My latest article, co-authored with Marilia Gehrke and Carl-Gustav Lindén and published in Journalism, examines how European codes of ethics do, or do not, regulate data journalism practices. Based on an analysis of 37 codes from 32 countries, this research shows that fundamental journalistic principles, such as accuracy, fairness, and transparency, are largely present in these texts. However, they remain formulated in general terms and offer few guidelines to support methodological choices specific to working with data, whether in collection, processing, analysis, or presentation.
A blind spot in journalistic ethics
The ethics of data journalism remain largely unexplored, probably because it is generally assumed that traditional ethical principles naturally apply to data-driven practices. However, professional practice is challenging this view.
Firstly, data is often perceived as objective and neutral. Yet it is produced within specific social, institutional and political contexts and serves particular purposes. Furthermore, processing it inevitably involves human decisions. One of the main ethical challenges of data journalism lies precisely here: data is often used to bolster the credibility and accuracy of information, even though its use relies on a series of interpretive choices that can influence the results.
Secondly, any data-based work involves a series of complex operations, from data collection to visualisation, including data cleaning, cross-referencing, and analysis. Each step can introduce specific risks to the quality of the information: biases in the datasets, measurement errors, or interpretation choices that often remain invisible to the public.
None of these choices is neutral or objective, since they rest on human decisions that require explanation, control, and critical examination. Simpson’s paradox demonstrates the interpretive nature of data analysis: trends observed in multiple data sets can reverse when combined, leading to significantly different conclusions based on methodological choices.
Moreover, although transparency is a value shared by most European codes of ethics, the concept is primarily associated with identifying sources, transparency regarding media ownership and managing conflicts of interest. Methodological transparency, which involves making the origin of data, cleaning criteria, analytical choices and their limitations visible, remains largely absent.
The recently updated codes in Flanders, Belgium, Kosovo, Serbia, and the United Kingdom ( Impress ) now include provisions on transparency in the use of AI. These provisions focus primarily on labelling automatically generated content and ensuring human oversight. This is a step forward, but it leaves out upstream data practices, where many of the risks associated with AI originate.
Epistemic governance, a bridge to AI ethics
Few of the examined codes in this study explicitly addressed data journalism, aside from a few specific provisions: the presentation of polls and opinion surveys (sample, sponsor, time period), the accuracy of graphs and illustrations, or the attribution of data from third-party sources. None, however, address the quality of the datasets, their biases, or the choices made during analysis. This lack of reflection on the conditions of knowledge production is all the more problematic given that these methodological choices directly influence the quality and credibility of the information produced.
To overcome this limitation, we propose to consider the ethics of data journalism as a form of epistemic governance, that is, as the set of rules, procedures and mechanisms that structure the production of journalistic knowledge.
While traditional ethical approaches emphasise individual responsibility, epistemic governance focuses on the conditions in which methodological choices are made, discussed and monitored. This perspective treats ethics as a requirement embedded in professional practice, not just a set of values or principles. Therefore, documenting methodological choices, collectively reviewing analyses, and sharing responsibility within teams become essential to guaranteeing the quality and reliability of information.
This process-oriented approach is particularly well-suited to AI-assisted journalism, where risks are distributed through automated chains. The probabilistic nature of AI systems amplifies risks that often originate in the data itself. Therefore, concepts from AI ethics, such as explainability and algorithmic accountability, can inform journalistic ethics, provided a common language is established, since accuracy, transparency, and objectivity do not necessarily mean the same things in journalism as in data science or AI engineering.
Corpus, code, and annotations available on OSF: https://osf.io/m4g9r/overview
Dierickx, L., Gehrke, M., & Lindén, C.-G. (2026). From normative principles to epistemic governance: A workflow-based analysis of European journalism ethics in data-driven environments. Journalism . https://doi.org/10.1177/14648849261492006
Original version in French.
Automatically translated with the help of Google Translate.
Humanly post-edited.
