Updated October 10, 2022

False similar

The "Recommendations" list is a popular engagement feature used in apps of different types, from music streaming to news outlets. Recommendations increase user satisfaction, make sessions longer, and boost retention rate. Even though the pattern looks quite similar in all contexts and the purpose is clear ("Give me more of this"), a one-size-fits-all algorithm may lead to different outcomes.

Let's go by example.

Imagine a music streaming service. You are happy if the app is continuously playing songs with similar acoustic characteristics. That's because you most likely want to extend your current state. Listening to similar songs leads to minimal deviation from the mood. Let's call this job "find other similar units".

Now imagine a news aggregator using the same algorithm. You are reading articles with similar ideas. Reading similar ideas reinforces your opinion since there's only little deviation from the original one. This creates resistance to other opinions and forms a bubble. What perhaps could be a better job is to "find other units on a similar topic". This would help form a better reader's opinion on the topic and will create better resistance to manipulation.

But hey, things that make us angry keep us more engaged.