Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

International Marketing › Troubleshooting

Translated Marketing Copy Reads Wrong

Why localised copy is accurate but flat or embarrassing: translation where transcreation was needed, no termbase or style guide and no in-country review.
Troubleshootingintermediate16 min readUpdated

First, label each disputed asset with what it was commissioned as

Start from the complaint and the assets it names, not from the translator. List each disputed asset and write beside it what it was sent out as. The translation and transcreation guide places every marketing asset on a spectrum: translation keeps meaning, structure and terminology, for specifications, terms and help articles; localisation keeps meaning and purpose but adapts formats, examples and idiom, for most web and email copy; transcreation keeps only intent and effect, for taglines, headlines, ad copy and campaign ideas, which a native copywriter rewrites from a brief. A label on every asset settles what work the supplier is doing and what the reviewer judges it by, and spares a translator the blame for a headline nobody asked them to rewrite. The glossary entry for transcreation gives the short version.

The label often explains the complaint by itself. A tagline that "means something odd" and went out as translation lost the pun or rhythm it relied on; a help page that reads "like a manual" and went out as translation is doing its job; a feature with two names on two tabs was never the translator's call. Group the disputed assets by what they show: wrong point on the spectrum, terms that differ, tone and address that shift, errors clustered in one supplier's jobs, fluent pages nobody edited, corrections that keep coming back. Each group is a row in the table below.

Two questions speed the diagnosis. First, who signed what off? The translation guide separates language, commercial and compliance sign-off, and a RACI matrix with one accountable person per decision per market shows at once which was never assigned. Second, is the complaint about the words? The brand identity and voice guide treats voice as the constant and tone as how it adjusts; "does not sound like us" is a style guide problem, while "looks foreign" is often about prices, proof, imagery and checkout, the path the page on an international site edition that is not converting walks. The troubleshooting hub for international marketing lists the neighbouring symptoms, including localisation that delays every launch, which often shares the unnamed-reviewer cause with this one.

Diagnosis table

Diagnosis

SymptomCheckLikely causeFix
Taglines, headlines and ad lines are correct and lifeless, or mean something odd; the complaint names the lines people were meant to rememberThe label on each disputed line in the localisation request, and whether a brief with the intended effect went with itPersuasive copy sent for translation when it needed transcreation from a briefBrief the lines as copywriting and have a native copywriter rewrite them from the brief, with options and a rationale
The same product, feature or category word appears in two or three forms across pages, emails and support articlesA search of the edition for each product and feature name against the list of approved terms, if one existsNo termbase, so brand, product and feature names drift between pages and suppliersBuild the termbase with the in-country reviewer, with definitions, do-not-translate names and rejected alternatives, and send it with every job
Formal and informal address, sentence length and formats change from page to page; the edition reads as several voicesA sample of pages from different jobs or translators read by the reviewer for address, register and formatsNo style guide for the language, so address, tone and conventions vary by translatorWrite the language's style guide with the reviewer, decide address once, and turn each voice trait into a habit with examples
Faults cluster in one supplier's jobs; term and register corrections repeat batch after batchThe reviewer's correction log by supplier and by kind, and what each job was sent withA supplier chosen on a portfolio rather than a paid test piece, or jobs sent without the briefRun paid test pieces judged blind by the reviewer, match suppliers to the label on the work, and send every job with the full pack
Help pages and product data read fluently and carry wrong terms, altered claims or odd idiom that customers find firstWhich pages went through machine translation and whether a qualified linguist post-edited them in fullMachine translation treated as finished because it read fluentlyPost-edit every machine draft in full against the termbase, keep machine drafts away from persuasive copy, and gate publication on a fluent reviewer
Corrections are made on the asset and the same errors return on the next job; no one can say who signs off languageThe sign-off record for recent jobs and whether reviewer changes reached the termbase and style guideNo named in-country reviewer, or one who corrected the asset but never the termbaseName a reviewer per market and content type with agreed time, record the three sign-offs, and make every correction a change to the files

Fix, verify and prevent each cause

Persuasive copy sent for translation

The translation guide sets out what transcreation needs: a proper creative brief (objective, the market's audience, the single thought the reader should take away, what must appear and what to avoid) and a native copywriter who writes to the brief, not to the source words, returning options with a rationale. Send only the source text and the guide's verdict is that you have ordered a translation under another name. Puns, rhyme, idiom, humour and cultural references rarely survive the crossing, so a tagline built on one arrives accurate and empty. Claims are treated separately: because advertising rules differ, a line that is fine at home may need evidence, qualification or removal elsewhere; the Advertising Standards Authority sets out the UK's rules and other markets have their own regulators and codes. That is general direction, not legal advice.

