Why content platforms dilute at scale?
In my last post on Product vs Distribution, I'd mentioned why some great products end up having bad distribution. In one of the comments, it was pointed out that companies like Quora and LinkedIn lost the product after having distribution. I think there's a deeper reason for that.
The fundamental block of content is topic. E.g. Some people are interested in Cricket vs some in Football. Each topic has a 'surface area' - essentially referring to how many people are interested in a topic. E.g. More people are interested in Cricket in India than football. So surface area for Cricket in India is higher. However, at a global level, the scales might be reversed.
When content platforms start, they have to have a niche - which basically means you have to choose a topic.
Too broad a topic and you don't know how to get content and people.
Too narrow a topic and you will not get enough traction.
E.g.
Sports >> Cricket >> Dhoni >> Dhoni best shots
Let’s say you start with Dhoni best shots videos, get people who are interested in this content to come there and consume the content.
Currently, your TAM is quite low - you are just attracting people who are interested in best shots of Dhoni. So after a while, you realise that people don't spend a lot of time on your platform. Now, you ask yourself, "Why just have best shots, why not have best stumpings as well?"
So you add that, by getting creators who post good stumping videos. Over time, you start adding different content categories and scale (e.g. getting creators who post Kohli videos, Dravid videos etc). At the same time, your creators also scale with you. They had 100 followers, which grew to 1K to 10K etc. Here's where things become interesting.
The creator who was posting Dhoni's best shots compilation videos realises that as his follower base grows, he cant keep posting the same content. So he also expands his niche. A fundamental trait in this expansion is choosing the "lowest common denominator content" - basically a content that all his followers can connect with. If his follower group is very large, he will not just have Dhoni but all types of cricket videos. If his follower group is even larger, he will start adding other sports also (if he can). In this entire expansion, the original niche (Dhoni's best shots videos) gets lost. This surface area expansion alienates the original followers, because now they are (usually) getting less of the content they are interested in.
This happens more in Twitter, Quora and Linkedin, than Tiktok or Youtube. Why? Follow graph. Follow graph implies that what you see is dependent more on who you follow. In contrast, interest graph implies that what you see is dependent on what you are interested in. You can read more about the evolution of feed algorithms here.
How does the follow graph play a role here?
The problem is more pronounced in the follow-driven platforms, because it treats each tweet or post by a person you follow almost1 equally. So if you follow Dhoni, his best shot video, best stumping video and even wedding video, will be treated equally and surfaced to you. Whereas, in interest-driven platform, the wedding video will be deprioritised if you have not shown interest to such videos before.
This means, just because you liked a creator's video on one topic, it's not assumed that you'll like the video on another topic2. Hence, the strategy for interest driven platforms is to get all kinds of content and then just personalise it better. In contrast, follow driven platforms, because they prioritise all of the creator's videos, end up becoming a mess as they grow.
This is a major reason why platforms like Quora and Linkedin lose interest over time. The double dilution of content, by the platform and by the creator, makes it harder for people to discover content they truly like.
One may think that the other reason could be moderation. E.g. In Quora, would we have been able to control quality if we removed 'bad questions'? Sounds good theoretically, but doesn't work practically. Because who's to say some question is bad or not (apart from obvious legal rules drawn on pornography etc). Someone might feel that "What did you do on your first date?" is a bad question, someone else may feel otherwise. Hence, you cant truly moderate question quality. But what you can do is just show these questions to people who are interested. So again, interest graph.
In a nutshell, building recommendations based on interests rather than follows is the panacea. Of course, it's harder than it sounds. Hence, Tiktok’s dominance so far -
Twitter seems to have an interest based component as well, so even if you follow someone, you are more likely to see the content that you are interested in. But only marginally.
The algorithms may decide that you have a general interest towards what the creator is posting, in which case other videos will also surface.