Press Realease: The FARE_DSA project will audit search engines and large language models for political recommendations and study possible risks for democracy, in line with the EU’s Digital Services Act.
The European Research Council (ERC) has awarded a prestigious Proof of Concept (PoC) grant to Joana Gonçalves-Sá, a researcher at LIP - Laboratory of Instrumentation and Experimental Particle Physics and NOVA LINCS, the Research Center of the NOVA School of Science and Technology, to lead the project FARE_DSA – A Framework for Auditing Recommendation Engines under the DSA.
FARE_DSA will develop and test a novel pipeline for auditing algorithmic platforms — specifically search engines (SEs) and large language models (LLMs)—that are increasingly central to how people access information online. The project will be the first to offer a concrete, policy-aligned tool for implementing Article 40 of the European Union’s Digital Services Act (DSA), which grants vetted researchers access to non-public platform data to assess systemic risks, such as election interference or political manipulation.
“Platforms such as Google and ChatGPT have become central gateways to information,” said Dr. Gonçalves-Sá. “But we are far from understanding how their algorithms might shape public opinion. During the 2024 EU parliamentary elections, we found that both search engines and LLMs consistently showed content biased toward one side of the political spectrum. With this project, we aim at testing if such biases exist at scale and, if so, why. This can only be done if we work both with regulators and with the platforms themselves, to understand root causes and implications”.
The project builds on prior ERC-funded research, namely the ERC-StG FARE – Fake News and Real People (launched in 2020 with €1.5 million in funding) and PoC FARE_Audit grants, which developed a suite of web-crawlers capable of simulating users with specific traits (e.g., country, age, gender) and analysed how platforms respond to their queries. These tools offered strong preliminary evidence that even slight differences in browsing history or profile can lead to significantly different and sometimes biased results. FARE_DSA aims at enhancing this methodology with access to real platform data under the DSA, offering unique insight into how personalization and monetization may distort information exposure, especially during critical periods such as elections.
“Our team has spent years developing ethical, privacy-preserving tools to audit algorithmic systems without collecting personal data,” Dr. Gonçalves-Sá added. “But to fully understand what is going on behind the scenes — how results are ranked, when and why certain links appear — we need access to non-public data. Article 40 opens the door for exactly that. This is a crucial time for research on algorithmic transparency. This grant will help us pave the way, and hopefully, help others follow.”
The project will contribute to clarifying what kinds of data are available, under what formats, and under what terms they can be accessed, breaking a longstanding transparency deadlock between researchers and platforms. FARE_DSA will also produce actionable policy insights, host training workshops for academics and NGOs, and contribute to other audits of Very Large Online Platforms (VLOPs) and Search Engines (VLOSEs), to be carried out by European regulators, NGOs, journalists and other relevant stakeholders.
Although the ERC Proof of Concept scheme is typically associated with commercial innovation, FARE_DSA highlights its potential for high-impact social innovation, bridging cutting-edge research, digital transparency, and policy implementation.
The project was supported with a 150k€ and will run for one year.