Fix: relabel every headline, tagline, ad line, campaign idea and video script as transcreation, brief each with its intended effect, and commission a native copywriter; route any claim through the market's compliance owner. Words inside creative count too: the guide to adapting creative for different markets has supers and end frames transcreated from the brief.

Verify: the reviewer and market owner judge the new lines against the brief, not the source, and local people read them aloud; the test is whether a line produces the response the brief asked for.

Prevent: label every asset in every request, using the localisation brief template's assets-and-approach and intent-of-the-source rows.

No termbase, so names drift

The tutorial on building a termbase and translation memory traces drift to a missing decision: left to choose, every translator settles on a sensible word for a product or feature, and the next one settles on another, until one thing carries several names and customers cannot tell whether they are the same. A termbase settles the vocabulary once per language: the approved term, the names never translated, a definition and context, and the rejected alternatives with the reason, which keeps a term from creeping back with the next translator. The translation memory does the same for whole sentences, holding only pairs a reviewer approved.

Fix: pull candidate terms from the edition and the source catalogue, mark brand and product names as do-not-translate, have the in-country reviewer approve one term per language with the alternatives recorded as rejected, seed the memory from reviewed pages, and correct the live edition from the termbase in one pass.

Verify: search the edition for each product and feature name and confirm one approved form on every page, email and help article; then count term corrections in the next job's reviewer log.

Prevent: keep the master files where the business controls them, send both with every job as the brief template's termbase row expects, and review them at every market gate and whenever a name changes.

No style guide for the language

The tutorial on writing a localisation style guide starts from the many small choices a translator makes on each page: how to address the reader, how formal to be, how to write a price or a date. Without a guide for the language, every translator chooses differently and the edition reads as several voices. The guide records those choices once, inheriting the house editorial style guide by reference and deciding only what differs per language. Address and register, which touch almost every sentence, are settled with the reviewer, and voice is written as habits rather than adjectives, because a translator can follow "short sentences, no exclamation marks in product copy" but not "warm".

Fix: write the guide with the in-country reviewer, who owns the language decisions: address and register with exceptions by content type, each voice trait as a habit with an approved and a rejected example, local formats for dates, numbers, money and units, the names that stay in the source language, and known sensitivities; then have the reviewer re-read the disputed pages against it.

Verify: the reviewer marks a sample of pages from different jobs where address, register or formats break the guide; a batch produced with the guide should come back with far fewer such corrections and no repeats.

Prevent: publish the guide beside the termbase with a version date, link it from the brief template's tone-and-style row, and turn every repeated correction into a rule with an example.

A supplier chosen on a portfolio, or jobs sent without the brief

The guide to choosing and managing translation suppliers sees the usual buying error as giving one supplier all the work and then faulting them for both flat headlines and inconsistent terms, though the first is a skill mismatch and the second usually a missing termbase. A portfolio shows work for other brands; a paid test on your own content and pack, scored blind by the in-country reviewer against weights fixed in advance, shows what the supplier will do for you. Every job then carries the same pack: the brief with each asset's label, the current termbase and memory, the style guide, context such as screenshots, and the review route; work sent without it is corrected at the reviewer's expense.

Fix: where the reviewer's correction log shows faults clustering by supplier, run paid tests on real content with one pack for every candidate, have the reviewer score the anonymised pieces, and match each supplier to the label on the work: an agency or translator for translation, a translator with marketing experience for localisation, a native copywriter for transcreation. Make the pack a condition of every job, and ask each supplier to state where machine translation sits in their process and who post-edits it.

Verify: the reviewer scores the chosen supplier's first full-pack batch on the test's weights, and changes per job, repeat corrections and on-time delivery are recorded from then on.

Prevent: review suppliers on a fixed cycle against those records, decide renewals in time rather than by default, and keep the master files with the business so a change of supplier changes no approved wording.

