Translation quality assurance (TQA) is the set of planned activities that makes sure translated content reaches an agreed quality level every time, not just when a good linguist happens to be on the job. It starts before anyone translates a word, runs through production, and ends with checks and feedback that improve the next job. This guide explains what TQA covers, how it differs from QC and LQA, the steps to build, what to score, and how marketing teams keep it manageable across many markets and local partners.
TL;DR
- QA is preventive and process-based (style guides, glossaries, workflows). QC is corrective and checks the output. You need both.
- TQA aims for a predefined quality level reached consistently, not the highest possible quality on every asset.
- Score translations by error type and severity (critical, major, minor) so results are comparable across languages and vendors.
- Automate mechanical checks, keep humans on meaning, context and naturalness, and match review depth to content risk.
- Feed every finding back into glossaries and style guides, or the same errors will return.
Table of Contents
- What translation quality assurance means
- QA, QC and LQA: what is the difference?
- The steps of a translation QA process
- What to assess: error types and severity
- What to automate and what needs a human
- Match review effort to risk
- Measure, then close the loop
- Where marketing teams struggle with TQA
- Putting quality into the workflow with GearBox®
- A starting checklist
- FAQ
- Sources
What translation quality assurance means
Quality assurance is defined under ISO 9001 and ISO 17100 as planned, systematic activities designed to provide confidence that a product or service meets specified quality requirements. Applied to translation, that means the work is organized so errors are unlikely to occur, instead of being hunted down at the end.
Two points from the research are worth keeping in mind.
First, timing. An academic paper on the translation quality process argues that QA should begin before translation, ideally when the source content is created, and continue throughout. A confusing English source sentence becomes a confusing sentence in every language you publish.
Second, the target. TQA is not about reaching the maximum possible quality on everything. It is about guaranteeing that an agreed, predefined level is reached consistently. A legal disclosure and a social caption do not need the same level, and a good program says so up front.

QA, QC and LQA: what is the difference?
People use these terms interchangeably, which causes real confusion when you brief a vendor or build a workflow.
QA is process-oriented. It aims to prevent errors by establishing procedures, standards and tools throughout the workflow. Think style guides, approved terminology, briefs, and automated checks.
QC is output-oriented. It focuses on detecting and correcting errors in completed or partially completed translations. Proofreading after translation is the classic example. QC is part of QA, but it cannot replace it.
LQA evaluates the translation itself. Linguistic quality assurance is the structured evaluation of translated content against defined criteria, often documenting findings and calculating scores. It is an assessment, and it sits apart from the preventive systems that QA describes. MultiLingual makes the same distinction: QA means systems that prevent problems at any stage, while assessment evaluates a finished translation against predetermined parameters. The same article notes that on large projects, QA or LQA is often reserved for the post-translation stage because of time and cost pressure, which makes it effectively QC.
Functional testing checks the final format. It confirms that content displays and behaves correctly, including layout, links, input fields and character rendering.
Here is how they fit together:
| Activity | Question it answers | When it happens |
|---|---|---|
| Quality assurance | Is our process set up to prevent errors? | Before and during production |
| Quality control | Does this output contain defects? | After production steps |
| Linguistic quality assessment | How good is this translation, scored against criteria? | After translation |
| Linguistic testing | Does the language work in this environment? | In context (page, app, print layout) |
| Functional testing | Does the content render and behave correctly? | In the final format |
| Localization QA | Is it culturally relevant and usable overall? | Combines the above |
The last row follows the definition that localization quality assurance combines linguistic and functional checks and adds cultural relevance and user experience.
The steps of a translation QA process
The sequence below follows the structure described in the Stepes governance guide: define requirements, prepare assets, produce, evaluate, approve, improve. Here is what each step looks like for a marketing team.
1. Define quality requirements and purpose
Write down what each content type is for, who reads it, and what “good enough” means. A brand campaign headline, a dealer price sheet and a warranty notice carry different risks. Decide the quality bar per type before work begins.
If some assets need creative adaptation rather than faithful rendering, decide that here too. Our guide on transcreation vs. translation shows how to choose by asset.
2. Prepare language assets
Build the glossary, the style guide and the brief before the first file goes out. Include approved product names, terms that must never be translated, tone rules per market, and formatting conventions for dates, currency and units. These are the preventive tools at the heart of QA. Style guides, glossaries and automated technical checks belong to QA, while human linguistic review and in-context review belong to QC.
