Why 1.5M Followers ≠ Page 1 Anymore: Observations on 2026 Follower Suppression

[BullMountainHawk]

Hey everyone and apologies to Chaturbate Support team on this. Correct me if my observations are off. It is not my intention to "hack" the Chaturbate algorithm, but leverage this for your benefit.

I’ve been spending the last few days deep-diving into the session data for a few veteran rooms and some high-performing independent boutiques. I wanted to share some observations on what looks like a major shift in the 2026 discovery logic. We’re seeing cases where even 'Stadium-level' rooms with over a million followers are being 'hidden' from their own fan base at the start of a show. If you’ve felt like you’re shouting into a void for the first hour of your stream, you aren't alone—the Chaturbate AI algorithm has fundamentally changed how it 'weights' your opening 15 minutes.

1. The "Shadow Phase" & Follower Suppression​

From what we can tell, the algorithm has moved toward a "Bandwidth Throttling" model. Instead of notifying your entire follower list at once, the system appears to test a small "Control Group" of active users. If that group doesn't hit a specific "Engagement Density" threshold (it’s not just about token volume, but the number of unique tippers), the AI suppresses the remaining 95% of notifications to manage site-wide user fatigue.

This creates a "Shadow Phase" where you’re online, but invisible to the very people who signed up to see you. We’ve observed that the AI only "releases" the full notification blast once it sees a specific rate of growth.

2. The "Sentiment Engine": 100 1s > 1 100​

One of the biggest misconceptions in 2026 is that total token volume is the primary driver for placement. The new "Sentiment AI" is actually looking for Activity Density.

• Unique Tipper Density: The algorithm now prioritizes the number of individual users contributing over the size of a single tip. 100 people tipping 1 token signals a "Viral Event" to the AI; one person tipping 100 tokens signals a "Private Transaction." While the big tip is great for the bank, those 100 small tips are what actually push you up the front page.
• Reciprocity Loops: The AI tracks "Model Response Time." When a tip hits and the model (or a moderator) acknowledges it immediately, the "Engagement Score" spikes. This is why thanking even the 1-token tippers is now a technical strategy to keep the AI from throttling your feed.

3. The "Outlier Effect": Rank vs. ROI​

We need to talk about the disconnect between viewer count and earnings.

• The Retention Anchors: We’ve seen rooms pinned at the top of Page 1 with massive viewer counts (the "_arry" effect) that are essentially being used by the site as "Retention Anchors" despite having almost zero conversion.
• The Boutique Model: Conversely, we see independent "Boutique" rooms capped at 300 viewers that are out-earning the front page because they focus on Daily Pillars—recurring users who provide a "Revenue Floor" regardless of where the AI places the room.

4. Strategy for the "Small Room" Breakthrough​

If you are sitting in a room with 5 to 50 viewers on Page 50+, you are fighting the AI’s "Stagnation Filter." To break out, you need a Metronome.

• The Metronome: If you have one or two loyal users who can start a "Metronome"—tipping just 1 token every 30 to 60 seconds—the AI detects a Continuous Engagement Stream. This signals that the room is "Heating Up," which is the only way to move from Page 50 to Page 15 for a "Discovery Test."
• The Navigator (Moderator): A moderator in 2026 isn't just a bouncer or a cheerleader, they are a Navigator. They can use "Pulse Tips" (small, timed bursts) to signal "Alpha Activity" to the AI, acting as a manual override to lift a shadow-ban or jumpstart stagnant growth. If the room gets stale - no comments, no tipping, they should be able to help it reindex and move.

Final Thoughts​

The 2026 AI is a machine, but it’s a machine that rewards Activity Density. Whether you are a "Stadium" room with millions of followers or a "Boutique" room with 50 regulars, the key is understanding that Interaction is SEO.

Don't let a slow start convince you that you're "hidden"—sometimes the AI just needs a consistent pulse to wake up. Curious to hear if other veterans or the CB Support team have noticed similar patterns with the notification throttling lately!

I’d love to hear from other veterans, and again, if CB Support is lurking, I’d welcome your insight on whether this 'Control Group' testing for notifications is the new standard. Let’s help more models get back on Page 1!

 

[BullMountainHawk]

I honestly haven't read it because I don't care about the subject.


