The next speaker in this session at the 2026 Association of Internet Researchers conference in Mexico City is the great Bruna Paroni, whose focus is on the 2018 and 2022 presidential elections in Brazil; the former elected the far-right anti-establishment candidate Jair Bolsonaro, while the latter saw the return of leftist Lula.
This same period also saw significant changes to the Facebook algorithm, from a downranking of news during 2018 through the emergency measures enacted after the 2020 US presidential election and coup attempt to a 2022 reduction of ‘political’ content (by some nebulous in-house definition) in users’ personal feeds. These algorithmic changes form the backdrop to this research.
The project collected some 460,000 URLs shared on Facebook in August to October 2018 and 2022, respectively, via the Facebook URL Shares dataset; of these, some 220,000 were identified as political via a fine-tuned LLM; in these, the analysis identified some 400 distinct narratives (200 per election), also via LLM; and finally, outlets were scored on a left-to-right scale; and these narratives were finally located on that left-right scale based on the average of the positions of the URLs involved in them.
A comparison across both elections shows that from 2018 to 2022, the political centre has shrunk: from 74% to 59% of narratives were still centrist, while right-wing narratives grew from 3% to 17% of the total. Left-leaning narratives accounted for some 23-24% in both cases. In 2022, only some 29% of the 2018 volume of political URLs were still present, and only 6% of mainstream URLs.
In 2018, 83 views were needed to produce one share; in 2022, this grew to 370 views. In 2018, each new share produced some 32 extra views, and in 2022 this was broadly unchanged at 30 extra views. In 2018, right-leaning content turned views into shares 1.3 times more easily, but this advantage disappeared in 2022.
Reducing the level of political content on Facebook does not have a neutral impact, then, especially when ‘politics’ is never fully defined. This shows that the platform is not simply a background to the debate, but actively shapes it too.
