Labeling headlines generated with artificial intelligence reduces credibility

Madrid. Adding tags indicating that a headline has been generated by artificial intelligence (AI) increases potential readers’ skepticism about its content, even if they are real or human-written headlines, since it assumes complete automation of publishing with AI.

With the arrival of AI tools capable of generating all types of content, whether text or images in a simple way, it is necessary to offer users an informative context about the content they see on the Internet, in order to avoid dissemination of false information and clarify which material is synthetic.

In this sense, multiple platforms such as Meta, YouTube, Google or TikTok, among others, have tools to label content that has been generated or modified using AI functions and visibly inform users of this.

Likewise, it highlights the use of the Universal Internet protocol offered by the Coalition for Content Provenance and Authenticity (C2PA), of which OpenAI, Microsoft, Intel and Adobe, among others, are already part.

Even the use of these AI-generated content labels is one of the guidelines included in the framework of the AI ​​governance law, approved by the government of Spain based on the European AI regulation.

Although its use is necessary to identify AI content, which can be very realistic, it must be taken into account how these brands can impact user perception, reflecting the need to label more specifically so as not to harm high-quality content and legitimate news.

Skepticism increases

A recent study shared by the scientific journal PNAS Nexuswhich investigates the impact of AI-generated labels, found that their use in headlines helps increase skepticism among users, reducing their credibility by assuming that all text has been generated by AI, even if they are real headlines or written by humans.

Specifically, the report was based on two experiments carried out with participants from the United States and the United Kingdom. Study 1 investigated how tagging AI-generated headlines influences both the perceived accuracy of the headlines and participants’ intention to share them, while Study 2 explores the mechanisms causing increased skepticism.

As a result, it was shown that while participants did not equate AI-generated content with fake content, labeling headlines as AI-generated did reduce their perceived accuracy, as well as users’ willingness to share them with others.

Thus, this “aversion” to AI is due to the expectations that the headlines generated with this technology do not have human supervision, so they have no confirmation that they are correct or true.

According to the data, this skepticism was identified among users regardless of whether the headlines offered true or false information, or whether they were created by humans or AI. That is, it was enough to use the AI-generated label to create doubts.

However, the study also shows that the impact of labeling AI-generated headlines was “three times less” than labeling them as false. Therefore, it can be seen that, if it is false content generated by AI, it is more effective to label it as false directly, instead of as generated with AI.

It is not a reliable space

With all this, the findings shared suggest that the use of AI-generated labels “should be approached with caution” to avoid unwanted negative effects, for example, in the event that it is real and harmless content or even beneficial to users, but it is generated by AI.

The conclusions also reflect the need to deploy this type of labeling measures in a transparent manner, clearly explaining what each label indicates, so as not to create confusion among users when reading or viewing content on the Internet.

“More generally, being exposed to AI labels can act as a kind of warning that the online environment is saturated with AI-generated content and is not a reliable space to acquire reliable information,” the study determines.