Different Search Engines as Vectors of Propaganda

The next speakers in this final ECREA PolCom 2023 conference session are Elizaveta Kusnetsova and Martha Stolze, whose focus is on computational propaganda and the broader relationship between algorithmic systems and mis- and disinformation. This has been highlighted especially by the use of algorithmic tools by the Russian propaganda machine, particularly in the context of the Russian war against Ukraine. This continues a long-standing tradition of Soviet and Russian propaganda by using new technologies.

The present study focusses on the ‘US biolabs in Ukraine’ disinformation story, and is interested in what information sources search engines provide in response to this query; it also explores how the search engine and search language affect these results; and how these results vary based on the searcher’s location and time. This was investigated through algorithmic auditing, using agent-based testing.

To collect the data, the study conducted four rounds of searching in the morning of evening of two consecutive dates from France, Germany, and Switzerland, and in English, French, German, Russian, and Ukrainian, for Google, Bing, and Yandex. The analysis focussed on the organic search results and coded the results for source type, source leaning, conspiracist content, and misinformation content.

All search engines prioritised news outlets as sources; Google focussed solidly on fact-checking content as well, while Bing was mixed (for English and Russian searches), and Yandex provided fact-checking content only for Ukrainian-language searches. The news outlets were also mixed; Russian-language searches across all engines returned a substantial number of Russian state media, while Google returned these only for Russian-language searches and Bing was again somewhere in the middle.

Yandex also provided far more mis- and disinformation-promoting or -mentioning content, while Google did so only for Russian-language searches, and Bing for French-language searches (for some reason). Oddly, across all search engines searches in the mornings also provided more misinformation content than in the evening – this may be a function of 24-hour news cycles.