An AI Label Can Change How Art Feels

Imagine finding a small painting that makes you stop scrolling. You like the balance of light and shadow, and the figure at its centre reminds you of someone. Then you read the caption: AI-generated. Would your response change? If it did, which part of your judgement would you revise?

This question matters for creators presenting their work, customers choosing commissions and readers trying to understand their own reactions. A useful discussion needs room for the experience of the work and the importance of its origins. Research on art and poetry offers a way into that discussion.

What changed when researchers changed the label

In a study published on 4 July 2023, Lucas Bellaiche and colleagues presented AI-created paintings with randomly assigned human-created or AI-created labels. The first study analysed responses from 149 participants; a second analysed 148. Across the studies, participants generally evaluated human-labelled images more positively on liking, beauty, profundity and worth, even though every image had actually been generated by AI.

The controlled label comparison is the revealing feature. Different judgements could emerge without a corresponding difference in the image. However, the study measured reported evaluations, including perceived worth. It did not demonstrate what collectors would actually pay in a sale, or how every audience would respond.

Poetry revealed a similar tension

Brian Porter and Edouard Machery’s paper published on 14 November 2024 compared poems by established poets with poems generated using ChatGPT 3.5. In its evaluation experiment, 696 US participants saw the same ten poems, five from each source. Groups were told the poems were human-written, AI-generated or given no authorship information. Overall quality ratings were lower under the AI framing than under the human framing. Yet the actual AI-generated poems received higher overall quality ratings than the selected human poems.

These findings concern particular poems, readers and the model used in that study. They do not rank all poetry or establish what expert readers, other languages or future audiences would prefer. For a creator, the useful question is how much feedback concerns the words themselves and how much concerns the account of their production.

Decide which judgement you are making

My practical suggestion is to give creative feedback three separate questions. These are editorial recommendations for clearer conversations:

• The work: What do you notice in its language, rhythm, composition or colour? Which specific choices support your response?
• The process: What do you understand about who made which decisions, and how does that matter to you?
• The relationship: What were you expecting from this artist, publication, gift or commission?

You might enjoy a piece while deciding that its production process does not match what you want to support. You might value the maker’s story while finding the composition unsuccessful. Both positions can be explained precisely. Calling the entire response “good” or “bad” makes it harder to identify the decision being made.

A poster review with two useful answers

Consider a hypothetical poetry event organiser reviewing an illustrated poster. The poster has a strong central image, but the event date is difficult to read. Its caption accurately explains that the illustration was AI-generated and that a designer arranged the typography and layout.

One reviewer objects to the use of generated imagery because the event is intended to showcase local illustrators. Another likes the atmosphere but wants the date enlarged. The organiser has two separate decisions: whether the production approach fits the event’s purpose, and whether the layout communicates the necessary information.

A clearer review would preserve both comments. If the organiser commissions a local artist, the date may still need enlarging. If the organiser keeps the generated image, the concern about the event’s purpose still deserves an answer. Agreement on one issue does not settle the other.

Ask for feedback that names its reasons

When you ask someone to review a piece, give them a specific task. For a poem, invite comments on a particular image, a shift in voice or an ending that feels unfinished. For an illustration, ask about the focal point, visual balance or suitability for its intended setting.

Then invite a separate response to the process information. Useful questions include:

• Which detail of the making process affects your response, and why?
• Does the description leave you uncertain about any human or generated contribution?
• Does this process fit what you expected from the project?

Keep the accurate disclosure available. The purpose of these questions is to distinguish reasons within a critique. Ask reviewers to connect each recommendation to something they observed or to a value they want the project to uphold. A small feedback session can guide revisions; it cannot establish what an entire market believes.

Give people a process they can understand

As a creator, consider what a short explanation needs to convey. A broad AI label may leave your reader wanting more information. Identify the contribution that matters to this piece: generating an illustration, suggesting alternative lines, selecting elements, redrawing a figure or arranging the final layout.

For example, a hypothetical process note might explain that an artist supplied an original sketch, generated background options and repainted the foreground. Use that description only when it accurately reflects the work. Keep the wording proportionate to the project and its audience.

The MaryChuks authorship ledger guide offers a fuller approach to recording creative decisions. For the viewer, a concise explanation can point to those records without turning every caption into a production diary.

A process note should help someone understand what happened. It cannot require them to admire the result or accept the workflow. Leave space for informed disagreement, including a preference for work made through a different process.

Notice what your first reaction leaves unanswered

As a reader or viewer, try naming one feature you appreciate, one question about the process and one expectation you bring to the encounter. You can then explain a changed opinion without dismissing your earlier response or pretending that origins have no importance.

For creators, the same discipline makes criticism more usable. A comment about imagery suggests a different revision from a concern about attribution or the purpose of a commission. Keep those reasons visible. They give the conversation somewhere to go after the label has been read.

Sources

• Bellaiche and colleagues, Humans versus AI: whether and why we prefer human-created compared to AI-created artwork, Cognitive Research: Principles and Implications, 4 July 2023.
• Porter and Machery, AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably, Scientific Reports, 14 November 2024.