How our translations are made

A medical directive only works if the person reading it understands it. That is why bloodless.org is available in more than one language — and why translation quality is a patient-safety issue for us, not a cosmetic one. A mistranslated blood product name or a subtly shifted legal phrase could change how a document is read at the bedside.

This page explains openly how our translations are produced today: what is done by AI, what is checked and how, and where the limits are. We would rather be honest about our process than pretend every word has been reviewed by a certified translator when it has not.

Languages we support today

English is the authoring language — every piece of content is written in English first. Spanish (Latin American) and Brazilian Portuguese are fully available across the site, the directive documents, and email. French is in preparation as our next language.

Honest status labels

Every language carries an explicit support tier, so its status is never overstated:

• UI-only: the site is browsable in the language, but legal documents fall back to English with a visible notice.
• Full draft: everything is available in the language, and legal document text is clearly flagged as an AI draft pending certified review.
• Certified: legal document text has been reviewed by a certified legal translator and the draft flag is removed.

Our multi-step quality pipeline

New translations are not produced in a single machine-translation pass. Each language goes through a tiered, multi-step pipeline designed to be the strongest possible process when native-speaker review is not yet available:

1. Terminology first. Before anything is generated, contested medical and legal terms are verified against authoritative sources in the target language — national health authorities, legal codes, and clinical usage. One wrong term would otherwise repeat consistently across thousands of strings.
2. Risk-tiered generation. Legal document prose gets our most capable AI model, section by section. Medical content is generated with the verified terminology locked into every prompt. Interface text is generated in batches.
3. Independent adversarial review. A second, separate AI pass acts as a critic — not a translator — checking terminology adherence, formality, false friends, and meaning fidelity. Findings are fixed, not waived.
4. Back-translation of legal text. Legal prose is translated back into English in a fresh context and compared line by line with the original. Any drift in meaning around refusal language or agent powers blocks release.
5. Mechanical checks. Automated scripts verify every string exists in every language, placeholders and formatting are intact, banned mistranslations are absent, and locale typography conventions are respected.

An honest note about Spanish and Portuguese

Our Spanish and Portuguese translations predate this pipeline. They were produced with glossary-guided AI generation and spot checks, but without the independent review and back-translation steps described above. A retroactive review pass using the full pipeline is planned. In the meantime, both languages remain clearly flagged as AI drafts, and every page includes a way to report a translation problem.

Provenance and freshness tracking

Every translated piece of medical content carries a record of how it was produced — human-written, AI-generated, or AI-generated and human-confirmed. When the English source text changes, every translation of it is automatically flagged as stale and queued for re-review, so an updated English question can never silently coexist with an outdated translation.

Help us improve

Native speakers are the best proofreaders. If you are browsing in Spanish or Portuguese and something reads wrong — a term, a phrase, a tone — you can flag it directly from the page while signed in, or use the feedback option in the footer. Every flag is reviewed.