Following our tests, we’re sharing significant updates to the Recommended For You algorithm to better recognize games that keep players coming back over time. This will help those games reach the largest relevant audience on Roblox.
As part of our continuing commitment to transparency, we’re sharing the signals and their relative contribution here and bringing them into Creator Analytics so you can see what matters, where your game is strong, and where to improve.
We’ll keep improving the algorithm and will share any meaningful updates with you.
What’s Changing in the Recommended For You Algorithm
We’re making 3 key updates to how the Recommended For You algorithm evaluates games.
Expanding familiar signals from 7 days to 28 days: The Recommended For You algorithm is evolving to directly measure longer-term retention across Day 1, Day 2–7, and Day 8–28, using signals you’re familiar with: playtime, play days, qualified play sessions, intentional co-play days, spend days, and Robux spent. Measuring across these three windows gives the algorithm a clearer view of which games continue to bring players back over time, not just which games drive early activity. Our testing has shown this change not only increases user engagement through the first 28 days, but also into the future, helping you build a more robust, longer tenured player base.
Removing QPTR and introducing PTR and first play bounce rates: We’re removing qualified play-through rate and adding play-through rate and first play bounce rates so we can better understand what happens immediately after a player decides to click on a game. This lets the Recommended For You algorithm measure play-through, session quality, and retention separately and more precisely. This change helps guide players towards high quality games for them, and prioritizes getting you players who play your games longer and come back to play again.
Sharing the relative importance of each signal: We’re sharing relative importance among different signals to help you understand where to focus.
Each signal is based on users who organically joined from the Recommended for You sort on Home. Here is a 2 min overview video of the overall algorithm and you’ll also find details in Creator Docs
What This Means for You
We’re rolling out these changes today, and soon you’ll see these updated signals in your Creator Analytics dashboard under the Acquisition tab > Home recommendations. Take a look at your data so you can better understand where your game is strong, and where to improve.
With this launch, you may see some fluctuations in your distribution. If your game keeps players coming back over time, this algorithm update is designed to reward that more clearly in distribution. There’s never been a better time to build a game with lasting value.
Join Our Webinar
We’re hosting a webinar on June 17 at 11am PST to walk through these updates in more detail and be available to answer questions live. Register here!
FAQs
What factors influence Home recommendation impressions?
There are 4 factors that impact how many home recommendation impressions each game gets:
What you do (changes to gameplay, updates) influencing signals of home recommendations,
What Roblox does or Discovery algorithm changes (we share any meaningful changes/updates transparently),
Total users or people visiting Roblox Home (seasonality). There are many forms of seasonality: weekly seasonality which usually peak on a Saturday and lower during weekdays, and summer/back to school seasonality, holiday seasonality, and
If other games are significantly more successful at engaging players, your game’s distribution may decrease, even if your own engagement signals remain steady.
How do benchmark games impact discovery algorithms and home recommendations?
Benchmarks and benchmark games do not impact the recommendation algorithm in any way, but are meant to provide you with a point of comparison on the creator analytics dashboard. Benchmark helps you to identify signals that are low as areas of opportunity.
Will thumbnail personalization also use PTR?
Yes, we will be testing updating thumbnail personalization to use PTR too.
You know what would be a massive improvement to Recommendations? Being able to mute words we don’t like. I have been on Roblox for nearly 18 years now. Not ONCE have I ever played a brainrot or Furry game. Yet I keep seeing those on my home page, even after marking them all as “Not Interested”… Every other social media platform allows us to mute keywords to stop seeing them. Please give us more control over the games we see, so we can tailor our algorithms.
This is all well and good, but the fact remains that we can’t find the games we want to play, because there is no discovery in the discovery algorithm. If I want to search for a scifi game or a platformer, I can’t. If I want to continue playing a game I played 9 days ago, I can’t because the continue sort gets buried. If I want to filter out brainrot games, I can’t.
The algorithm is telling me what I “should” play, instead of the other way around. Until you guys fix this all of these changes are meaningless for new projects.
Not to mention the continue play sort also getting resorted NOT in chronological order, it makes absolutely no sense to me why a game I played a month ago would appear before one I played a few days ago
When I go to the continue play section, it’s because I want to play a game that I played recently, so it ONLY makes sense to be in chronological order
Could we get a banner in Studio or the Creator Dashboard linking to this post please? This is a major algorithm change that should be spotlighted as much as possible
So much THIS! The current recommendations system makes way too many assumptions. Just because a game is popular with some audiences does not mean everyone else is going to like it. A proper algorithm would prioritize genres that we already play. Like, why is my home page full of furry and brainrot, when I have exclusively played sci-fi and military games for nearly two decades…? I constantly mark recommendations I dislike as “Not Interested”, yet new ones keep popping up like an endless game of Whack-A-Mole… If we could just mute or block those words entirely, I would have a much cleaner home page.
I can see there are three different charts for each metric, are they each considered separately for the algorithm with one type weighted more than others? or are they combined and the breakdowns are just to help the developer identify where to make improvements?
How about players joining with 1.5GB RAM devices who end up crashing in first few minutes? These will affect the bounce rate because of their device limitations.
I may be missing something, but I’m confused about how this works for new games.
If these signals are based on users who organically joined from Recommended For You, wouldn’t a new game need RFY traffic first before it can build strong RFY stats?
Ads don’t fully solve that either, because ad players are usually lower-intent than algorithmic players, and the post says these signals are based on RFY organic users anyway.
So is there a cold-start system for new releases, or some kind of fair testing period before 28-day stats matter heavily?
WAIT hold on, so if I understood correctly, now the algorithm takes into account ONLY organic (Home Recommendations) visitors statistics, over any other players who return daily, through search, ads, or joining a friend?
So what I am supposed to do when Roblox home recommendations keeps sending me players with devices that crash upon joining the game?
Edit: It would be good to have an option to limit game access for under performing devices, this would solve the bounce rate.
Looking for some clarification here, but this change doesn’t seem like it will make the algorithm push your game slower than usual, instead rewarding well designed games while also giving a more transparent timer per say for developers to implement changes to their underpreforming games before their discovery is impacted.
