AI la­bels re­duce so­cial me­dia en­gage­ment

When companies, organisations and influencers use AI to create social media content, transparency comes at a cost.

A new study from Copenhagen Business School shows that users engage less with posts labelled as AI-generated or AI-enhanced than with posts created by humans. This applies both to their emotional response and their willingness to like, share or otherwise interact with the content.

The researchers also find that the effect is strongest for emotionally driven content. When a post aims to create a sense of connection, identification or emotional engagement, users respond more negatively if they know AI has been used.

“Across our studies, we found that engagement declined whenever content was labelled as AI-generated, even though the content itself was identical, and the decline was substantial for fully AI-generated content. AI-enhanced content also saw lower engagement, but the effect was significantly smaller,” says Michael Wessel, Associate Professor at CBS.

The study highlights a specific dilemma for companies and platforms: Being transparent about AI may be essential for credibility and responsible communication. However, the same transparency can also make your content less effective.

The study has just been published in the journal Electronic Markets, and has been conducted by Freya Seeger, Michael Wessel and Christiane Lehrer from the Department of Digitalization at Copenhagen Business School.

Emotionally driven content affected the most

The researchers compared responses to rational, information-based content as well as emotionally driven content. The results showed a clear difference.

Rational content might include a post that objectively presents a new product, explains a feature or provides practical information. 

However, emotionally driven content could be a travel influencer sharing a beautiful sunset, a personal moment or an atmosphere designed to make the users feel something and imagine themselves in the experience.

While users reacted less negatively to AI in rational posts, emotionally driven content was affected far more by the AI label. According to Michael Wessel, this reflects what people expect from social media.

“The effect was clearly stronger for emotional content than for more rational or informative content, which makes perfect sense because social media is, at its core, social. People use these platforms to connect with other people, so they react more strongly when the human element appears to disappear. At the same time, we are already becoming accustomed to using AI services such as ChatGPT and Google's Gemini for factual information.”

This suggests that users are not only judging the content itself, but also what the label says about authenticity and human presence.

AI as a tool is received better than AI as the creator

The study also shows that users distinguish between AI-enhanced and AI-generated content.

In other words, it matters whether AI simply helps refine a post or whether it is the primary creator. According to Michael Wessel, the research findings suggest that users are much more comfortable with AI as a tool than as the author.

“Users seem far more willing to accept AI as a tool than as the actual creator. As long as there is still a person behind the content and AI is simply used to refine it, the social contract is not broken to the same extent.”

This is an important distinction because much of today's content does not fit into a simple either-or category. In practice, much content is now created through collaboration between people and machines.

“Much of the existing research relies on a binary distinction between AI and humans, but that is no longer how the world works. Today, much content is created by people and then edited or enhanced using AI, and we wanted to capture that reality.”

Timing of AI labels also matters

The researchers also examined whether it makes a difference when users see an AI label and found that timing can change how people respond to AI-enhanced content. If users first experience the content and are only later told that AI helped enhance it, their reaction is less negative than if the label appears from the outset.

For fully AI-generated content, however, this strategy does not appear to make a difference.

“Timing matters for AI-enhanced content. If users first experience the content and only afterwards learn that AI helped refine it, they respond more positively than if they see the label immediately, but for fully AI-generated content, revealing this later does not really help. It is still penalised heavily.”

A possible explanation is that people respond differently to discovering that AI has assisted a human creator than to realising that content was created by AI, which may leave them feeling misled.

“Users may feel deceived when they only later discover that something was entirely AI-generated. With AI-enhanced content, the reaction is different because the human contribution is still clear.”

A dilemma for companies and platforms

According to Michael Wessel, the study highlights a growing challenge in digital communication.

On the one hand, expectations for transparency are increasing as companies and platforms make greater use of AI. On the other hand, the study shows that transparency itself can reduce engagement.

The dilemma has become even more relevant since new transparency requirements under the EU AI Act took effect on 2 August 2026. Among other things, the rules require certain types of content created or manipulated using AI to be clearly identified.

So what can organisations learn from these findings?

“In the short term, transparency may come at the cost of lower engagement. If content relies heavily on authenticity or emotional value, the use of AI can weaken the very quality of the content. In those cases, it becomes a genuine question of how much authenticity is needed.”

At the same time, he believes the trend could also create a countermovement. As social media becomes increasingly saturated with cheap, rapidly produced AI content, human-created content may become more valuable.

“Because producing content has become so inexpensive, social media is increasingly being flooded with low-quality AI content. Ironically, that may make genuinely human-created content more valuable and help it stand out.”

Platforms often make AI labels difficult to spot

The study examines how users respond when they notice AI labels. According to Michael Wessel, however, it is equally important that many platforms deliberately design these labels to be difficult to see.

“Platforms understand very well how design influences behaviour, and many AI labels today are easy to overlook. In practice, this allows platforms to meet transparency requirements without making it likely that users will notice the label.”

According to him, this reflects a broader strategic tension between transparency and performance.

“Some platforms have moved towards clearer labelling and a more proactive approach to AI-generated content, while others have done the bare minimum and kept labels as discreet as possible. That reflects a broader strategic tension between transparency and engagement.”

Meta, TikTok and YouTube are taking different approaches

The differences become clear when looking at the largest platforms. According to Michael Wessel, Meta has long adopted a highly discreet approach with labels so easy to miss that they rarely become a meaningful part of the user experience. He points to Meta's own figures, which, according to him, show that only around 0.006% of users engaged with AI labels.

TikTok has taken a more proactive approach. The platform requires labels on realistic AI-generated content, also applies automatic labels and is testing tools that give users greater control over how much AI-generated content appears in their feed.

YouTube has also stepped up its efforts. In May 2026, the platform announced that AI labels would become much more prominent. On standard videos, they will appear directly below the player above the description, while in YouTube Shorts they will be displayed as an overlay on the video itself.

According to Michael Wessel, these differences show that platforms are still searching for the right balance between transparency and protecting engagement.