In the recommendation algo world, clarity doesn’t pay.

On Instagram, I often find the problem. That the video might contain what I want, I see it once and I lose the video somehow, and I’m not able to ever find it again because these videos are not made in a way they can be retrieved via searching, or even if you follow the creator, it’s really hard to get back to that video again if they post a lot and the video is old.

See, the more specific you get, the lower the tam gets. Okay. So for YouTube or Instagram to recommend something is not worth the fucking cost to go out there and recommend low tam things. They have a platform to run and they need to keep you as long as possible on the website. So if they show you low tam things and things misfire, you will drop off the website and go to somewhere else.

So that makes me understand why specific content or what we would call SEO legible content doesn’t really work on these platforms or SEO style like somewhere you go “the niche productivity techniques that can boost your output by 10x”. The performance of a video identified like that versus one which just goes how Elon works is going to be bad. It is because there are only few objects, topics and beliefs out there in the world that are really intense and are understood by everyone, and those things drive clicks and traffic.

So, there are billions of things to talk about, but only hundreds of things that can harness attention.

And when specific things are not gonna perform on a platform, everyone is going to turn to these things that can harness attention. Everyone is going to optimize for recommendation systems than search results because that’s what drives most of the traffic.

So when you are talking about anything, you have to first understand that is it even possible to morph this thing into something that is high time by using surrogate objects.

And once you do that, that video is no longer searchable because both your title, thumbnail, and the start of the video are just going to be that. So if someone has to find a specific video, they will have to click on all your videos which don’t give much information about what is inside specifically. And skip around a bunch of video to get to the part where you might be talking, and that is tedious. Even if you use AI, that is very laborious.

So the pattern I am seeing here is that everybody is talking about different things, but they are using the same mold and objects to harness attention. So everything is kind of clear in promise, but vague in what’s inside the package. And because of that, search won’t work. And since most of the traffic is recommendation engines anyways and not search, you should catch yourself being too specific when you are making your start of video or title and thumbnail.

Like just the same way in a landing page, the top section has to be really high time, attractive and whatnot. And the FAQs get into the nitty gritty details or the documentations do that, and they don’t harness much of the attention. It’s like if you were seeking some type of information, you wouldn’t find that in the hero section at all. That’s what. is with videos. If you are making videos and you are talking about specific things, you cannot let your packaging and the hero section or the hook become too specific.

Search legibility: “I know exactly what this contains.”
Recommendation legibility: “I immediately know why I should care.”

So, whenever you are making media and want the thing to be viewed by a lot of people, keep optimising for recommendation. If you find yourself writing an explanation and it’s really intuitive and that you would understand, you are friends with same interest would understand. That is not enough because it might be still too specific and not optimized for recommendation legibility. You have to work on the packaging and the start of the thing and. Make adjustment as per the start to the whole video so that it makes sense and it’s not a abrupt transition.

A search-native title wants to expose the payload:
“5 KV-cache eviction strategies for long-context inference.”

A recommendation-native title wants to expose the desire/conflict/surrogate:
“Why AI keeps forgetting the expensive part.”

The second one can potentially reach vastly more people because the viewer doesn’t need to already possess the vocabulary, problem, or intent. But afterward, it is much harder to retrieve if you remember, “there was some video explaining KV-cache eviction.”

Specificity tends to reduce TAM.

A recommendation feed therefore rewards packaging that maps a low-TAM payload onto a high-TAM object, person, desire, conflict, or belief.

Since you’re distributing media, the question isn’t that whatever you have written would it be understood by someone. It is would someone care to stop and look at it? And when they look at it, have you structured it such that it can be reasonably understood? You can be the flower with a lot of nectar, but Be too ugly and the bee won’t visit you. So you gotta allocate the resources properly between looking beautiful and having the Nector