Enrique Nieto Arranz - Fact Checking Tool

Challenges

Fake news and misinformation can take many forms, so providing tools that address different cases is essential. These tools need to present evidence to help determine whether something is real or not, but they can’t simply give a yes-or-no answer. That’s why human evaluation remains necessary. At the same time, we had to consider that the end users are journalists, often without a technical background, who need to understand how the tools work in order to trust them and confidently use them in their reporting.

Solutions

We focused on improving the most frequently used tools and adding new functionalities to better support journalists’ workflows. The redesign covered the Image Forensic tool, the GIF Comparator, and Twitter Analysis. We introduced clearer explanations, enhanced usability, and improved visual feedback, all without compromising speed or overwhelming users. A key priority was making the technology’s inner workings more transparent to build user trust, while maintaining a fast and efficient experience for both expert and beginner users. Additionally, we delivered a general UI redesign to modernize the app’s overall look and feel.

To tackle the redesign, we followed a focused process that balanced user needs, technical realities, and design goals

Research of the context of use.
Conducted workshops, focus groups, and interviews to understand how journalists verify information and where they encounter difficulties.
Understanding of the technology.
Collaborated closely with the data scientists responsible for the anomaly detection algorithms to understand the underlying technologies.
Design of prototypes.
Developed wireframes and interactive prototypes to explore improved workflows and user interfaces.
Implementation of designs.
Actively participated in the development to bring the designs into the live product.
Validation and adjustments.
Tested the new designs with real users and iterated to address usability issues and refine the solution.

UEQ Questionnaire

First Prototypes of the Tool

Forensic Tool

The forensic tool is one of the most important features of the app. It allows journalists to detect alterations in images that could have been used to spread misinformation. However, the algorithms or filters used to identify these alterations are not 100% accurate and can sometimes produce false positives. This, combined with the complexity of explaining how they work, presented several challenges that we aimed to address with the following features.

Interactive presentation of results.
All filters were displayed together, and to compare them with the original image, users simply had to hover over each one to see the affected areas highlighted in the original image. This helped them easily identify regions that might have been altered.
Clear and informative entry point.
The tool’s entry point was designed to be as user-friendly and informative as possible, setting expectations and clearly communicating its limitations.
Contextual explanations and guidance.
Next to each filter, we provided information such as detection scale, explanations of how the filters work, and tips for identifying real versus false positives. This helped journalists make informed decisions and include any necessary context in their reports.

Gif showing the use of the filters in the forensic tool

Screenshot showing the initial screen of the forensic tool

Example of explanations for the forensic tool

Gif Comparator Tool

Once alterations were identified, journalists needed a tool to display the results in a clear and visual way. Simply showing the original and altered images side by side, or highlighting the altered area, was often not effective, if the alteration was subtle, it could be difficult to spot the differences. That’s why we introduced the GIF Comparator Tool, which allowed journalists to upload the original and altered images, and automatically generated a GIF that overlapped both, switching between them every few seconds and showing only the altered part. This proved especially useful for their web reports, as they could publish the GIF and allow readers to easily see the detected alterations.

Screenshot showing the Gif generator tool

Twitter analysis tool

Twitter is often a place where a lot of misinformation is created and shared. However, when a tweet becomes popular, a large volume of comments and related content is generated, and Twitter does not provide tools to easily analyse all that information. That’s why we introduced the Twitter Analysis Tool, where journalists could simply paste the link to a tweet and receive a detailed analysis. Insights such as the types of comments people were posting, the most active users, and the links being shared helped journalists examine the conversation and identify any false information that other users might have already pointed out.

Other Tools

Other more experimental tools were also developed. Deepfakes have become a new trend in fake news and misinformation, and they are very difficult to detect because the entire image is AI-generated, making it nearly impossible to spot alterations using traditional forensic tools designed for manual edits. However, some experimental algorithms became available, and we provided an easy-to-use tool to explore them for both images and videos.

Foreign languages can also be a challenge for journalists, especially when they involve different alphabets, and extracting text for translation becomes difficult when it’s embedded in images where it can’t be easily copied. To address this, we developed an OCR tool specialised in translations. The tool detects any kind of text in an image, attempts to identify the language, provides the extracted text in plain format for easy copying, and suggests a translation into a preferred language. This greatly streamlined the process of analyzing this type of information for journalists.

Screenshot showing the deepfake detector for images

Screenshot showing the deepfake detector for video