12 Checks Before You Send a Translation

TABLE OF CONTENTS

The translation already looks “done.” This is the short gate between that feeling and hitting Send—whether you are emailing a client, posting product copy, submitting homework, or shipping UI strings.

Machine translation and AI drafts save time. They also leave a predictable set of last-mile mistakes that a spellcheck will not catch: a US-style date that flips the day in Europe, a product name that got translated, a {name} placeholder that broke, or a link that still opens the source-language page. The American Translators Association stresses completeness, accuracy, readability, and a final pass in the real delivery format before release. You do not need a full localization program for that—you need twelve concrete checks you can run in about 5–15 minutes.

This is not a deep bilingual QA workflow. For longer document review, PPTX decks, or pro proofreading, use the deeper guides linked at the end. This page is the pocket gate: open the source and target side by side, then work the list top to bottom.

How to run this 5–15 minute gate

Situation	Suggested depth
Short email, chat reply, social caption	5 minutes — checks 1–6 + 12
Product page, proposal, homework essay	10 minutes — full list
UI strings, app store text, FAQ	15 minutes — full list + placeholder/link focus

Keep one browser tab (or window) on the source and one on the target. Fix issues in the file you will actually send—not only in a chat draft.

1. Dates and locale order

Confirm every date matches the reader’s locale, not yours.

US English often uses month–day–year (11/01/2026 = November 1).
Much of Europe and Latin America uses day–month–year (01/11/2026 = 1 November).
Many East Asian locales prefer year–month–day.

Unicode CLDR documents locale date patterns precisely for this reason: short numeric dates are ambiguous across borders. Prefer an unambiguous form for mixed audiences (1 November 2026 or ISO 2026-11-01) when the channel allows it. Also check whether the weekday still matches after reformatting.

2. Leftover source-language text

Scan for untranslated fragments that AI tools often skip:

Headings, captions, alt text, footnotes, headers/footers
Text inside tables, text boxes, speaker notes, or image overlays
Stub phrases left in the source language (TBD, lorem, TODO, N/A that should be localized)

ATA’s review guidance calls out hidden text under graphics and in tables as easy misses. Search the target file for distinctive source words (brand-less ones) to surface leftovers fast.

3. Names and do-not-translate (DNT) items

Protect items that should stay as-is unless the brief says otherwise:

People and place names (confirm spelling against the source)
Product names, feature names, legal entity names
Acronyms on a DNT list, code identifiers, SKU/model numbers

Build a tiny DNT list before you translate—even five terms help. For more on what usually should not flip languages, see What Not to Translate.

4. Currency symbols and number separators

Numbers are not “universal.” CLDR number and currency patterns treat . and , as placeholders that become locale-specific symbols.

Locale habit	Example meaning of `1.500` / `1,500`
en-US	`1,500.00` → one thousand five hundred
de-DE / many EU	`1.500,00` → one thousand five hundred
Mixed drafts	`1.500` can mean 1.5 or 1500—disastrous in a quote

Check currency code vs symbol (USD / $ / US$), symbol position, and decimal places (JPY often has none). Never change the numeric value while “fixing” separators.

5. Tone and register

Ask: would this sound right from this sender to this reader on this channel?

Cambridge Grammar defines register as varieties of language for different purposes, contexts, and audiences—and contrasts formal language (serious situations, people you do not know well) with informal language (relaxed, familiar contexts). A chatty AI draft in a first client email, or a stiff textbook tone in an app onboarding tip, both fail the send gate.

Quick test: read the greeting, one body paragraph, and the sign-off. If they disagree in formality, fix the mismatch. For cross-language tone patterns, see Formal vs Informal: 12 Tone Tips.

6. Placeholders and variables

Search the target for broken or translated placeholders:

{name}, {{count}}, %s, %1$s, ${user}
ICU / MessageFormat style {count, plural, ...} blocks

W3C internationalization guidance on composite messages warns that variables must stay intact and reorderable; translators (and AI) must not “helpfully” translate the token or hard-code English word order. Compare placeholder count and names against the source string. If the UI shows Hello, {name}! literally, you are not done.