3. Produce the translation with qualified resources
Use professional linguists, give them the assets, and make sure they see the context: where the text appears, which image sits beside it, what the call to action is. Translators who see only isolated strings produce isolated strings.
4. Evaluate the output
Run automated checks first, then human review. The next two sections cover what to check and who should check it.
5. Approve
One accountable person or role signs off, and the approval is recorded. For regulated or franchise content, approval often involves legal or brand reviewers in the local market. A clear approval chain matters more than a long one.
6. Improve
Turn findings into updates. Findings, corrections and reviewer decisions should become updated language assets and better future performance. Without this step, you are doing QC and calling it QA.
What to assess: error types and severity
A TQA process should assess accuracy, completeness, terminology consistency, fluency, grammar, style, locale appropriateness and functional performance. Errors are normally classified by type and by severity: critical, major or minor.
Typical checks at the QC stage are spelling, grammar, omissions, numbers and comprehensibility. Numbers deserve special attention in marketing: prices, percentages, phone numbers and dates are where a small slip does the most damage.
A simple severity scale for marketing content:
- Critical: changes the meaning in a way that creates legal, safety or financial risk, or is offensive. Example: a wrong rebate amount or a missing disclaimer.
- Major: misleads or confuses the reader, or breaks brand terminology. Example: a product name translated when it should have stayed in English.
- Minor: a style or fluency flaw that does not obstruct understanding. Example: an awkward but correct phrase.
There is also an international standard for scoring. ISO 5060:2024 provides a framework for segment-by-segment assessment of human, post-edited or unedited machine translation output, and its error typology is inspired by Multidimensional Quality Metrics (MQM) and classifies errors by type and severity to improve agreement between reviewers. Using a shared typology means a “major” error in German and a “major” error in Portuguese mean the same thing.
What to automate and what needs a human
The most reliable pattern is a combination. Automated checks cover missing content, spelling, tags, placeholders, numbers, dates, glossary compliance and formatting, while humans review meaning, context and naturalness.
In practice:
| Automate | Keep human |
|---|---|
| Missing or untranslated segments | Whether the message lands with the local reader |
| Glossary and terminology compliance | Tone, humor, cultural references |
| Numbers, dates, currency formats | Whether a claim is acceptable in that market |
| Tags, placeholders, broken formatting | Overall naturalness and flow |
| Spelling and basic grammar | Final sign-off on high-risk content |
For measuring quality, the research literature groups automated methods into n-gram surface matching, deeper linguistic features, and deep-learning models, and manual methods into traditional criteria such as intelligibility, fidelity, fluency, adequacy and comprehension. Scores from automated estimation are useful for triage. The EUATC points out that ISO 5060:2024 keeps human expertise at the center of the evaluation, which is a good rule for anything customer-facing.
Match review effort to risk
Reviewing everything at the highest level is slow and expensive. Tiered systems solve this by aligning review effort and methods with content type and risk level.
A workable tiering for a brand that markets through local partners:
- Tier 1, high risk: legal, pricing, warranty, regulated claims, anything with financial or safety consequences. Full linguistic review, scored, with local approver sign-off.
- Tier 2, brand-critical: national campaigns, hero landing pages, packaging copy. Linguistic review and in-context check before launch.
- Tier 3, high volume, low risk: social posts, local ad variations, internal communications. Automated checks plus spot sampling.
The tiers also tell your vendors what to expect and tell your local partners which assets they can adapt freely. Our marketing localization workflow guide goes deeper on how to organize this at scale.
Measure, then close the loop
A score is only useful if it changes something. A scalable loop is measure quality, prioritize high-impact issues, apply corrective actions and measure again, so recurring problems shrink instead of being fixed one at a time.
A quality-audit framework from a language services provider treats this as a continuing cycle of planning, assessment, documentation, corrective action, follow-up and improvement, and tracks quality over time with Errors Per Thousand, which allows comparison across languages or content types.
To run a lightweight version:
- Sample a fixed share of each tier every month.
- Log each error with type, severity, language and vendor.
- Review the top three recurring error types per quarter.
- Fix the cause: a missing glossary entry, an ambiguous source sentence, an unclear brief.
- Re-measure the next cycle.
Where marketing teams struggle with TQA
Translation QA is relatively simple when one team translates one product into three languages. It gets hard when content is created and adapted by many hands.