Yeah it's something that makes us cringe. There's a certain vibe to "AI speak", and your formatting and tone make it obvious there's a lot of it. This forum was made by, and is carried by, real people behind their cameras and computers. The recent shift in the world to AI doesn't sit the same with everyone.

Outside of that, don't care.

I understand ... and I get that. If it helps, I am from the technology world and as an innovator / designer... so my OWN language tends to be a bit AI-like. So it is hard to tell when it's me or when it's AI... BUT... I agree with you. I am a very real person, who knows and admires very real people behind the cameras and computers and has supported many of them very well. I WANT them to do well. I want you ALL to do well. When I'm asked, Hey B******, what's going on here? You have insights that none of us have. Tell us. So I did. And yes, I DID use AI to find out. And I shared output with my words. So yes... my words, and AI words, will sound similar.

Here's what I will do. Going forward, I will put all AI output in italics -- and specifically call it out... stating "My AI research says...." (or something to that effect), rather than weave it into my narrative.

And maybe that should become a standard. It's certainly something I do professionally and probably should have been more clear about here. But it's my pledge going forward should anyone has any further questions or observations.

 

[SCRIPTA]

...I read "typical AI drivel" or whatever insult you want to throw at me. Then by all means insult me...

It was a simple, legitimate question.

I was already aware that you'd said you use AI. My question wasn't whether you use AI as a tool. My question was whether your forum posts were being generated by AI, because they read that way. I didn't realize we were having a discussion with AI-generated responses rather than directly with you. Those are two very different things.

What surprised me was how quickly you interpreted a straightforward question as an insult and responded to arguments I never made. I wasn't criticizing your use of AI or the information you posted. I was simply asking who, or what I was actually conversing with. A simple "Yes, I use AI to help write many of my posts" would have answered the question.

 

[rockin_rod]

The comment I'd like to make OP is that this all seems to be based on surmising what is going on in CB's algorithm from what you have observed.
Even the thread title includes 'Observations'.
I don't doubt you have spent many hours observing, but only the CB software devs truly know how the algorithm operates, and as we know, it doesn't remain static either.
The problem I have with using AI (in my paid job) is that the output is so dependent on the input.
Not saying it's the case here, but when you enter speculation, the result can seem very credible but may not be factually accurate.

 

[BullMountainHawk]

It was a simple, legitimate question.

I was already aware that you'd said you use AI. My question wasn't whether you use AI as a tool. My question was whether your forum posts were being generated by AI, because they read that way. I didn't realize we were having a discussion with AI-generated responses rather than directly with you. Those are two very different things.

What surprised me was how quickly you interpreted a straightforward question as an insult and responded to arguments I never made. I wasn't criticizing your use of AI or the information you posted. I was simply asking who, or what I was actually conversing with. A simple "Yes, I use AI to help write many of my posts" would have answered the question.

And my apologies for going off on a tangent. I should have divulged that some of my content is quoting AI - but I didn't mark it as such. You are not wrong. But you ARE conversing with a real human most of the time, who uses AI in places to make his point. As I pointed out, I will certainly italicize that in the future, when I am referring to AI output. I tend to get a bit too defensive sometime. It's one of my many flaws.

Thank you for pointing that out. I always strive to do better.

 

[BullMountainHawk]

The comment I'd like to make OP is that this all seems to be based on surmising what is going on in CB's algorithm from what you have observed.
Even the thread title includes 'Observations'.
I don't doubt you have spent many hours observing, but only the CB software devs truly know how the algorithm operates, and as we know, it doesn't remain static either.
The problem I have with using AI (in my paid job) is that the output is so dependent on the input.
Not saying it's the case here, but when you enter speculation, the result can seem very credible but may not be factually accurate.

Absolutely correct. And we must continue to observe and evolve where we need to. And you're right. Only the CB software devs truly know what's going on. But I think the thousands of models who work hard here every day need to know a little about how their rooms are "presented" (if that would be the right word)? And if we have the tools to observe and inspect, and yes, (as you would do), use AI to try to interpret what we are seeing. And then test our hypothesis. Because the CB developers are not going to tell us. For example, I am told that CB patterns room placement based on similar algorithms for Twitch, TikTok, YouTube Shorts, and Spotify... among others. But I guess I can see that. (I'm curious if there is some off-the-shelf SW / SaaS that these platforms use, that CB also deployed).

