Nieman Lab: Research findings on how newsrooms should label AI use for readers

Researchers Mark Coddington and Tamar Wilner summarized findings from several new studies on how news organizations should communicate their AI use to readers. The piece, published June 17, addresses a question newsrooms have been handling inconsistently: not whether to disclose AI involvement in content, but how to do so in a way that maintains rather than undermines audience trust.

What the research found

The studies found that AI-generated or AI-assisted journalism provokes distrust from readers when labeled in certain ways, and that audiences show a clear preference for content that signals meaningful human involvement. Broad disclosures — “this article was produced with AI” — tend to raise more concerns than they resolve. More specific labels that describe what AI did and what humans verified or decided perform better with audiences.

The research suggests that the language of disclosure matters as much as the fact of disclosure. Framing that positions AI as a tool used by a journalist, rather than a system that produced the content, better preserves the credibility signals readers are looking for.

Why this matters for writing teams

Most editorial AI policy discussions focus on what uses of AI are acceptable internally. This research addresses the outward-facing question of how to communicate those uses, which is increasingly relevant as AI assistance in drafting, summarizing, and translation becomes standard. Teams that have policies in place but haven’t thought about how to describe them to readers are the primary audience for these findings.

The distinction between “AI-generated” and “AI-assisted” is not just a semantic choice — it affects how readers evaluate the accuracy and fairness of the content they’re reading.

Practical implications

Newsrooms currently using boilerplate AI disclosure language should review whether that language is specific enough to communicate what humans actually did in the reporting and editing process. The research points toward disclosures that name the task (transcription, translation, initial summary) and specify what human review followed, rather than generic statements about AI tools being used.

For writers and editors working in AI-integrated workflows, the findings reinforce the value of documenting which steps involved AI and which involved independent human judgment — both for internal accountability and for disclosure purposes.