In 2025, the Danish Translators' Association and the Danish Authors' Society made it official: post-editing AI-generated text does not qualify as translation under Denmark's Public Lending Rights Act. 

That might sound like a bureaucratic detail about book royalties. It is not. It is the first clear legal line drawn between “translation” and “cleaning up after a machine” — and the language industry should be paying very close attention.

The ruling came after it emerged that some people had incorrectly received public lending fees for books where they were listed as translators but had only edited machine-translated output. 

The Danish Agency for Culture and Palaces updated its guidance accordingly: post-editing AI output does not meet the requirements of the Public Lending Rights Act. 

The professional associations went further, issuing new industry guidelines requiring that books produced through post-edited AI translation carry a specific disclosure statement, and that post-editors must not be credited as translators anywhere in the book or its metadata.

Why this matters beyond Danish publishing

This ruling is a microcosm of a tension playing out across every corner of the language industry. 

The core question — when a human edits an AI translation, who did the creative and intellectual work? — is not abstract. It determines how projects are scoped, priced, credited, and evaluated. 

However, in practice, the distinction between “light post-editing” and “retranslation from scratch” that delivers real translation by professionals is murky at the best of times. Industry research from 2025 found that 90–98% of professionals using machine translation or LLMs do some level of post-editing on the output. 

Additionally, 84% of language service providers had clients specifically request human editing to improve AI-generated content. That is not a niche workflow. That is most of the market, operating without agreed definitions of what the work actually involves.

Our Danish Lingonaut regularly receives client briefs that say: “This has already been machine translated, just check it over.” In half of those cases, the MT output is so uneven across tone, terminology, and structure that “checking it over” means a near-complete rewrite. The invoice is the same. The cognitive load is not.

What post-editing actually is in 2026

Post-editing has moved well beyond correcting grammar. It now requires understanding how specific AI models fail — hallucinations, tone drift, domain-specific errors — and working with quality estimation tools that flag suspicious segments at the word level. 

It involves prompting LLMs to produce better drafts before editing starts, and recognizing when a segment should be discarded and rewritten rather than patched. This is strategic, skilled work. 

It is also genuinely different from the creative and interpretive act of translation from sources. Both deserve professional rates and honest job titles.

If you are commissioning localization and planning to send AI pre-translated content for “review,” be upfront about it. Ask what the quality of the MT output is per segment and per language pair. 

Also, ask whether the brief is “post-edit to a publishable standard” or “translate from source”, as well as whether your linguists are being compensated for the actual cognitive load involved. 

Clients who treat post-editing as a bargain-basement version of translation get bargain-basement results. And as the Danish ruling illustrates, the legal and reputational implications of blurring these lines are starting to crystallize. 

At LikeLingo, when we localize your content, a Lingonaut owns the output — not as a corrections officer cleaning up someone else's work, but as the author of the target text.