I look at this way. If we publish something, like a blog or an article, or maybe a YouTube video, we are going to employ SEO to make sure our content is found. This is really the same thing. Google doesn't tell us exactly HOW content is presented during searches. But over time, we can use Search Engine Optimization to better our chances of the content being found. For CB models, it's the same. Using experimentation, it behooves us to make a few logical hypothesizes to determine how CB might be working, based on observations. And yes, it changes. It's not their job to tell us. It's their job to provide a quality product - the models, that are tailored to the users' tastes and interests. But let's both HOPE they are continually updating - not only for the users' sakes, but also the fairness to the models. NO ONE came out and said. "Hey, your room might be suppressed from your best followers from time to time." These models in many case, count on their regulars to enjoy their shows. If those regulars don't see those models online, when they are, in fact, are online - then we should find out why. It was and still is a problem for many.

When we first saw this last year, we thought it was a bug. It was "fixed" but then kept reoccurring. So that prompted me to dig deeper. Gather data. Talk to models, test hypothesis, try out solution. And YES, consult AI to help us determine what is happening.

The more we dug, the more I was able to learn. But this is not particularly static, as you rightfully point out. I'm sure CB is tweaking and changing. But I believe, I've uncovered enough evidence of what is happening and how it can be mitigated (just like we employ SEO to help our content or our business be found on the web). Are my conclusions 100% accurate? Of course, not - but I believe they are highly accurate. Are the algorithms changing? Of course, we should think that CB developers and product managers are smart folks They will look for improvements.

The goal of me posting the original post here was to simply say. Hey - this is happening to CB - here's some things that are going on. We've observed them - across multiple rooms. And upon research, here is maybe WHY. And what can a model (or even a model's moderator) do to work within these parameters.

I'd love for you to take a look yourself. See if you can come up with further solutions. This is an incredible forum, full of serious people who all want to see the modeling community do well. And if there are ways to do it, even if they are NOT 100% accurate -- but even 70% accurate, then I think it is worth the effort. And if we can help improve things, then let's do it.

 

[cbhours]

The comment I'd like to make OP is that this all seems to be based on surmising what is going on in CB's algorithm from what you have observed.
Even the thread title includes 'Observations'.
I don't doubt you have spent many hours observing, but only the CB software devs truly know how the algorithm operates, and as we know, it doesn't remain static either.
The problem I have with using AI (in my paid job) is that the output is so dependent on the input.
Not saying it's the case here, but when you enter speculation, the result can seem very credible but may not be factually accurate.

This.

You aren't feeding your AI enough information, its just speculating, CB makes baby step changes, they are usually small and simple and layered on top of existing systems. In general they are very careful about it too even though they do small baby step changes they usually only apply it to a small percentage of users/models first.

The Retention Anchors: We’ve seen rooms pinned at the top of Page 1 with massive viewer counts (the "_arry" effect) that are essentially being used by the site as "Retention Anchors" despite having almost zero conversion.

For example this indicates that the AI you are using doesnt understand the affiliate program.

 

[pbonACF]

So then explain to me why I, as a model who is not overtly energetic or doing things for free, has gained more new followers in the last six months than I ever have before and higher traffic.

Another change CB introduced was to allow filtering on Room Size. Here are data points from two different days:

Setting Rooms Available
----------------- --------------------------
No Filter 87 94
Intimate 27 32
Mid-Sized 41 44
High-Traffic 20 19

I haven't seen a precise definition of the three categories, but if Mid-Sized then CB reduced your competition by over half for spenders who are worn out from watching high-traffic rooms but still looking for capable models offering an experience they like.

Visitors can filter for and also see base rate for Private Show on the thumbnail pages which may impact visitor choice. Lower it to get more budget-minded guys, raise it to get fewer budget-minded guys.

 

[BullMountainHawk]

his.

You aren't feeding your AI enough information, its just speculating, CB makes baby step changes, they are usually small and simple and layered on top of existing systems. In general they are very careful about it too even though they do small baby step changes they usually only apply it to a small percentage of users/models first.

I don't disagree with you. And absolutely it is speculating. There will ALWAYS be some level of speculation here. Chaturbate is not going to hand us the code for us to see how they do it. And as for incremental changes. I agree there too. BUT, by the time I found out a few things, it wasn't incremental. I'm not saying they didn't start this months and months before it made it clear something different was going on. But the impact began to become clear, in late Q3, Q4 of 2025. When I look at the performance of multiple rooms, looking at tip velocity vs. placement. Number of members in the room. Chat interactions, member suppression, model interactions and even activities, and try to describe those to an AI - over and over and over, we can draw some conclusions.

And let’s be entirely realistic about the technology here: Chaturbate doesn't need to invent a brand-new, proprietary AI from scratch. They live in the same engineering world as the rest of us. Today, any enterprise platform can pull off-the-shelf, open-source machine learning libraries—like NVIDIA’s Merlin framework (specifically Transformers4Rec for session-based tracking) or Google’s TensorFlow Recommenders—and layer them directly onto their existing database infrastructure to achieve the exact predictive matching that you or I can observe today for CB rooms.

When a deep-learning analysis points to these specific algorithmic behaviors, it isn't guessing out of thin air; I believe it is identifying the unmistakable digital footprint of industry-standard recommendation pipelines. (in other words, one AI is recognizing elements of other AIs). And why would it not? When I tell AI ... okay this model complained to me this week, that while she was online from 8:00AM her time until 1PM her time, that 4 or 5 of her regular followers did not see her online. Her room was suppressed to them, even though she was streaming a great show. One ought to ask "Why." Or if you closely watch a room's member numbers creep up (or down), depending on what she's doing... as well as what page / row she appears on (that's where your programs are very useful), describing THAT behavior can lead to some SPECULATIVE conclusions.

Of course we're not going to be 100% accurate. Hell, we've had SEO for years and it, at best, is effective only 65-75% of the time for something that's been around for years, such as Google Search.

BUT if we can, leverage this speculation and hope we can achieve some level of accuracy: 60-70% (Or more, of course), then maybe we can give models SOME insights -- that is ACTIONABLE insights -- that might help their rooms grow and hard work be better appreciated. But they need to be FOUND first. And this is maybe, just maybe a start.

Oh and I won't even begin to speculate WHY the particular member suppression even occurs. My AI says this:

Possible Reasons an Algorithm Hides a "Followed" Model​

1. The "Scroll-Past" Penalty (Implicit Negative Feedback)In legacy systems, if you followed a model, they were permanently pinned to the top of your page when online. In modern AI, if a model is online and you scroll past their thumbnail two or three times without clicking, the system logs an Implicit Negative. The algorithm assumes you are "fatigued" by that model today. To protect your engagement, it drops them from your feed, assuming you simply don't want to see them right now.

2. Session-Based Overrides (Now vs. Then)Tools like NVIDIA’s Transformers4Rec prioritize your current session over your historical account data. If you followed a model six months ago, but your click behavior in the last ten minutes suggests you are looking for a completely different vibe, category, or energy level, the system will actively suppress your followed model because she doesn't match your immediate real-time vector.

It is a ruthless system that prioritizes real-time psychology and wallet-extraction over a static "Friends List."

Okay now. back to me (my voice). To me... that's the most brutal part. That I (or any other member) doesn't see a particular model as "online" when the model clearly IS. That was the original observation that I asked "WHY" about.

You are right, CB doesn't do things quickly. I expect they have been working on this LONG before it started to manifest. This didn't happen overnight. Hell, it hasn't changed all that much since Jan/Feb when I really started digging into this.

All I am saying is this. IF me (or you) or ANYONE can provide some insight as to how / why rooms are placed / suppressed / not suppressed / etc. then that is a good thing. NO ... this is not perfect, and YES, it absolutely IS speculation. But even if we are only partially right it is a start. But I believe it is NOT blind-speculation either. An AI, when shown what is going on. What the observations are. Over and over. It's going to make some connections. In this case, it at some point said... oh this is very similar to how NVIDEA's product or Google's product works, well that makes sense to me. So knowing or SPECULATING that they are using this or that, might give us some ideas and tools to figure out some insights.

We're all on the same team here. We want an ecosystem where hard working models have the opportunity to do well. And for everyone to have a positive experience. CB wants to make money as do the models. If the models can help influence their visibility and placement, then they should be offered the knowledge to do so.

Thanks for keeping me honest.

Oh and the other ... about the affiliate program - I know a bit about it. And No I didn't explain it to AI... but this is, outside of that. More to say on that later. I always cringe when I see a room with 30K users in it. I find it less helpful to the room in the long run. In fact, sometimes, it is downright harmful to the room's overall health for a while after they rotate out of an ad.

We can talk about that another time.

