the paradox of YouTube recommendations

Over the past several months, I’ve noticed something funny about what kinds of video recommendations I get when I watch something on YouTube. I have watch history turned off on both my personal and professional Google accounts, so if I’m logged in to Google in the browser where I’m watching the video (usually on a desktop/laptop), I get pretty generic recommendations, with an obvious connection to the video at hand but no awareness of my past viewing. On my phone, though, I’m not logged into Google in my main browser, so if I bring up a video there, I get way more personalized recommendations that are very tied in to what I’ve previously watched on my phone.

None of this is terribly surprising, I suppose, but there are a couple of observations that have emerged from this that are worth making. First, YouTube’s algorithmic recommendations are actually pretty good, and I resent it. The number of times that I’ve clicked on algorithmically recommended videos on my phone demonstrates not only the quality of the recommendations but also how clearly the purpose of the recommendations is to get users to stay on the site longer. It’s creepy, and I hate it. Second, though, there’s a seeming paradox at the center of these differing experiences. If I want YouTube to be less creepy about my watch data, I have to create an account with them to register that preference. Not only is it opt out, but it’s “hey, register with us so that you can opt out.”

Again, I can see with some distance how this all makes some kind of sense, but it feels wrong—enough that I finally got cranky enough today to write a short blog post about.

• YouTube
• privacy
• surveillance
• algorithmic recommendation