YouTube Algorithm 2026: Net Information Gain — Step-by-Step Guide & Transcript

[00:02] back this up, right? This is not just a clickbait title. In my last video, I did mention some of this new information and strategies that I do, um and I got a lot of comments of people saying that my

[00:15] information was BS, or that I was lying. So, let me make this very clear. I'm not here to convince anyone. If you want to do YouTube the old way, go ahead. I don't give a [ __ ] But, just be aware that in a month or two, your favorite

[00:27] guru is going to be telling you the same exact information. I'm just the first name is Romero. I've been doing YouTube for the last 8 years, and across all of my YouTube channels, I've made over $2 million in ad revenue. And in this

[00:40] video, I'm going to show you exactly how I doubled my views in the last 2 months. So, strategy number one is net information gain. Right now, if you search something on Google, the first thing you're going to see is an AI

[00:52] overview. Now, the majority of those AI overviews that Google gives you are sourced from like Reddit and random articles, but Google is now heavily leaning towards YouTube videos as sources for these AI overviews. And this

[01:08] [music] is Google's objective, right? They want to make YouTube the go-to source for everything on the internet. But, the issue that Google was having is that there was too much copycat content, right? Ever since ChatGPT came out and

[01:21] all of these LLMs, it became very easy for people to rewrite someone's video itself was already annoying, but it got even worse. Because when you post a YouTube video, Gemini, which is YouTube's AI reviewer, has to analyze

[01:36] visuals, the transcript of your video, and that costs Google money. They were like, "Okay, well, we have Gemini reviewing the same exact video over and that's costing it power and money." So,

[01:50] the solution of this was something called gist, which is essentially like a mini version of Gemini, whose job is to like filter all of the copies before it reaches Gemini. Now, this new filter was added on Friday, January 3rd of 2026.

[02:04] So, literally only a few weeks ago. And the OG patent is called [music] contextual estimation of link information gain. That's for those of you that want to go and review this yourself. So, now this

[02:16] categorizes your videos now. When you post a video, step one on Google's end is that the ingestion system scans your video. Step number two, it stores your >> And step number three, which is the new step that Google added, is the gist

[02:32] filter. So, before your video can reach step number four, which is Gemini, it has to pass the gist filter. And what the gist filter does is that when it's reviewing your video, it asks, "Okay, do I have this anywhere else?" Meaning, is

[02:46] there a similar video out there that I already have? So, if you make a video that is too similar in the wording or the topic or the I guess conclusion, your video is not going to pass the gist filter and it's never going to recommend

[03:00] it. So, that's what a shadow ban is. The reason that the majority of people are giving new information to Google. They're not giving net information gain. I mean, think about it, man. Google knows everything about everything. So,

[03:14] if you're just rewording people's videos and giving it the same information that it already has in its brain, it's not going to reward you anymore. Now, that doesn't mean that you can't post a video because someone else already did, but

[03:26] approach this. Let's say you want to make a video about the Titanic. And the title of your video is, "What really happened to the Titanic?" If the script talks about the iceberg, the lifeboats, or that one band that was playing while

[03:40] the ship was sinking, when the algorithm starts scanning your video, aka gist, it's going to compare your video to 50,000 other YouTube videos that talk about the Titanic. I guarantee that at least 90% of those YouTube videos talk

[03:54] about the band, the lifeboats, and the iceberg. So then the algorithm is going to say, "Okay, we already have this. This is nothing new." And in return, it kills your video's push. Now, the correct way of approaching this is that

[04:07] knows, right? You can still make a video about the Titanic, but you're not going everyone else does. I mean, it was in the [ __ ] movie. Instead, you're going to title your video the six-word radio message that sank the Titanic. The

[04:21] entire topic of your video has to be about Jack Phillips, who was the radio operator in the Titanic, who essentially started beefing with another ship nearby, and then Jack and the ship nearby started going back and forth, and

[04:33] essentially, when the time came for him to give distress calls, the other ship didn't respond cuz he turned him down. And because of this beef, the final the captain. [music] And that's a true story, by the way. So

[04:47] now, when the algorithm goes and scans this new Titanic video, it's going to see Jack Phillips, distress signal, radio conflict. And when it compares to say, "Wait. Yeah, that this information is true. I don't really see

[05:01] another YouTube video out there that focuses only on this, you know, radio beef." And this creates [music] net information gain, meaning that when someone Googles the radio fights of the Titanic, your video is going to come up

[05:14] as sourced with the AI overview. Pretty cool, huh? Now, let's go ahead and say gist filter. This takes us to step number four, which is Gemini. But now there's a new thing with that as well, and that's called semantic IDs. Now, I

[05:28] might lose a lot of you guys in this one cuz it's a lot of nerdy [ __ ] this is actually important. I'm going to try to explain this as easily as possible, okay? Right now, when you post a video, the algorithm does not go based

[05:43] title. It doesn't care. It cares about what's inside of your video. What words something you need to understand is that words are not words to Google and

[05:55] YouTube. It is code. For example, P. Diddy. Right, to you and me, P. Diddy is is Diddy, right? But to Google, he is M01VRN. That's true. And the word knowledge is

