How the YouTube Algorithm Works in 2024

A brief history of the YouTube algorithm

What is the YouTube algorithm? To answer that question, let’s do a quick overview of how YouTube’s Algorithm has changed over the years. And how it works today.

2005-2011: Optimizing for clicks & views

According to founder Jawed Karim (a.k.a. the star of Me at the Zoo), YouTube was created in 2005 in order to crowdsource the video of Janet Jackson and Justin Timberlake’s notorious Superbowl performance. So it makes sense that YouTube’s algorithm started off by recommending videos that attracted the most views or clicks.

Of course, this led to an increase in misleading titles and thumbnails (a.k.a. clickbait). User experience plummeted as videos left people feeling tricked, unsatisfied, or plain old annoyed.

2012: Optimizing for watch time

In 2012, YouTube adjusted its recommendation system to support time spent watching each video. It also included time spent on the platform overall. When people find videos valuable and interesting, they watch them for longer. Or, so the theory goes.

This shift to reward watch time was a game changer. According to Mark Bergan, author of Like, Comment, Subscribe: Inside YouTube’s Chaotic Rise to World Domination, “[Watch time] had an immediate impact. Early YouTubers were basically making TikTok videos… but watch time created gaming, beauty vlogging, alt-right podcasts… all these verticals we now associate with YouTube.”

Accounts that were big performers previously (like videos from eHow, or MysteryGuitarMan) dropped off almost immediately.

YouTube’s algorithm change led some creators to try to make their videos shorter in order to make it more likely viewers would watch to completion. Others made their videos longer in order to increase watch time overall. YouTube didn’t comment on either of these tactics and maintained the party line: make videos your audience wants to watch, and the algorithm will reward you.

That said, as anyone who has ever spent any time on the internet knows, time spent is not necessarily equivalent to quality time spent. Soon, YouTube changed tack again.

2015-2016: Optimizing for satisfaction

In 2015, YouTube began measuring viewer satisfaction directly with user surveys. It also prioritized direct response metrics like Shares, Likes, and Dislikes (and, of course, the especially brutal “not interested” button).

In 2016, YouTube released a whitepaper describing some of the inner workings of its AI: Deep Neural Networks for YouTube Recommendations.

In short, the algorithm had gotten way more personal. The goal was to find the video each particular viewer wants to watch, not just the video that lots of other people have perhaps watched in the past.

As a result, in 2018, YouTube’s Chief product officer mentioned on a panel that 70% of watch time on YouTube is spent watching videos the algorithm recommends.

Hootsuite

2016-present: Dangerous content, demonetization, and brand safety

How does the YouTube algorithm work in 2024?

When deciding what to recommend to each user, the YouTube algorithm takes into account the following:

• What videos have they enjoyed in the past? If you’ve watched a 40-minute video essay about the flags of the world or gave it a like or comment, it’s probably safe to say you found it interesting. Expect more flag content coming your way.

• What topics or channels have they watched previously? If you subscribe to the Food Network’s YouTube channel, the algorithm will likely show you more cooking content.

• What videos are typically watched together? If you watch “How to change a monster truck tire,” and most people who watch that also watch “Monster truck repair 101,” YouTube might recommend that as follow up viewing.

Hootsuite