A fluent machine draft published as finished

The translation guide's warning is that a smooth machine draft conceals its mistakes until customers find them, and that those mistakes gather where marketing is most exposed: brand and product terms, claims, tone, idiom and anything needing context the model never had. Its answer is machine output plus human post-editing, set by content type: a first draft constrained by the termbase for product data, help articles and long informational content, post-edited in full by a qualified linguist; a first draft or a reference only for web pages and emails, edited to the style guide with every claim and term checked; no starting role for headlines, ads and taglines; and professional translation with local compliance review for legal and claims text. The AI-assisted content guide supplies the editorial controls: every claim checked by a person against a source they can open, a review gate before publication that deadlines never bypass, and a named person who answers for the substance. The translation guide adds that for localised content the reviewer must be fluent in the language, because a reviewer who cannot read the output is checking formatting rather than quality.

Fix: find the machine-drafted pages nobody fluent post-edited in full, send each to a qualified linguist for a complete post-edit against the termbase and style guide, and pull any persuasive line that began as a machine draft for transcreation.

Verify: the reviewer's sample of those pages in context shows no term outside the termbase, no claim that departs from the source and no foreign-sounding idiom, and complaints about them stop.

Prevent: state the machine's role by content type in every brief, as the brief template's review-and-sign-off row asks, and keep it away from headlines, ads, taglines and claims.

No named reviewer, or corrections that never reach the files

The translation guide keeps in-country review separate from the translator's own check: the translator compares the text with the source, while the reviewer, a native speaker who knows the market, reads the asset in its final setting and judges it against the market. Reviewers are named per market and content type, briefed on the brand, termbase and style guide, and given time agreed with their managers, because sales or support colleagues reviewing in spare moments become the bottleneck. The rule most often broken is that a reviewer who changes a term corrects the termbase too, or the same fix is needed on every later job. The tutorial on reviewing localised copy in a language you do not speak covers the central owner's side: reading the reviewer's notes for patterns, asking why a change was made, settling disputes by the brief and the files, and recording the three sign-offs with a name and date.

Fix: name the reviewer for each market and content type in the RACI matrix, agree their time, give them the brief, termbase and style guide, and have them re-review the disputed assets in context, routing every change into the termbase or style guide as well as the asset.

Verify: the next job's correction log repeats no earlier correction, and every live asset carries recorded language, commercial and compliance sign-off.

Prevent: make the sign-off record part of every job, log corrections by kind, and treat any correction seen twice as a missing rule.

Prevention: a termbase and style guide before the next batch, test pieces for suppliers and named reviewers

Every cause above is a document or decision the first batch went out without, and each is cheaper to put in place now than to repair later. Build the translation guide's three documents in the order complaints arrive. The termbase comes first, because inconsistent names are what customers notice most and what spreads fastest; the style guide second, written with the in-country reviewer so address, register, formats and voice are decided once; and the brief, which travels with every job carrying each asset's label, the intent of each persuasive line, tone, termbase, target-language keywords, local checks and named reviewers. The localisation hub shows how those three sit beside the asset work.

Then choose the people. Run paid test pieces, judged blind by the reviewer on weights agreed in advance, before any supplier gets a batch, and match suppliers to the label on the work. Name the reviewer per market and content type with time agreed in writing, and write a RACI matrix for the edition: who is accountable for language, commercial and claims sign-off, who owns the termbase and style guide, and who may only propose changes. One accountable person per row is what stops review by whoever has time, which is review by nobody.

Make the files the edition's memory. Every reviewer correction becomes a termbase entry, a memory pair or a style guide rule with an example, and the correction log is read by kind at every market gate: rising term corrections mean the termbase is not being used, rising register corrections mean the style guide has a gap or is being ignored, and rising meaning corrections point at a supplier or a machine draft. Review suppliers on that log and their delivery dates on a fixed cycle.

Finally, keep the split straight between what is carried and what is adapted. The standardisation and adaptation comparison, which lists the Chartered Institute of Marketing among its sources, holds identity and tone of voice constant and transcreates the execution; the style guide turns that voice into each language's habits, and the termbase protects the fixed names. If the edition still reads wrong with all of this in place, the problem is creative rather than linguistic, and the next step is the adapting-creative guide's test of a local version against the master.

