We ran our audit by searching for 32 different queries multiple times in over 3,000 counties across the United States. The queries were chosen to be of local interest (e.g. "mayor" or "school board") and more general national interest (e.g. "president" or "immigration").
The actual searches were conducted using Selenium to manage the browser and change our location for each query. We randomized the order counties were sampled and took precautions to make sure cookies and other traces did not bias subsequent searches.
The first striking result we found was the gross level of inequality in the number of times some outlets were returned compared to others. @NYTimes, @WashingtonPost, and @CNN combined to make up over 16% of search results.
To put this in perspective, the distribution of the number of times each outlet was returned was more unequal than the income distribution in the United States. Worse, this distribution was more unequal than the distribution of newspaper circulations collected by @CISLMUNC.
We can also look at these results in the context of just how much local news was returned in each search. When we did so, we found that local topics included more local news outlets in their results than general topics.
But, looking at the whole set of results may be an overly optimistic decision. Most users only look at the first few search results for any query. When we look at the top of the search results, we see that the share of local news outlets drops for almost every query.
When we modeled the likelihood of finding a local or regional news outlet, we found that features of the local news economy were not associated with seeing either type of outlet. What really matters is the type of query and how far down you look!
The lack of sensitivity to local conditions led us to ask whether GN was sensitive to any economic features. We found that a newspaper's circulation is strongly associated with the number of times it was returned. This process ultimately reinforces existing inequalities.
So, what should we make of these results? Is everything as bad as it seems for local news outlets? Maybe not!
To see whether locally oriented or generally oriented terms were more popular, we collected Google Trends data for each term. After standardizing against a term with near-constant interest, we found that local terms actually received more interest than general terms!
There is still a lot working against local news outlets, though. For starters, being part of a healthy local news economy where people are more likely to want to read the local news is not enough to increase the probability of being positioned advantageously in the results.
This insensitivity means that there is no clear and obvious set of practices that outlets and communities can adopt to help make their local news coverage more sustainable.
Rather, the GN algorithm appears to reward existing audience reach or readership. Local outlets that are already struggling to gain new readers cannot count on GN to help direct them there.
Luckily, there is plenty of room for more research! We need to synchronize audits of publishers and platforms, track and classify the content being posted to sites, audit actual user behavior, and test whether our results replicate on other platforms like Twitter and Facebook.
We know that social media platforms promote content that is popular and leads to engagement, but these posts are not always accurate and often require clarification. https://twitter.com/TwitterSupport/status/1300872087872176128?s=20
But, popular posts can also fail to capture the lived reality of people in the communities being covered. https://twitter.com/mattyglesias/status/1300469764146573312
In a bit of good news to end this story, we've already started to see Google engage with these issues. They've been working to promote local news during the pandemic and have been publicly communicating some of their decision making logic in this area. https://twitter.com/searchliaison/status/1257368114645393411
Finally, we'd like to thank @davidlazer, @gregmartinphd, @RERobertson, @DanielTrielli, and @nikkiusher for their feedback as we worked on the project. Thanks also to Brian Weeks, @s_soroka, and @UMichPCWG for inviting us to discuss the paper at their February symposium!
You can follow @seanafischer.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: