THEIR REPLY - what a polite decline looks like in their culture, what enthusiasm looks like, and what silence is likely to mean. This is as important as sending the right message and is almost always omitted. 6. NAMES AND TITLES - how to address them correctly, including name order and whether titles are expected. 7. WHAT NOT TO DO - the specific mistakes people from my culture commonly make with theirs. 8. THE CAVEAT - business cultures vary by company, generation and individual as much as by country. Note where you are describing a generalisation, and say that an individual may not match it. Do not produce cultural stereotypes. Describe communication conventions, hedge appropriately, and say where you are uncertain.
What you get: An adapted message with the changes explained, plus guidance on reading their reply and an explicit generalisation caveat.
Tip: Section 5 is the half everyone forgets. Knowing that 'that would be difficult' means no in some business cultures prevents months of waiting for a deal that already died.
Build a translation glossary for consistent terminology
Lock down how key terms are translated before translating anything.
Build a translation glossary. SOURCE LANGUAGE: [LANGUAGE] TARGET LANGUAGE(S): [LANGUAGES] OUR CONTENT: """ [PASTE REPRESENTATIVE CONTENT] """ INDUSTRY: [SECTOR] BRAND TERMS THAT MUST NOT BE TRANSLATED: [PRODUCT NAMES, TRADEMARKS] EXISTING TRANSLATIONS: [PASTE, if any] Produce: 1. TERM EXTRACTION - every term needing a consistent translation. Group as: - BRAND AND PRODUCT - do not translate. List them explicitly so translators do not. - DOMAIN TERMS - industry vocabulary with an established target-language equivalent - PRODUCT CONCEPTS - things we named ourselves. These are the hard ones: translate, transliterate, or keep in source? Recommend one per term with reasoning. - UI ELEMENTS - buttons, labels, statuses, where platform conventions apply - COMMON PHRASES - recurring sentences worth standardising 2. THE GLOSSARY TABLE - Source term | Part of speech | Definition/context | Target translation | Do not use | Notes The 'do not use' column is the important one: the plausible wrong translations to rule out, especially where the obvious cognate is wrong. 3. AMBIGUOUS SOURCE TERMS - words used with more than one meaning in our content. Each meaning needs its own entry. Flag these; they are the main source of inconsistency. 4. TERMS THAT DO NOT TRANSLATE CLEANLY - where the target language has no equivalent, or splits one source concept into several, or merges several into one. For each, the recommended approach. 5. STYLE DECISIONS to apply consistently: formal or informal address, whether we use the imperative for instructions, capitalisation conventions, how we handle English loanwords. 6. INCONSISTENCIES IN OUR EXISTING TRANSLATIONS - if I provided any, where they disagree, and which to standardise on. 7. MAINTENANCE - who owns this, how a new term gets added, and how translators query an entry. The glossary is only useful if it is decided before translation starts. Flag anything needing a decision from us rather than guessing.
What you get: A grouped glossary with a do-not-use column, ambiguous terms split by meaning, and style decisions locked before translation begins.
Tip: The 'do not use' column is what makes a glossary actually work. Telling a translator what the wrong obvious choice is prevents more errors than telling them the right one.
Where AI actually helps here
- Translation with an explicit register instruction — formal, casual, regional
- Flagging what does not translate and offering alternatives
- Back-translation as a check: translate, then translate back, and compare
Where it falls down
- Low-resource languages, where quality falls off sharply
- Legal, medical and safety text, where an error has consequences and a human professional is the answer
- Knowing your audience’s variety — Urdu for Karachi and Urdu for Delhi are not interchangeable
The mistake almost everyone makes: Not naming the variety and register
“Translate to Spanish” gives you a neutral Latin American Spanish that reads as translated everywhere. “Translate to Spanish for a professional audience in Madrid, formal register, keep the product names in English” gives you something usable. Always: flag anything that does not translate cleanly rather than substituting silently.
Free tool: Prompt Templater
Runs in your browser. No sign-up, nothing uploaded.
Questions people ask
Is AI translation good enough to publish?
For internal and low-stakes content, usually. For anything customer-facing in a market you care about, have a native speaker read it — the errors are not grammatical, they are tonal, and those are the ones that cost you.
How do I check a translation I cannot read?
Back-translate it with a different model and compare against your original. Differences are not proof of error, but they show you exactly where to ask a human.