Issue 237,  published December 3, 2019

A Third Way for Discovery

Thanksgiving week is a weird time on the internet for me. A lot of the people I work with and follow log off for the American holiday, yet where I live in northwest England there’s no particular reason that explains the online tumbleweed. The whole predicament reminds me just how US-centric my internet consumption is. Anyway, if there is one trend that I’ve observed with some interest in recent years, it’s the rise of the “if you’re travelling for Thanksgiving, here are some podcasts you should listen to” list. I saw plenty on Twitter and there were also articles doing a similar thing published everywhere from the NYT to the WSJ to Elle and more.

Most of these recommendations have a few common traits, ostensibly designed around the Thanksgiving break hook. Multi-part limited-run series were very much at the forefront, being bingeable and totally finishable within a single drive/journey. The avoidance of politics was a seeming priority, either as a way to take a break from their regular routine or in anticipation of heated conversations at the dinner table. These recommendations also tended to be narrative, non-fiction, and American.

If you were in the market for podcasts along these lines, it was a pretty solid few days to be online. I, for one, certainly found some new stuff to listen to, and I wasn’t going anywhere at all. For a hot second, the oft-decried “discovery problem” in podcasting didn’t feel like a problem at all. There were even people trying to personalise their recommendations: some were even asking people what they already like to listen to and tailoring suggestions accordingly.

It was especially interesting to watch this pop of human-generated podcast recommendations in the same month that we saw the launch of two algorithmic discovery tools by major tech corporations, namely: Your News Update from Google and Your Daily Podcasts from Spotify. Both features use collected on-platform data about a listener’s interests, location, user history, and preferences to build a playlist of audio content tailored specifically for them. In the case of the former, the lists generally involve news bulletins. The latter, meanwhile, is typically made up of whole podcast episodes and trailers. But the end result for the listener is broadly the same for both features, I think. It’s something to listen to.

This kind of algorithmic curation from large corporations is just the kind of thing that can cause heckles from those who believe in podcasting’s legacy as a freer, more open alternative to the rest of the internet — not least because of the ever-present, underlying suspicion that Spotify could always lean on the feature to funnel listeners towards their own original content. (Techcrunch raised this possibility in their initial write up of the feature, which, frankly, will probably always linger no matter how much Spotify may attempt to suggest otherwise. Possibility may not be probability, but it’s still Possibility.) It also works as a way of persuading people to listen more on the platform so as to improve the recommendations. For what it’s worth, mine are a total mess, but that’s because I currently only use Spotify to listen to things I can’t get anywhere else.

If these automated systems are one extreme, then the analogue “peer-to-peer” discovery model emphasized by these Thanksgiving lists is the other. This dichotomy feels like the end-point of where the so-called “discovery problem” seems to be drifting towards at the moment: either you’re outsourcing your listening to a machine that’s progressively learning more about you, or you’re getting suggestions from humans based on their own experiences. (On that front, the choices that we’re open to are reflective of our personal constitutions, even if we’re not always fully conscious of it.)

Which raises the question: is it truly either/or? Is there some middle way? We’ve already talked about the concerns regarding algorithmically-driven recommendations earlier in this column, but the prime concern with human recommendations is perhaps best represented in a type of complaint I hear every so often, from smaller or independent podcasters about how hard it is to get their show heard in a list-making climate that prioritises the same few dozen heavy hitters.

I spend a lot of time thinking about this, as I’m constantly trying to beat this problem myself. Alongside my work for Hot Pod, I write a daily podcast recommendation newsletter for the curation service The Browser, and part of the mission of that project is to surface episodes that you are unlikely to find any other way. Listening to enough episodes to find three excellent ones for each day, while still balancing all the other things I care about (such as diversity of creators, a wide variety of subjects, different approaches and lengths, overall quality) while also avoiding being US-centric or duplicating other services, is really hard.

Sometimes it feels like I’m playing a constant, unwinnable game of chess with myself. I was interviewed by Lifehacker back in the summer about the tools I use for this process. That list includes a bulging RSS reader, multiple ListenNotes feeds and about a million random Google Keep notes. In my experience, this combination works best when I’m able to effectively combine machine and human elements.

I have an excellent international editor in Lindelani Mbatha, who feeds me the best episodes he encounters from where he’s based in South Africa, but I also get a lot of leads from the “listeners also subscribed to” module on the desktop version of iTunes and the random episode generation on the Just Listen app. I comb through past lists and reviews from other critics and I’m experimenting with the new Spotify feature. For me, the two different modes of discovery are perfectly compatible, and both make my job easier in different ways.

In an age of ever-greater data collection and personalisation, it’s not unreasonable to be worried about feeding yet more of yourself into algorithms online. But every decision we make about what to consume is already being directed and influenced in more ways than we even realise. There has never really been such a thing as a “free” choice when it comes to media, has there? I’ve never been convinced that there’s any one solution to the so-called discovery problem, anyway. It seems to me that the route to more podcast listeners overall involves a middle way between more Thanksgiving-style personal connection over recommendations and better algorithmic performance. There’s good things to be had from both.