Jacob O'Bryant
Home · Archive
The pros and cons of algorithmic curation
27 April 2021

Here’s a mental model I’ve been using for thinking about content consumption. Consider “content sources,” where a source could be a newsletter, someone you follow, a YouTube channel, an RSS feed, etc—anything that generates a list of content items. There are at least four ways to interact with sources:

There are a lot of different methods for coordinating who sees what content, and they handle the different cases with varying degrees of goodness.

Algorithms are great (in theory, often not in practice) for filtering and discovering content. They can analyze a huge number of sources and reduce it to a manageable load. They can introduce you to things you would never have found otherwise.

It’s a different story for subscriptions. If you want to see every one of Tim’s posts, then whatever app you’re using has one job: show you Tim’s posts. An algorithm sample will only mess that up. Out of all the complaints about algorithmic feeds, this is in my opinion the most legitimate one.

Of course, using these definitions, you don’t really want to subscribe to everyone you follow on social media. The death of the newsfeed describes how we need both filtered and subscribed content, and that those needs are addressed respectively by algorithmic news feeds and chat apps. There’s an interesting oscillation between the two (group chats grow until you need an algorithm, algorithmic feeds grow until you get sick of the noise and go to chat instead).

Blocking is an interesting case. In one sense, it’s simply the opposite of subscribing; e.g. “don’t show me any tweets from Mitt” (Tim’s evil twin). But a lot of blocked content is also yet-to-be-discovered content, like email spam or posts from Reddit trolls (yeah, it’s not a perfect mental model). Explicit blocking isn’t good enough; you don’t want to even see it in the first place. Algorithms can be great here. In fact, email spam filters are one of the best examples we have of successful algorithmic curation.

Newsletters have their place in this model too. They’re fabulous for subscribed content, and this is one of the big advantages for newsletter authors: you can build an audience of people who want to get all your content, without an algorithmic filter getting in the way. Mostly, that is.

You could even think of curated/list-of-links newsletters as a filter/discover source: Tina subscribes to 10 sources and then sends you an email with links to her favorite items. And Tina herself is one of your subscribed sources. (Do you have 10 filtered sources or 1 subscribed source? Is it pronounced potato or potato?)

If you’re thinking about improving your content consumption habits—or if you’re designing a product related to content consumption—it might be helpful to think about how you’re addressing these different needs. For example, I’ve designed The Sample specifically for algorithmic discovery via email. This makes it a bit different from other newsletter aggregators: it doesn’t try to manage any subscriptions for you. It only tries to introduce you to newsletters you haven’t seen before. It’ll forward each newsletter to you at most one time, and after that it’s up to you to subscribe if you want to keep getting it.

That will remain the focus for a while, but at some point I do think it’d also be nice to filter content. I built a feature for that into Findka Essays, before I pivoted to The Sample: you can “subscribe” to a set of RSS feeds, and then it sends you a random-ish sample from those feeds. I don’t yet have a workflow for doing that with newsletters. I wouldn’t be surprised if something in that space already exists; I should do some more searching.

One last fun example: your bookmarks should probably be a filtered source, but most apps like Pocket treat them like a subscribed source. The result is that you never actually read the bookmarks because there are too many. (Or perhaps you never read them because you don’t have a habit of going to Pocket, which is why it’s ten times easier to retain users if you deliver your product over email).

Anyway, I’ve thought for a while that it’d be swell to have a service that periodically emails you a small handful of your bookmarks, selected randomly. The bookmarks stay in the pool until you actually read them or remove them. I finally did a search for this today, and immediately it showed up as the top result: Mailist. A win for humanity indeed.

I don't write this newsletter anymore but am too lazy to disable this signup form. Subscribe over at tfos.co instead.