 

[cbhours]

When I look at the performance of multiple rooms, looking at tip velocity vs. placement

How are you determining placement?

Possible Reasons an Algorithm Hides a "Followed" Model

A followed model may disappear from a recommended slot position and returned to their original position.

The model list you see is has a ton of recommended model lists that's unique to you and that system applies to anonymous users as well.

 

[Vixxen81]

How are you determining placement?

A followed model may disappear from a recommended slot position and returned to their original position.

The model list you see is has a ton of recommended model lists that's unique to you and that system applies to anonymous users as well.

That's my question, too. Not that I have ever once cared about page rank and I've been extremely vocal about that on here, but what I see isn't going to be what my neighbor sees because we didn't click on/look at the same rooms. And you and I have had a similar conversation before where I was talking about how I don't trust all of the sources that say a model ranked this at this time.

For instance, the Chaturbate Contest Stats page has me at a reasonable number, while Lovense's weird little rank calculator has me at 6000ish. If I recall correctly, your site was the closest match to CB's numbers.

AI is trained and I can't help but wonder if OP is using reinforcement learning when plugging data. Kind of the equivalent of shaking a magic 8 ball again when you don't like the answer?

 

[rockin_rod]

When I look at the performance of multiple rooms, member suppression, model interactions and even activities, and try to describe those to an AI - over and over and over, we can draw some conclusions.

okay this model complained to me this week, that while she was online from 8:00AM her time until 1PM her time, that 4 or 5 of her regular followers did not see her online. Her room was suppressed to them, even though she was streaming a great show.

How are you observing and quantifying the term 'member suppression'?
If I have a room in my followed list, when that room begins broadcasting, I get a notification and I see the 'followed room' tab light illuminate as a visual cue that a room has gone online.
But it's still up to me to react to the notifications and go seek out those rooms.
How are you deciding the rooms have been deliberately suppressed from some regs?
Just because the model says she was online for hours, doing a great show, but those guys didn't show up, doesn't draw any conclusion that they were being purposely denied the opportunity to join the room.
If the model contacted the regulars and they told her they weren't notified her room was online, that could be BS, and for lots of reasons.

 

[KingMarti]

I understand ... and I get that. If it helps, I am from the technology world and as an innovator / designer... so my OWN language tends to be a bit AI-like. So it is hard to tell when it's me or when it's AI... BUT... I agree with you. I am a very real person, who knows and admires very real people behind the cameras and computers and has supported many of them very well. I WANT them to do well. I want you ALL to do well. When I'm asked, Hey B******, what's going on here? You have insights that none of us have. Tell us. So I did. And yes, I DID use AI to find out. And I shared output with my words. So yes... my words, and AI words, will sound similar.

Here's what I will do. Going forward, I will put all AI output in italics -- and specifically call it out... stating "My AI research says...." (or something to that effect), rather than weave it into my narrative.

And maybe that should become a standard. It's certainly something I do professionally and probably should have been more clear about here. But it's my pledge going forward should anyone has any further questions or observations.

You know what happens when AI dosnt have the information to give a real answer through right? Unless you already know the answer to what you are looking for or know where to fact check it, AI is a fiction generation machine, it will make correlations based on hallucinations which completely invalidates the data you get back rather than tell you it dosnt have enough data to give you an answer. The ranking algo is not public information, so AI has no way to actually know anything about it, so it's going to just make shit up bassed on other sites it does have data for, but the thing is other sites are not chaturbate so what other sites do is irrelevent.

People (not specifically talking about you) need to stop taking everything AI gives them as absolute fact, which is why when anyone is just copy and pasting AI output it's not even worth taking the time to read because it hasnt been fact checked. Not to mention there is so much unnecessary language added into responses to increase the token count it makes it a slog to try and read through.

 

[BullMountainHawk]

You know what happens when AI dosnt have the information to give a real answer through right? Unless you already know the answer to what you are looking for or know where to fact check it, AI is a fiction generation machine, it will make correlations based on hallucinations which completely invalidates the data you get back rather than tell you it dosnt have enough data to give you an answer. The ranking algo is not public information, so AI has no way to actually know anything about it, so it's going to just make shit up bassed on other sites it does have data for, but the thing is other sites are not chaturbate so what other sites do is irrelevent.

