Dataharvest 2025 - the European Investigative Journalism Conference

AI has moved from niche tech coverage to leading headlines worldwide. Opaque algorithms make life-changing decisions about millions across the continent in health, banking and social security systems. Meanwhile, Big Tech increasingly uses Europe as a base for the resource-guzzling data centres that power modern generative AI development. So where do you start? This hands-on session will walk through a framework for different story archetypes and an understanding of how to generate and pursue story leads. Then, we'll home in on a specific story and how to build an AI accountability investigation from scratch — including following the money behind digital infrastructure, using freedom-of-information laws and collaborating with data scientists to uncover algorithmic bias. Participants will leave the session with an understanding of how to investigate automated systems even when access to code is difficult. Participants who want to go deeper can attend the Level 2 session.

Investigating algorithms is not just about obtaining code. On the contrary, there's so much we can learn from an automated system – whether Al-driven or rule-based – by looking the impact it has on individuals, minorities and marginalized communities. We can examine the people involved in its development and deployment; the companies who profit; the data it generates; and the weight given to the parameters it relies on to make decisions. How do we find this data and approach the investigation? This session is aimed at journalists with previous experience in reporting about automation and AI and those who attended our beginner level session (Saturday, 0930). In this practical workshop we will explore two real, successful cases of investigative algorithmic accountability reporting. These investigations uncovered how automated systems are used for welfare allocation in different European countries. This will be a hands-on session – don't expect ready-made answers. Instead, all participants will start with the same, limited information about the case. Together, we will explore the different methodologies and approaches we can use to uncover how an algorithm works, even when we have little data to start with. During the workshop, we will aim to obtain as many techniques and methods as possible and inspire new investigations.

In this session, we will show the methodology developed by The Guardian to gather and measure content published on TikTok during the UK election campaign. Using this methodology, we revealed how far-right Reform UK leader, Nigel Farage, outperformed all other parties and candidates on the platform. We also exposed the political content shared during the campaign aimed at young audiences.
During the session, we will also show you how to set up accounts located in countries different to the one where the reporting team is based. We will also show the scraper for automatically gathering content associated with particular hashtags at various times over several days. TikTok is relatively hostile to scrapers, forcing us to resort to advanced scraping techniques including residential proxies. We will demonstrate how we identify the main content creators, using some metrics from TikTok as well as the number of videos recommended by the algorithm, and how we used the algorithm manually to find more related videos - and therefore Tiktokkers - similar to those we were investigating.