[06:10] M04_TB. Now, that is what Gemini sees. Now, these codes here are called nodes. And when the YouTube algorithm analyzes your video, it creates one long semantic ID, which is made of nodes. Now, let me give

[06:26] you an example of how this works. Let's say we're making a YouTube video about Elon Musk. If in the script of this video, I use terms like rich tech CEO instead of Elon Musk or egg cars instead of saying Teslas are ugly,

[06:41] I would be using human language. That's not something Google goes based off of. I understand slang or satire is not necessarily um translated correctly to Google. But, if in my script, I say SpaceX or Starship or Tesla or Mars, and

[06:58] I keep hitting nodes that link back or relate to Elon Musk, they are going to be added in the semantic ID of my video. So, in YouTube's eyes, the more nodes I hit in the script of my video, the more authority my video is going to have.

[07:13] trust my channel and it's going to push it because of that, because I give a lot of context to a specific topic. Now, here's the tricky part. Take the term young shoppers. Right, when I say young shoppers, you probably think of a kid or

[07:27] a teenager buying groceries or buying clothes. But to Google and YouTube, the term young shoppers does not exist. It's not a real term. It doesn't have a semantic ID, meaning that to Google, it's not a real word. But, if in the

[07:42] script of your video, instead of saying young shoppers, you say Gen Z consumers, guess what? The algorithm is going to understand exactly what you're saying because Gen Z consumers does have a semantic ID. So, how do you fix this?

[07:56] Well, you can do it manually or you can use Vid Ninjas. Now, this is the first publicly, but this is a software that I created that I shared to members of my school community. And not only can you use this as production workflow, but you

[08:10] can also track up to 30 competitors where you can see their total views, how many views they get per hour. You can see this month, this day, and then go based off of their best performing videos. It also has a tool built in

[08:22] where it helps you create scripts, descriptions, tags, and even thumbnails. But as for semantic IDs and hitting those nodes, we added a section called The Sauce where all you have to do is paste the script of your video, give it

[08:36] a target topic or primary entity, and click analyze sauce. What Vid Ninjas is compare your video to Google's entire [music] knowledge panel, and it's going to tell you exactly what's missing and what you have to reword. It's also going

[08:51] to tell you exactly what semantic IDs hit, including the codes. It also searches Google and YouTube to see the originality score of your script, you know, to see if anyone else has done it before. And if you scroll down, you can

[09:03] and it's going to automatically apply those changes for you. So, with this, you're going to have original scripts that every semantic ID / node. It's also going to tell you the net information gain score of your video, and it fixes

[09:17] That's pretty cool. If you want to check this out, you can click the link in the below my school community, where if you don't know, I I >> It sounds like I'm plugging everything right now.

[09:31] I feel like [ __ ] Billy Mays right now. I'm the owner of Views for Income, which is one of the biggest YouTube communities in the world. We use school over 80 plus videos where I teach you

[09:43] I have [music] extensive updates on the Google and YouTube algorithm. And it's essentially a blueprint on how I made over $2 million with YouTube. And I also I speak with you guys directly, answer any questions, review channels, all that

[09:58] good stuff. My camera ended up dying. I'm wearing the same shirt because I was just pretend like it's not a different day, but you know, whatever. But the last one. And that would be the invisible ceiling, which is GARM Brand

[10:13] and Safety, which stands for Global Alliance of Responsible Media. So, what this means is that YouTube is a business. And they have to make a advertisers. And advertisers and YouTube, they have something called

[10:27] GARM, where advertisers can pick specific words or topics that they don't want their ads to be shown on YouTube. And if advertisers don't want their ads running on your video, YouTube is not going to push it. Now, my hypothesis is

[10:40] bit. And what I mean by that is that advertisers might be pulling away from AI stories, from very AI generated content, from brain rots. And because this type of content is so new, they're probably adding like specific words or

[10:55] phrases that are essentially killing the momentum of a lot of channels, which explains why YouTube might be terminating and demonetizing a [ __ ] ton something happened in the back end with this whole GARM thing. And you know,

[11:07] advertisers started saying, you know what, I don't want my videos to be in political videos, in violent videos, of course. But now, I also don't want my videos to appear in AI generated content. And because there was so much

[11:19] of this content made and posted on YouTube every single day, they probably noticed that a lot of channels were being skipped by by advertisers. But again, that's just a theory. Again, I don't think YouTube is 100% against AI,

[11:31] but I do think they're against monetizing AI slop. You got to remember that these AI softwares are tools. I don't think the intention was for AI to replace, you know, the actual visuals in the production workflow. So, myself, I

[11:44] but I know a lot of people have. I don't know if I sound crazy right now, but true, at least from my experience, right? Like, I have this little notebook Google theories, right? Like, if they update anything new, I'm like, "Okay,

[11:58] what would the change be?" Again, I don't believe that they're going to complicated [ __ ] or a little bit confusing. So, then I start experimenting myself. And with semantic IDs and that information gained with my

[12:11] experimentation, it seemed to be 100% effective. But yeah, that's pretty much doing with other Google updates, you can go ahead and click on this video here. I give a bit more context. And yeah, I guess I'll see you guys next time.