People (not specifically talking about you) need to stop taking everything AI gives them as absolute fact, which is why when anyone is just copy and pasting AI output it's not even worth taking the time to read because it hasnt been fact checked. Not to mention there is so much unnecessary language added into responses to increase the token count it makes it a slog to try and read through.

Oh absolute-fucking-lutely YES. It's the biggest thing about AI that pisses me off. BUT... in many occasions I will challenge results. Call it out. Being aware of that is a good first step when committing data to AI for analysis. And yes, I fact check the best I can. And you are right. NONE of us know for SURE what CB is doing. But I was asked about a specific situation. I had no reason to doubt that model's input and questions. And soon, the "suppression" happened to me. Then I watched user counts vs. placement vs. tip velocity in a room... for days and weeks. Just took data down. And THEN fed it to my AI ... I created a few specific prompts to try to diagnose what is happening. That's when it said... "oh this sounds like that NVIDA's Merlin architecture does. And then it brought up other software tools. And come on, CB developers are going to use the same tools that other sites use to manage millions of concurrent users to sort thousands of streams in real-time. Some one else said, CB does not move quickly. I'm sure they don't. These changes have probably been in the works for months and months, before they manifest.


And yes, I've check and cross-checked and checked again. I've spent enough time with AI in my professional life (creating useful agents to solve difficult business problems) to know how to deal with it.

One reason that I posted here was .... MAYBE, it could offer insights. Little things that models or moderators can do that could improve their room placement as well as attract more users and hopefully good paying users are attracted. And ALSO - I invite others to observe, try to do the same thing I did. I'm not going to make a cent on this. I have friends who are models and they asked. I tried to help. That's it.

If it improves the performance of one or more models, then it was worthwhile. But like any novel query, it's certainly worth the time to TRY to find out. And if others can also correlate and see if they get the same (or different ) results, then everyone wins.

 

[KingMarti]

Oh absolute-fucking-lutely YES. It's the biggest thing about AI that pisses me off. BUT... in many occasions I will challenge results. Call it out. Being aware of that is a good first step when committing data to AI for analysis. And yes, I fact check the best I can. And you are right. NONE of us know for SURE what CB is doing. But I was asked about a specific situation. I had no reason to doubt that model's input and questions. And soon, the "suppression" happened to me. Then I watched user counts vs. placement vs. tip velocity in a room... for days and weeks. Just took data down. And THEN fed it to my AI ... I created a few specific prompts to try to diagnose what is happening. That's when it said... "oh this sounds like that NVIDA's Merlin architecture does. And then it brought up other software tools. And come on, CB developers are going to use the same tools that other sites use to manage millions of concurrent users to sort thousands of streams in real-time. Some one else said, CB does not move quickly. I'm sure they don't. These changes have probably been in the works for months and months, before they manifest.


And yes, I've check and cross-checked and checked again. I've spent enough time with AI in my professional life (creating useful agents to solve difficult business problems) to know how to deal with it.

One reason that I posted here was .... MAYBE, it could offer insights. Little things that models or moderators can do that could improve their room placement as well as attract more users and hopefully good paying users are attracted. And ALSO - I invite others to observe, try to do the same thing I did. I'm not going to make a cent on this. I have friends who are models and they asked. I tried to help. That's it.

If it improves the performance of one or more models, then it was worthwhile. But like any novel query, it's certainly worth the time to TRY to find out. And if others can also correlate and see if they get the same (or different ) results, then everyone wins.

"you were right to push back on that"

I wrote a program 2 maybe 3 years ago (so defiantly out of date now) that kept track of where my room was ranked along with logs of what was happening in the room, in regards to # messages sent, total number of tokens, total amount of tip actions that occurred between updates. Ultimately findings were not very insightful because I would have also needed to track every other room on the site since rankings isnt just about what happening in your room but whats happening in all the others. Some days an action would cause me to jump pages ahead of where I was, other days when those same actions happened I got knocked pages backwards. Even if you were to track everything it would end up being far to much data to actually go through.

I do think that tokens per second (well 15 minutes because it think thats still the update time but per second just sounds better) is still the most important ranking factor, no reason other than it makes sense from a business perspective that you want the rooms that earn the most money at the front of the site since those are the rooms that are getting members to spend the most money. I would assume secondary rankings would then be % of the room tipping, number of messages in chat, number of people in the room etc. Again no reason or evidance to think that, just kind of makes sense to me that's how it would be.