Translation QA for agencies and localisation teams. Upload a bilingual file. Rules check tags, numbers, terms and formats. A model then reads every translated segment for a shifted meaning, a dropped sentence or the wrong register. Your reviewer gets the evidence on each finding and signs the report off. Your original file is never changed.
No card. 10,000 AI-reviewed words and 500,000 rule-checked words a month. Trados, memoQ, Phrase, Crowdin, Smartling, XLIFF and software string files.
The refund will eventually show on your statement.
Le remboursement apparaîtra éventuellement sur votre relevé.
Length, punctuation and terms all pass. No issues found.
Éventuellement means possibly, not in the end. A promise became a maybe, on a refund.
Le remboursement finira par apparaître sur votre relevé.
One bilingual file in, a report out: the rules, the AI review, the fix and the profile behind it. It works without sound.
Tag checkers pass copy that is mechanically perfect and still wrong. Nobody in the pipeline reads the sentence after the translator, and with machine output there was no translator. That is the half your client reads. What's Wrong reads it for you and shows your reviewer the evidence.
Rules settle what is certain: tags, placeholders, numbers, units, URLs, terms and locale formats. They run in seconds and give the same answer every time. Spelling runs beside them in 67 languages.
A model then reads every translated segment the rules passed and reports mistranslation, omission, addition, terminology, consistency and register, each with the words concerned and a proposed fix. A dropped negation or a reversed meaning is always critical.
The same model labels each rule finding it can assess: agrees, disputes or cannot tell. A disputed finding stays in the report at half weight until your reviewer decides, and the decision is recorded. No second model removes an AI finding.
Inside Phrase, XTM Cloud, Smartling and WorldServer through the browser extension, which inserts a fix into your editor in one click. As a Crowdin app and on a Smartling webhook. On a watched Dropbox folder, by email to the profile's own address, by REST API with webhooks, as a CI step that fails the build on the findings you choose, or from your chat assistant through the MCP connector.
Files from Trados, memoQ and Wordfast, XLIFF 1.2 and 2.1, spreadsheets, packages, gettext files and .json, .resx, .strings and .properties string files with their source. Nothing to migrate.
Every run is a record: the file, the profile, every finding with its segment, rule, severity and decision, and the name and date on the sign-off. A run that could not finish says not checked. The report, in HTML, Excel or CSV, is the proof that a person reviewed the machine's output, signed and dated.
Agencies attach it to the delivery and sell it as what it is: risk management for the client's content. One profile per content type says which checks are critical for a UI string, a help page or a campaign, so the record matches the risk. Your original file is never changed; a corrected copy with the fixes you confirmed is a separate download. For programmes above five million words a month, the Enterprise page covers per-word pricing, the API and the audit trail.
Rules are never metered. Unlimited seats from Team up. Annual billing is two months free and your price is fixed for 12 months.
Your own profiles, rules on 500,000 words. No card.
Start freeYour own profiles and term lists. One seat.
Choose SoloUnlimited seats, roles, API and webhooks, Crowdin and Smartling, email and watched folders.
Start 14-day trialEverything in Team, LQA scorecards on your own templates, white-label reports, a DPA on request.
Start 14-day trialNo. It reads everything so your linguists spend their time on the findings that matter, and it writes their decisions down.
No. Your segments are never used to train our models and your workspace admin chooses which providers may receive your text. Uploaded files are removed within 24 hours unless you agree a longer keep with us.
Trados, memoQ, Wordfast and Phrase bilingual files, XLIFF 1.2 and 2.1, spreadsheets and packages, gettext files and software string files (.json, .resx, .strings, .stringsdict and .properties, sent with their source file), and any segment pair through the API. The Help page on uploads lists each format; the pricing page answers billing questions.
The rules keep running, AI review pauses and the report says so. Free has 10,000 AI-reviewed words a month, Solo 50,000, Team 300,000 and Business 1,500,000. Premium models count 10 words per word. Every plan, Free included, can add a word pack or move up: 250,000 words for £125 / $165 / €145 or 1,000,000 words for £400 / $529 / €465, valid for two years. A word counts once per check, whatever runs on it. Rules never use your allowance.
Keep it for what it does well. Xbench and What's Wrong shows what each one catches, from someone who ran QA on Xbench for years.
Keep it for files that must stay on your computer. Verifika and What's Wrong shows what each one catches on our own delivered jobs.
Bring a delivery your current check passed. You get the complete report, every finding with its evidence, and the file back untouched. Free, no card, no integration.