AI-Generated Content: Is It Okay Not to Disclose It?

Creating Content with AI and Disclosing AI Use Are Different Questions

How would you feel if a photograph in a news report had been generated by AI?

Would you respond in the same way if AI had created a scene in a film?

And would your judgment change if the producer had not told you?

Discussions about AI-generated content often bring together two distinct questions. One concerns whether using AI to create content is acceptable. The other concerns whether consumers should be informed that AI was used.

A 2024 study by Seoungmin Han, Sein Hong, Jaeseo Choi, and Yoonhyuk Jung separates these questions. The researchers examined consumer perceptions of AI-generated content and AI disclosure across different content values and formats.

One notable finding is that even when consumers show little concern about whether AI was used, they may still regard disclosure as a basic expectation.

Comparing News and Entertainment Across Text, Images, and Video

The researchers distinguished between utilitarian and hedonic content values.

Utilitarian value refers to the usefulness of content for obtaining information and making judgments. Hedonic value refers to experiences such as enjoyment and pleasure. In this study, news represented utilitarian content, while films and television dramas represented hedonic content.

The researchers combined these two categories with three formats—text, images, and video—to create six combinations. Their approach allowed for the possibility that consumers might respond differently to AI use in a news article, a film image, or a television drama.

The online survey took place on April 28–29, 2024, with 71 participants who had previously experienced AI-generated content. Of these, 53 participants, or 74.6%, were in their twenties. University and graduate students accounted for 41 participants, or 57.7%.

For each combination, participants were asked how they would respond when AI was used and when it was not, as well as when AI use was disclosed and when it was not. The researchers analyzed these paired responses using the Kano model.

What Creates Satisfaction Is Not Always What Prevents Dissatisfaction

The Kano model examines responses both when an attribute is present and when it is absent. It goes beyond a simple distinction between liking and disliking, recognizing that satisfaction and dissatisfaction do not necessarily develop symmetrically.

An unexpected feature, for example, may delight consumers without causing dissatisfaction when it is missing. The model classifies this as an “attractive” attribute.

Other features are taken for granted when provided but cause dissatisfaction when absent. These are classified as “must-be” attributes.

An attribute is classified as “indifferent” when its presence or absence makes little difference to satisfaction. A “reverse” attribute is one whose presence is associated with a negative response.

Applying these categories helps distinguish whether AI use is perceived as an appealing feature, a production method consumers pay little attention to, or something they would prefer to avoid. It also allows disclosure to be examined as either an added benefit or a basic expectation.

AI Use in News Video Drew Negative Responses

Responses to AI-generated content were generally classified as indifferent. News images and video, however, showed a different pattern.

For news text, indifferent responses were the most frequent, with 39 responses. For news images, indifferent and reverse responses were tied at 29 each. For news video, reverse responses numbered 32, slightly more than the 30 indifferent responses.

Although the negative category was the most frequent for news video, the difference was small. The findings should not be interpreted as showing that all consumers reject AI-generated news video.

Nevertheless, the negative responses to visual news content deserve attention. News photographs and video may be understood as evidence of real events. The researchers discuss the importance of factual accuracy and objectivity in news as a possible explanation for these findings.

The study did not, however, measure the accuracy of actual news videos or participants’ ability to detect false information. It examined consumers’ perceptions of using AI in news production.

Indifference Was the Most Common Response to Films and Dramas

For films and television dramas, all three formats—text, images, and video—were classified as indifferent. Whether AI was used did not emerge as a central factor distinguishing satisfaction from dissatisfaction.

Some responses classified AI use as attractive. However, indifference was the most frequent category across all three formats. This should be distinguished from an active preference for AI-generated films or dramas.

The contrast with news can be interpreted in relation to the value consumers seek from content. A production method may carry different implications when people are trying to understand real events than when they are enjoying a fictional story.

The findings suggest that a single category called “AI content” is insufficient to explain consumer responses. Evaluations may vary according to what the content communicates and the experience it is intended to provide.

Disclosure Was a Basic Expectation in Both Categories

Although responses to AI use differed across content categories, disclosure produced a consistent pattern. For news and for films and dramas, AI disclosure was classified as a must-be attribute across text, images, and video.

This means that disclosure was perceived less as a source of additional satisfaction than as a basic condition whose absence could cause dissatisfaction.

The expectation was stronger for news. The dissatisfaction coefficients calculated by the researchers ranged from approximately -0.72 to -0.76 for news, compared with approximately -0.52 to -0.57 for films and dramas.

A coefficient closer to -1 indicates a stronger tendency toward dissatisfaction when the expectation is unmet. These figures are not percentages of dissatisfied consumers.

Acceptance of AI use and expectations of disclosure can therefore differ. Consumers may pay little attention to the production method itself while still expecting producers to reveal that information.

Here, “basic expectation” describes the consumer perceptions identified in the survey. The study did not determine the scope of any legal disclosure obligation.

Perceived Authenticity Was Also Associated with Different Responses

The researchers conducted an additional analysis of video responses by dividing participants into two groups according to their perceived authenticity of AI-generated content.

Authenticity was measured through participants’ perceptions of whether AI-generated content was authentic, reflected facts, and was trustworthy.

For news video, the most frequent response in the lower-authenticity group was the reverse category. In the higher-authenticity group, it was indifference. For film and drama video, both groups were classified as indifferent.

This suggests that trust in AI-generated content may be related to attitudes toward its use in news production. It does not establish that increasing perceived authenticity will necessarily reduce resistance.

A practical implication is that producers may need to consider consumers’ existing trust in AI content alongside the question of whether and how to disclose its use.

The Small Sample and Classification Method Require Careful Interpretation

This was an exploratory study involving 71 participants. Most respondents were in their twenties, and many were university or graduate students. The results therefore cannot readily be generalized to all consumers.

Participation was also limited to people who had previously experienced AI-generated content, which should be considered when interpreting the findings.

The Kano classification is determined by the most frequent response category. When categories are close—as with the 32 reverse and 30 indifferent responses for news video—the representative classification can obscure the diversity of responses. An indifferent classification does not mean that negative responses were absent.

The customer satisfaction coefficients also require care. Reverse responses are excluded from their calculation. As a result, the coefficients alone may not fully reflect negative attitudes toward AI use.

The distinction between utilitarian and hedonic content is another limitation. People may consume news for enjoyment or obtain useful information from films and dramas. Differences between these genres should not automatically be extended to all utilitarian and hedonic content.

Finally, the study did not experimentally compare the placement, size, or wording of disclosure labels. It cannot establish which disclosure design is most effective.

Content Planning Must Consider Consumer Expectations Alongside Technology

The study’s contribution lies in moving beyond a simple question of whether consumers like or dislike AI-generated content. It distinguishes content values and formats while treating AI use and disclosure as separate issues.

For producers and brands, this provides a basis for considering two questions together: What do consumers expect from this content, and what do they expect to know about how it was produced?

Accuracy and trust may be especially important when content communicates facts. In entertainment, AI use itself may not be central to consumers’ evaluations. Yet disclosure was connected with a basic expectation in both categories.

Alongside deciding what AI can create, content producers must consider how to explain its role to their audiences.