DeepL and KDP treat machine translation as a starting point

The writer's problem

A publisher considering machine translation for a foreign-rights edition needs to know whether the vendor is selling a finished translation or a draft that still needs a translator's pass, because that answer changes both the workflow and, on some retailers, the disclosure required at publication. DeepL's quality page and Amazon KDP's content guidelines, both retrieved 16 September 2026, address the workflow from two different angles.

What the documents show

DeepL's quality page, describing benchmark results from what it calls a March 2026 review with 48,000 blind evaluations judged by professional native-speaking linguists, claims its translation reduces work rather than eliminating it: one stated benefit is 50% lower translation workload that frees linguists and content teams from repetitive post-editing. The phrase post-editing is DeepL's own description of the human labor its tool is built around, not a claim that its output is publication-ready without a translator's review. KDP's content guidelines take a different, disclosure-focused position: they define AI-generated content as including translations created by an AI-based tool, and state that classification holds even if substantial edits are applied afterwards. The two documents describe complementary but distinct facts: DeepL frames its own product as designed for a post-editing stage that reduces but does not remove the linguist's task, while KDP's rule states that substantial human editing of an AI-produced translation does not change its disclosure category on that retailer.

The editorial choice

A publisher who treats DeepL output as a first pass with a post-editor reviewing every sentence is following the workflow DeepL's own benchmark language assumes, but that same publisher still needs to check KDP's disclosure rule separately, since editing after the fact does not, per KDP's own definition, move the content out of the AI-generated category for a translation specifically.

What stays with the author

Neither document tells a publisher how much post-editing is enough to call a translation reliable, and neither substitutes for a bilingual reader's judgment about whether meaning, tone and idiom survived the pass. Verifying the finished text against the source language remains the human translator's responsibility.

• Is machine translation output being treated as DeepL's own benchmark language assumes, a first pass needing post-editing, or as a finished text?
• Does the retailer's disclosure rule for AI-generated translations still apply after substantial human editing, as KDP's guidelines state it does?
• Has a bilingual editor checked meaning and idiom, not just fluency, in the finished translation?

Read together, a vendor's quality claims and a retailer's disclosure rule describe two separate obligations in the same workflow, and satisfying one does not resolve the other.