Infinite Scroll Is (Usually) Just a Bad Recommendation System

Stairs going down

Only one way to go with infinite scroll.

Infinite scroll, the web design practice that lets readers scroll down into more stories on a news site after reaching the end of a post, became popular with news publishers after the rise of feed-based sites like Instagram, Facebook and Twitter.

Infinite scroll’s not a simple thing to implement. URLs have to change in the address bar as the reader moves to the new story. Analytics packages have to know when to recognize a new read. Ad units have to be timed to fire again.

Infinite scroll is a recommendation system.

But infinite scroll looks like social networks do, so publishers keep adopting it. (Though some, like Forbes, have abandoned it).

I’ve never seen an infinite scroll implementation that’s smart. Or to put it more kindly, I’ve never seen an infinite scroll site that uses data smartly to decide what to show next.

Why does that matter?

Infinite scroll is a recommendation system.

In fact, it’s a high stakes recommendation system. The publishing system has to decide somehow what’s the next best story for a given reader to get next. There’s no choice for the reader.

Social networks have always known this model requires some sort of data or user selection.

Facebook has always built its feed using data, tweaking what to show to a user based on previous behavior.

Twitter’s feed for a very long time was reverse chronological only, but Twitter users customized that feed by choosing who to follow or unfollow. Instagram worked the same way.

Both have now adopted a more data-centric approach.

News sites, by contrast, seem to have no real data backing to their infinite scroll recommendations.

One possible option is just to show the most recently published stories, which appears to be the approach taken by Digiday, for instance. Another seems to be showing what’s popular or a selection of editors’ picks (which I believe is how works).

Others seem to go for some weak version of a related algorithm, by choosing stories that share a tag or category with the original story.

None of these are compelling or sophisticated ways of making a recommendation for a reader – especially since we live in the age of data.

To be honest, I’m not a big fan of infinite scroll on publisher’s sites, which unlike social networks, have longer content to get through. And infinite scroll doesn’t take you through a list of browseable posts – it drops you into the next one.

That seems presumptuous.

I prefer being presented with some choices (cough-cough) like Contextly’s mix of related, personalized, evergreen and popular stories.

But if publishers stick with infinite scroll, they should do something smarter that is data-driven and personalized. That’s Facebook’s real advantage.

(Final shameless plug: If you are a publisher that wants to make your infinite scroll smarter, drop us a note at We have an API you could work off. And, yes, our recommendation system does work with infinite scroll.)

Photo: CC licensed by Jan Fidler

Comments are closed.