A brand that sells through dealers, franchisees or distributors faces specific problems:
- Local partners create or edit their own versions, outside the central review process.
- Approved glossaries and style guides live in shared drives where no one finds them.
- Final files exist in five places, and nobody is sure which is current.
- Vendors in different regions work to different standards.
None of these are linguistic problems. They are logistics problems, and they are why TQA for marketing is as much about workflow as about language. If your content pipeline is growing across many locations, it also helps to look at multi-location content automation alongside your review process, so volume does not outrun your checks.
For franchise and multi-region programs, see our guides on multilingual brand management and brand consistency across global markets.
Our perspective
We have spent more than 20 years helping brands that sell through other people: dealers, distributors, franchisees, agents and advisors. What we see again and again is that translation quality breaks down at the handoff, not at the translator’s desk.
A glossary that is not in front of the local marketer will not be used. An approved asset that cannot be found will be recreated. A review step that lives in someone’s inbox will be skipped when a campaign deadline hits.
That is why we built GearBox® around the idea that campaigns, vendors, assets, orders and approvals belong behind one login. Corporate keeps control of the brand and the budget, and local partners get approved, localized materials they can use without starting from scratch. Approval hierarchies make sure the right reviewer signs off, and asset management, including integrations with Adobe DAM and Widen, keeps the current version the only version people can find. GearBox supports multilingual and multicurrency programs, which is where preventive QA matters most.
Our view: treat translation QA as part of marketing logistics. Prevention is cheaper than correction, and the easiest errors to prevent are the ones caused by people working from the wrong file.
Client
Putting quality into the workflow with GearBox®
If your translation checks keep failing because of process, not language, the fix is a platform that holds assets, approvals and vendors together. GearBox® brings campaigns, vendors, co-op funds, orders and brand assets behind one login, with approval hierarchies so localized content is reviewed before local partners use it.

See how GearBox® works for brands that market through local partners, or read about digital asset management at scale to see how approved, current files stay findable. For regulated programs, our page on secure and compliant localized campaigns explains how we approach control across regions. Ready to talk through your program? Schedule a demo.
A starting checklist
Use this to stand up TQA in a month:
- List your content types and assign each a risk tier.
- Write or update the glossary and style guide for your top three markets.
- Create a one-page brief template for every translation request.
- Turn on automated checks for numbers, tags, terminology and missing text.
- Choose an error typology with three severity levels and use it for every review.
- Name one approver per market and record approvals.
- Sample, score and log errors monthly.
- Update language assets from the findings each quarter.
FAQ
What is translation quality assurance?
Translation quality assurance is the planned, systematic set of activities that gives confidence a translation meets agreed quality requirements. It covers preparation (glossaries, style guides, briefs), production, checks and feedback. It begins before translation and continues after delivery.
What does QA mean in translation?
QA means the preventive side of quality: procedures, standards and tools designed to stop errors from happening. Examples are approved terminology, style guides and automated checks. It differs from QC, which inspects finished work and corrects defects.
What is QC in translation?
QC is the checking that happens after production steps, such as proofreading a finished translation for spelling, grammar, omissions and numbers. It is corrective and output-focused. It is part of QA but cannot stand in for it.
Can machines do translation QA on their own?
Machines handle mechanical checks well: missing content, tags, placeholders, numbers, dates and glossary compliance. Human reviewers are still needed to judge meaning, context and naturalness, especially for customer-facing and high-risk content.
Sources
- What Is Translation Quality Assurance? A Seprotec Guide
- Quality assurance process in translation (ACL Anthology, 2014)
- QA vs. QC in translation (POEditor)
- Translation Quality & Governance Guide (Stepes)
- Linguistic quality assurance in localization (MultiLingual, 2017)
- Do you still need Language Quality Assurance? (Acclaro, 2025)
- Localization quality assurance explained (Lokalise, 2024)
- Translation Quality Assurance guide (Crowdin)
- ISO 5060:2024 overview (EUATC)
- Translation Quality Assessment: A Brief Survey on Manual and Automatic Methods (ACL Anthology, 2021)
- Translation accuracy: Measurement, challenges, & improvement (Smartling, 2026)
- How to Build a Scalable Translation Quality Assurance Process (Smartling, 2026)
- Translation Quality Audit Process (Translated)


