Human Judgment Remains Non-Negotiable: How Editorial Teams Use AI Without Outsourcing Responsibility

When the European Union's 2024 AI Act's transparency provisions came into force in August 2026, they introduced a significant requirement: disclosure of AI-generated text on matters of public interest. However, the regulation carved out an important exception for content that "has undergone human review or editorial control" and for which a person or organization bears editorial responsibility—meaning accountability for published material.

This distinction matters deeply. At Nielsen Norman Group, artificial intelligence plays a role throughout the editorial workflow. While AI may generate language and contribute concepts to articles, humans retain ultimate authority over editorial decisions. Experts with domain knowledge—not machines—have reviewed the material and vouch for the accuracy and value of what appears in print.

The Editorial Workflow at NN/G

Since 2013, the editorial team has maintained a demanding, cyclical approach to publishing. An author proposes a subject, and editors assess whether the organization already covers it, whether the field requires the perspective, and whether the author brings a distinctive angle or organizing principle unavailable elsewhere.

Revision rounds follow. Rarely does an initial manuscript reach publication without substantial changes. Two editors typically examine logical flow, factual accuracy in user experience contexts, and prose quality. The cycle repeats until authors and reviewers reach consensus on a publishable version. Some pieces undergo two or three rounds; others require more. The rigor mirrors academic journals rather than typical blog publishing.

Everyone—visiting contributors and senior staff alike, whether new or with decades of tenure—follows identical standards. The process itself has not shifted since generative AI arrived, though editors report meaningful efficiency gains.

Where AI Improves Clarity

The author pool spans varied writing abilities. Many lack professional writing training. Some, including editors themselves, write in English as a non-native language. Previously, editors invested substantial time restructuring paragraphs for brevity and comprehension. Tools like Copilot in Microsoft Word and Grammarly now assist in condensing passages while maintaining or sharpening meaning.

Yet AI revisions can obscure reasoning or overlook subtle distinctions while trimming length. Editors must judge whether the machine-suggested version genuinely improves the text.

AI Handles Rigid Formatting Constraints

Every NN/G article opens with a summary of fixed length—currently 160 characters, though this threshold has shifted over time. Crafting a paragraph that fits precisely within such bounds demands skill. Most writers overlook character limits or miscalculate space counts. AI excels at generating summaries meeting exact specifications.

Summaries represent just one rigid structure. Study guides and glossaries impose their own constraints. Study guides, for instance, link to articles with descriptions following particular patterns—complete sentences with active verbs, for example. AI readily transforms author drafts into these standardized formats.

Repurposing Content Across Platforms

NN/G distributes material through self-paced courses, reports, videos, and podcasts. Occasionally, valuable material from one medium—a report finding, a course topic, a video segment, a podcast discussion—warrants adaptation into an article. AI can extract pertinent information and reshape it for the new medium. A human editor always performs the final review, but AI handles much of the mechanical work involved in reformatting.

AI as a Tool for Critical Thinking

Beyond rewriting and formatting, AI supports editorial judgment itself. It can scrutinize both the evidence backing an article and the coherence of its reasoning.

For instance, AI can assist in fact-checking and validating assertions. When an article references a study or legislation, generative AI can flag potential misstatements or overreaching claims warranting investigation. The opening sentence of this piece underwent multiple refinements with AI assistance to ensure it accurately captured the EU AI Act's specifics.

AI does not, however, verify correctness. When it identifies a potential problem, editors or authors must consult original sources, verify the claim, and fix any errors discovered.

Similarly, AI can help with argument-level analysis. Spotting logical flaws—contradictions, unsupported assumptions, unstated premises—ranks among an editor's most challenging and time-intensive duties. While human editors will continue performing this work at NN/G, AI can provide assistance.

An earlier version of this article centered on the principle that outsourcing thought to AI should be avoided. Yet the same draft elsewhere described AI as a valuable thinking partner capable of participating in reasoning. When the editor asked ChatGPT to critique the draft, the system identified this tension. That feedback prompted sharper articulation of the final principle.

This pattern reflects typical AI use in editorial work: the editor shares their own assessment and asks AI to challenge or expand on it. Sometimes AI identifies overlooked issues; sometimes it questions the editor's judgment; sometimes it suggests changes the editor rejects. In all cases, the editor weighs suggestions rather than accepting them as final. Recommendations that strengthen the argument and seem applicable are retained; others are discarded.

Editorial Judgment Cannot Be Delegated

Critics caution that assigning writing to AI amounts to outsourcing thought itself. Writing offers an opportunity to examine ideas, organize them, and through that process uncover weaknesses and strengthen arguments. It represents a chance to develop and exercise judgment.

Thinking, however, occurs through multiple channels. It happens via sketches, conversations with colleagues, dictating unstructured ideas to an AI system, or prompting. Prompting itself can constitute part of the thinking process.

NN/G does not mandate how authors should compose or reason. They may work through arguments independently, collaborate with peers, dictate thoughts for AI to rephrase, or use AI as a thinking partner. What matters is that a human specialist has performed the intellectual labor required to comprehend, assess, and take responsibility for the article.

AI may participate in writing, but NN/G does not transfer editorial responsibility to it and does not list it as an author. Human specialists determine whether an idea merits publication, whether reasoning holds up, and ultimately whether NN/G will attach its reputation to the work.

The organization's AI content policy itself demonstrates this principle in practice. AI generated much of the policy using this article as source material; humans then reviewed, revised, and approved the final version.