Visual digital intermediaries and global climate communication: Is climate change still a distant problem on YouTube?

PLOS ONE Visual digital intermediaries and global climate communication 4.5 Examining ranking assemblages To examine the conditions of visibility in the underlying ranking assemblages across con- texts (RQ2), we undertake systematic large-scale observation [11,66] with focus on platform-, user-, and channel-side factors. As noted earlier, it is difficult to disentangle the various agencies that contribute to rank- ing assemblages. Nevertheless, we argue that it is possible to systematically probe the multi- sided contributions in a comparative perspective. The systematic analysis pivots on a logic of potential position. In brief, we assume a space of possibilities for how platform, channel, or user (i.e. audience) agency could in principle influence the ranked content, then locate the actual position in the top-ranked results. We focus on the potential of a particular side to con- tribute to diversifying the ranking in ways relevant to dissolving the green ghetto on a plat- form such as YouTube. To dissolve the green ghetto, platforms need to surface fresh channels and themes, but it is also the case that channels need to bring fresh content and audiences need to engage with it. For the platform side, we analyse the extent to which the same query yields different results across time and place, with an eye to the visibility given to particular types of channels and the conventional arctic visual themes. We assume that the platform in principle could actively diversify results for the same query at different times and could return locally relevant results

at different locations. For the temporal analysis, we execute the same queries at different times, from a few seconds to three months apart, then compute the overlap between top- ranked results as the number of common top-20 videos divided by 20 (see supplementary material: diversity, time). We only requested videos published until December 2020 (several months prior to the data collection) in order to avoid contaminating the results with recent videos not available to previous searches. To assess localness in the platform context, we take all available region/language pairs query results, and examine location indirectly, through lan- guage (expecting there to be a correlation between channels and videos with metadata from regions where that language is spoken), and directly, by analysing the effect of the regionCode parameter, which returns only videos that can be viewed in the specified country (videos can be blocked in a specific region either by the channel or by the platform). For the channel side, we assess if diversification of (channel) voice brings diversification of visual content by examining whether all channel types in the same context have the same probability of showing the arctic visual themes. Similarly, to probe the role of user audiences, we consider diversification in channel preferences. We assume that if audiences strongly pre- fer a channel, this would favour its visibility in the top (even if channels with little engage- ment also appear). We take engagement in views, comments and likes as proxy for the broader

set of user popularity signals that platform ranking algorithms may draw on. The YouTube data only provides global metrics for each video, but we can still compare relative preferences across contexts. 5 Findings 5.1 Diversity and proximity in top-ranked results We first consider how YouTube articulates climate change to global audiences: what charac- terises the top-ranked results on YouTube Search with respect to diversity and proximity, and does it differ across contexts (RQ1)? The findings show that on YouTube, the issue is still typ- ically a distant problem delivered by a constrained circle. In general, the issue is defined by a delimited set of channel types, who do in fact expand beyond the usual suspects but still sus- tain a traditional distancing visual repertoire. This said, it depends on where you look. Closer PLOS ONE https://doi.org/10.1371/journal.pone.0318338 April 8, 2025 10/ 23