kultura42.hu / Articles / Checklist

Labelling AI-Made Artworks in Culture Coverage: A Checklist for Journalists

A practical guide for journalists to transparently document AI involvement in creative works while maintaining editorial integrity and rights standards.

By Zoltán Varga2026-11-055 min read

Ask About the Production Process

Start every review or feature by asking the creator how the work was made. Do not assume the final output tells the whole story. A podcast might use AI for script drafting but rely on human voice acting and editing. A painting could begin with a generated base image that an artist then paints over by hand.

Document the specific stages where artificial intelligence intervened. Did the tool suggest plot points, generate background textures, or assemble a music track? Gerard Bisbal, creator of the NUVAGAIA transmedia universe, uses different models for distinct tasks. He employs ChatGPT for worldbuilding organization and DALL-E 3 for visual concepts, keeping the narrative coherence under human control.

Avoid vague terms like 'AI-assisted' without further detail. Instead, specify the exact contribution of the technology. State whether the AI acted as a brainstorming partner, a draft generator, or a final renderer. This level of granularity allows readers to understand the actual creative labor involved. It also prevents the conflation of entirely automated spam with genuine hybrid artistry.

Verify the timeline of creation if possible. Note when the human creator reviewed, edited, or rejected the initial outputs. Bisbal discards a significant portion of AI suggestions to maintain consistency within his fictional universe. This filtering process is a critical part of the work that distinguishes professional creation from raw generation.

Name the Human Role Explicitly

Identify the human creator and their specific function in the project. Are they the director, the editor, the prompt engineer, or the final artist? The value of the work often lies in the decisions made by the human, not the raw data produced by the machine. Leontius Manovich notes that true AI-driven culture has not yet arrived because humans still decide what gets shown to audiences.

Clarify who holds the responsibility for the final content. Under current regulations, a natural person must accept editorial liability for AI-generated material to avoid mandatory machine labeling. This means a journalist or curator must be willing to stand behind the work as if they wrote or created it themselves. The human role is the anchor of trust in the piece.

Distinguish between the creator of the concept and the operator of the tool. Sometimes the person typing the prompts is not the original idea generator. Other times, the artist uses the tool to execute a vision they fully conceived. Both scenarios require clear attribution. Do not credit the software as the author unless the work is entirely autonomous, a state that remains theoretical for most cultural products.

Highlight the skill required to manage the AI effectively. Operating a complex stack of tools requires technical literacy and artistic judgment. Bisbal treats his AI tools as an exploratory workshop rather than an autonomous creator. This framing emphasizes that the human architect directs the process, ensuring the output aligns with a coherent vision.

Describe the Tool's Specific Role

List the specific software or models used in the creation process. Mentioning 'an AI' is insufficient for accurate reporting. Specify whether the team used a large language model for text, a diffusion model for images, or a text-to-speech engine for audio. Different tools produce different artifacts and carry different ethical implications.

Explain how the tool was integrated into the workflow. Was it used for rapid prototyping, final rendering, or post-processing? In the NUVAGAIA project, Gemini handles podcast scripts while DALL-E 3 generates visuals. This specialization shows that no single tool does everything. Describing this division of labor helps readers appreciate the complexity of the production.

Note any limitations or known biases of the chosen tools. Some image generators struggle with hands or text. Language models may hallucinate facts or mimic specific styles without permission. Acknowledging these constraints demonstrates editorial diligence. It shows that the publication understands the medium and its potential pitfalls.

Avoid treating the tool as a black box. Explain what inputs were provided and what outputs were selected. Did the creator iterate on prompts multiple times? Did they combine results from different models? Transparency about the inputs and selection process provides a clearer picture of the human effort involved than simply stating 'made with AI'.

Avoid Hype Words and Binary Labels

Reject terms like 'revolutionary,' 'game-changer,' or 'fully automated.' These words obscure the reality of the work. Most cultural products today are hybrids of human intent and machine execution. Labeling them as purely one or the other misleads the audience and devalues the actual labor performed.

Do not rely on AI detection software to verify authorship. Current detectors fail frequently, especially with hybrid content where humans edit AI drafts. Studies show that detection accuracy drops significantly when text is modified by a human. Relying on these tools creates false confidence and risks publishing incorrect attributions.

Focus on the quality and context of the work rather than the method of production. A story is valuable if it resonates, regardless of whether a computer helped write it. The debate should center on transparency and ethics, not on policing the presence of technology. Hype distracts from these substantive issues.

Resist the urge to categorize works as 'AI art' or 'human art.' These categories are increasingly meaningless as workflows merge. Instead, describe the contribution of each party. This approach respects the nuance of modern creation and avoids the polarization that often surrounds new technologies in the arts.

Establish Provenance and Evidence Standards

Require creators to provide documentation of their workflow when submitting AI-assisted work. This might include prompt logs, version history, or statements of intent. Such records serve as evidence of the human role and the extent of machine involvement. They protect both the creator and the publication from future disputes.

Understand the legal requirements for disclosure. As of August 2026, the EU AI Act mandates that providers of generative AI systems must label their output. However, if a human editor exercises control and accepts responsibility, the strict machine labeling requirement may be waived. Publications must know these rules to remain compliant.

Recognize that technical provenance markers like C2PA are not yet foolproof. Security analyses indicate that metadata can be altered or stripped during upload to social platforms. Therefore, human verification and editorial statements remain more reliable than digital watermarks alone. Trust must be built through transparent communication, not just code.

Build a 'non-slop zone' by prioritizing verifiable origins over volume. The goal is not to ban AI but to ensure that every piece of content has a known source. Readers pay for trust and accountability, not just for the absence of technology. Clear provenance standards elevate the quality of cultural journalism and media coverage.

AI-assisted: drafted with a language model from the author's research notes, then edited and checked before publication. Author: Zoltán Varga, AI visibility researcher and AI ecosystem architect, Budapest.