7. Links still pointing at source-language URLs

Click—or at least hover and inspect—every important link:

Does /en/ or ?lang=en still appear when the page should be Spanish?
Do help-center and pricing URLs resolve to the target-locale equivalents?
Are tracking or UTM parameters intact, and is the destination live (not 404)?

A perfect sentence that sends readers to the wrong-language page still fails the send gate.

8. Numbers, units, and measurements

Beyond currency separators, verify:

Units kept or converted per brief (km vs mi, °C vs °F)
Phone numbers, postal codes, and times (12-hour vs 24-hour)
Percentages and ranges (10–15%) still match the source values

If the brief says “do not convert units,” leave them and only localize the surrounding wording.

9. Glossary and product-name consistency

Search the target for every required glossary term. Confirm:

Preferred term is used every time (no synonym drift)
Banned or legacy terms do not reappear
Capitalization matches brand rules (OpenL, not Openl)

Inconsistency is one of the fastest ways a short marketing page looks unedited. If you do not have a formal glossary, make a three-line list of must-keep names before you send.

10. Completeness vs the source (bullet and section count)

Do a structural count, not only a vibes check:

Same number of H2/H3 sections?
Same bullet or numbered-list length?
Tables: same row/column counts? Captions present?

ATA’s completeness check is blunt: skip one small word like “no” and the meaning reverses. Counting list items catches omissions that fluent reading misses. For a full bilingual document pass, use The Bilingual Document Review Checklist.

11. Formatting and length overflow

Longer target languages (German, Finnish, Spanish) often overflow:

Buttons, menu labels, slide titles, email subject lines
Table cells, mobile UI, PDF text boxes

Check at real size (100% zoom / device preview). Watch for truncated strings, overlapping text, broken line wraps, and fonts that do not support the target script. If you translated slides or decks, also skim PPTX Translation QA Checklist for speaker notes and chart labels.

12. Final read-aloud (or text-to-speech)

Read the target out loud—or use Word / OS read-aloud—while glancing at the source for numbers and names. ATA reviewers recommend read-aloud specifically to catch wrong numbers and unnatural phrasing. You are listening for:

Stumbles, calques, and sentences that need two passes to understand
Wrong numbers spoken aloud (fifteen vs fifty)
Tone that sounded fine silently but feels off spoken

If anything makes you pause, fix it before Send.

Quick copy-paste checklist

When you need a draft or a quick re-check of a short passage after edits, OpenL can help you regenerate or compare wording—then run this gate again before anything leaves your outbox.

When you need a deeper QA pass

Use this send gate for last-mile confidence. Switch to a deeper checklist when the stakes or length grow:

Translation QA Checklist: 15 Must-Check Items — fuller project QA
Translation Proofreading Checklist for Pros — linguist-grade proofreading
Bilingual Document Review Checklist — source-vs-target document passes

Twelve checks will not replace a paid review on a legal contract. They will catch the embarrassing misses that turn a good AI draft into a bad send.

Sources

How to Review a Translation — American Translators Association (ATA) — completeness, accuracy, readability, read-aloud, hidden text, final format check
Quality Control in Translation — ATA — register, numbers/proper nouns, client instructions, bilingual review habits
CLDR Date/Time Patterns — locale date format conventions
CLDR Number and Currency Patterns — decimal/grouping separators and currency patterns
Register — Cambridge Dictionary Grammar — register as purpose-, context-, and audience-driven language variety
Formal and informal language — Cambridge Grammar — formal vs informal usage cues
Working with Composite Messages — W3C Internationalization — placeholders/variables must remain intact and reorderable
Multidimensional Quality Metrics (MQM) — open issue typology used across localization QA (accuracy, fluency, locale convention, and related dimensions)