A garden tools brand finds its headline was translated, not rewritten (illustrative)

Illustrative example — not a real result

Ferndrosse Garden Tools

Ferndrosse Garden Tools, an invented maker of secateurs, spades and hand forks, receives a message from its distributor in a new market 6 weeks after the edition launches: the product pages read like a manual, and the tagline means something close to "cut yourself better". The marketing lead lists the 23 assets the distributor names and writes beside each what it was commissioned as.

The tagline, 4 campaign headlines and 9 paid social lines had been sent to the agency as translation, with no brief; the agency's work was accurate. Three product names appear in two forms each across the product pages, the emails and the 40 help articles, because no termbase existed and two translators had made different choices. The help articles had come from a machine draft that a colleague in the market, who reads the language but had not been asked to edit, skimmed for fluency; 11 of them use a term for "blade" that the market's gardeners do not.

Ferndrosse briefs a native copywriter to transcreate the 14 persuasive lines from a one-page brief and chooses one tagline from 3 options after the distributor's staff read them aloud. The marketing lead builds a termbase of 160 terms with the distributor's product trainer as in-country reviewer, records the 6 rejected forms, and writes a 5-page style guide with her, including the informal address for product pages and the formal one for warranty text. The 40 help articles are post-edited in full by a qualified linguist against the termbase. Before the next batch, 2 agencies and 3 freelance copywriters are paid for test pieces on real content, scored blind by the reviewer, and one of each is chosen. The reviewer's name goes on every job as the owner of language sign-off and of every termbase correction. The second batch of 35 pages comes back with 9 corrections, none of them a repeat, against 71 on the first.

Common mistakes

  • Blaming the translator for a flat headline. They delivered what was asked for; the line needed a transcreation brief.
  • Calling the source text a transcreation brief. Without the intended effect, the writer can only translate it.
  • Fixing names page by page. Without a termbase the next job brings the drift back. Decide each term once, with the reviewer, and record the rejected forms.
  • Giving translators adjectives instead of habits. A word like "warm" cannot be followed; "short sentences, no exclamation marks" can.
  • Picking a supplier from past work for other brands. Pay for a test on your own content and let your reviewer score it blind.
  • Publishing a machine draft because it reads smoothly. Smooth output hides wrong terms, altered claims and off tone; a linguist who reads the language post-edits all of it.
  • Reviewing by whoever has time. Name the reviewer, agree their time, and record the three sign-offs.
  • Corrections that stay in the asset. A correction that never reaches the termbase or the style guide is needed again on the next job.

Frequently asked questions

The translator is professional and the translation is accurate. Why does the market still say it sounds wrong?

Because accuracy was the wrong brief for the lines people are complaining about. A headline, a tagline or an ad is meant to produce a feeling, and a faithful rendering keeps the words and loses the effect, so a correct translation of a persuasive line reads as stiff or odd. Find out what each disputed asset was commissioned as. If it went out labelled as translation, the translator did the job asked of them and the fault is in the request; the line needs rewriting from the brief by a native copywriter, which is a different piece of work with a different supplier.

Can we fix this by asking a colleague in the market to go through the whole edition?

A colleague can find the problems; they cannot fix the cause by correcting pages. A reviewer who changes a term on one page and never records it will change it again on the next job, and a supplier who never sees the correction repeats the mistake. Name the reviewer formally, agree their time with their manager, and make every correction they make a change to the termbase or the style guide as well as the asset. Then the next batch arrives with fewer of the same faults, which is the only sign the fix has worked.

Is machine translation the problem, and should we stop using it?

Machine translation is not the fault; publishing its draft unedited is. A fluent draft hides its errors, and they cluster where marketing is most exposed, in product terms, claims, tone and idiom. Used as a first draft for help articles, product data and long informational copy, constrained by the termbase and post-edited in full by a qualified linguist who reads the language, it works. Used as a starting point for headlines, ads and taglines, or published after a fluency check by someone who cannot read the language, it produces exactly the complaint this page is about.

Terms used on this page

Frameworks

Where this page is used

Sources

  1. Advertising Standards Authority — ASA
  2. Chartered Institute of Marketing — CIM

Related