@inproceedings{01HGAM3MY8P48B517R3RTW3PC2,
  abstract     = {{The advancement of technologies has made it increasingly easier to manipulate images. Recently, Adobe added a feature that allows to change images using textual prompts. These new technologies facilitate manipulated images as they allow users to realistically add or remove objects in an image without the need for advanced technical skills. As a result, the distinction between fake and real images is blurring (Khan et al., 2023; Qian et al., 2023). Consequently, growing concerns about the veracity of online messages and information have come to the fore in public debate. The study aims to investigate how audiences assess the credibility of information on Instagram, by studying the impact of manipulated images, source and verification label.

To date, most images are manipulated (or generated) for artistic or humorous reasons (where no harm was intended), however, other images are meant to show reality (GCFGlobal, 2023). Here, manipulation can create big problems, especially when shared on social media, where images often lack context. The main concern is that manipulated and false images will be used for disinformation or misinformation purposes. As more and more people rely on social media for their information (e.g., Newman et al., 2023), these manipulated images are deemed to result in widespread misinformation. Users are faced with the challenge of determining what images and information are credible and which are not (Amaral & Silveira, 2018; Farkas & Schou, 2018; Heuer & Breiter, 2018). What remains unclear is how audiences evaluate online images on social media in relation to different manipulation techniques. To test users’ visual literacy and credibility assessment of Instagram messages, we conducted a user experiment in which we asked users to rate messages on several criteria related to credibility and trust.

The purpose of this study is twofold. First, we want to measure the extent to which participants were able to detect different manipulation techniques. Secondly, we want to measure the effects of the source and the presence of a verification label on the credibility of Instagram messages.

Using a 232 between-subject design, participants were exposed to an Instagram post that stated Flanders is suffering from flooding due to climate change. This post varied among participants in terms of(1) manipulation techniques (i.e., a photoshopped image or AI-generated image), (2) the source sharing the information (i.e. the public broadcaster, a well-known voluntary association focussing on nature conservation, and a troll-like account), and (3) the presence of a verification label for the source (i.e., with or without blue check). The participants were asked to rate the shown image as being believable, authentic, fake and manipulated.

The results show that participants believe information shared by trustworthy sources such as the public broadcaster or voluntary association more than troll-like accounts. As expected, information from a troll-like account scored much lower in terms of message credibility. Remarkably, the control image – without the Instagram interface – was found to be more credible still, hinting that audiences are sceptical toward all information on social media, even from trustworthy sources. Furthermore, the study found no difference between messages with and without a verification label. As such, so-called verified "blue check accounts" aren't more credible than accounts without a verification label. Related to the manipulation technique, the AI-generated images score better in terms of message credibility, compared to posts with a photoshopped image.}},
  author       = {{Mareen, Hannes and D'haeseleer, Stephanie and Van Damme, Kristin and Evens, Tom and Lambert, Peter and Van Wallendael, Glenn}},
  booktitle    = {{MISDOOM 2023, the 5th Symposium on Multidisciplinary International Symposium on Disinformation in Open Online Media, Abstracts}},
  language     = {{eng}},
  location     = {{Amsterdam, the Netherlands}},
  pages        = {{2}},
  title        = {{COM-PRESS : an image manipulation analysis dashboard for fact-checkers}},
  url          = {{https://event.cwi.nl/misdoom-2023/}},
  year         = {{2023}},
}