DeepL's privacy terms are more checkable than its quality claim

The writer's problem

A publisher weighing machine translation for a foreign-rights edition has to separate a vendor's marketing claim about quality from what actually happens to the manuscript once it is uploaded. DeepL's own product page and privacy policy, as retrieved on 16 September 2026, address those two questions separately.

What the documents show

The product page markets DeepL Translator as offering unmatched accuracy and describes the company as delivering human-like quality and enterprise-grade security to more than 200,000 business teams worldwide; both statements are DeepL's own description of its product, not findings from an independent evaluation cited on the page. The same page lists separate tiers for individuals, teams, businesses, and enterprises under DeepL Translator, alongside distinct tiers for DeepL Voice and the DeepL API. The privacy policy draws a sharper, more checkable line than the marketing copy: on the free tier, we process the content you upload and their translations or improvements for a limited period of time to train and improve our neural networks and algorithms, and the free tier may not be used for the translation or improvement of texts containing personal data of any kind. On paid Pro accounts, by contrast, the policy states your texts will not be used to improve the quality of our services, and that submitted texts and documents will not be permanently stored and are deleted after complete performance of the contractually agreed services.

The editorial choice

This is an editorial position, not something either document states: a publisher should treat DeepL's accuracy language as a starting claim to be tested against the specific manuscript's register and idiom, not as a substitute for a bilingual editor's read of the translated text, particularly for literary or long-form prose where DeepL's own documentation makes no accuracy claim specific to that genre.

What stays with the author

Neither the product page nor the privacy policy states whether a given translated passage preserves tone, ambiguity, or a deliberate stylistic choice from the original; that judgment sits with whoever reviews the output, not with the tool. The privacy policy's data-handling distinction between free and paid tiers is a contractual fact a publisher can verify; a translation's fidelity to the source is not something either document measures.

• Is a free-tier account being used for a manuscript, given the stated restriction on personal data and the stated training use of uploaded text?
• Has a bilingual reader checked the machine translation against passages where tone or ambiguity matters most?
• Does the contract for this foreign edition specify who is responsible if a licensed translator later disputes a machine-translated draft?

A vendor's own accuracy claim and a vendor's own data-retention policy are two different kinds of statement, and only one of them is something a publisher can check directly against a contract term rather than against a manuscript.