Boost Your Discovery with the Improved Recommended For You Algorithm and Analytics for Creators

Hi Creators,

Today, we’re excited to announce the global release of an improved Recommended For You algorithm, along with a new Home Recommendations Creator Analytics tab to give you full visibility into the algorithm and to help you optimize and grow your experiences.

A Breakdown of Our Recommended For You Algorithm Signals

We’ve outlined the recommendation signals used by the new Recommended For You algorithm in the table below, including each signal’s definition and its role in influencing recommendations.

Recommendation Signals Definitions
Qualified play-through rate (qPTR) The number of engaging plays divided by the number of impressions from Recommended For You.
7 day playtime per user Total playing time per user in your experience within the last 7 days (maximum of 60 minutes per user, per experience, per day).
7 day play days per user Total unique days users engage with your experience over the last 7 days.
7 day spend days per user Total unique days users spend Robux in your experience over the last 7 days.
7 day Robux spent per user Total Robux spent per user in your experience over the last 7 days.
7 day intentional co-play days per user Frequency of users intentionally playing with friends over the last 7 days (through join, invite, or private server rather than through a matchmaking).

Optimizing Your Experiences with Recommended For You Signals

Roblox is more fun when you’re in experiences with friends, and users are often looking for new experiences to join with established connections. From our testing, we’ve seen that users with higher frequency of play, spend, and interaction with friends have a higher likelihood of long-term retention. Therefore, we’ve added an intentional co-play recommendation signal, which detects users manually joining experiences with friends via invites, sessions, or private servers and strongly indicates user interest.

To encourage co-play, focus on designing gameplay that naturally motivates users to spend time together and communicate meaningfully. For example, you can include player invite prompts and private server options, making it easier for users to invite their friends and increase co-experience gameplay.

Keep in mind — intentional co-play is not the only signal to pay attention to. When all of these signals are improving together in an experience, the recommendation algorithm will pick it up. Improving your retention, engagement, and monetization directly enhances our recommendation signals, resulting in better Home visibility.

And, to foster sustainably fun sessions, we limit recommendation benefits to the first 60 minutes of daily playtime per user, per experience.

Using Home Recommendation Analytics to Grow Your Experience

Our new Home Recommendations dashboard helps you monitor these recommendation signals. To access it, go to the Analytics > Acquisition page on Creator Dashboard, and select the Home Recommendations tab. Here’s how to use it:

1. Analyze Home Recommendation Impressions and Plays Trends

For example, if you observe a dip in Home Recommendation impressions starting March 3rd, you can analyze the corresponding recommendation signal trends.

2. Analyze Recommendation Signal Trends for Optimization

Keep in mind, changes to your recommendation signals usually show up in impression changes a few days later. So, if you noticed your qualified play-through rate dip on February 28th, that likely led to the impression drop on March 3rd.

It’s also important to identify recommendations signals that are below benchmark as areas of opportunity. Note that:

Continuing with our example, the charts below show that 7 day playtime and 7 day play days are both below benchmark. In this case, you may want to consider improving your onboarding funnel to encourage users to visit your experience more.

3. Discover Experiences Your Users Also Enjoy

We’ve added a list of similar experiences at the bottom of the page, based on relevant benchmarks, to show you other experiences your users enjoy.

Using Recommended For You to Get Discovered

Empower your creative vision and build experiences users will love discovering on Home, where over 90% of traffic begins. To maximize discovery via Recommended For You, leverage its recommendation signals to understand user preferences and adapt your experiences to drive engagement and long-term retention. Consistent updates, accurate experience details, and optimization with Creator Analytics are also key to improving discovery and building a loyal community.

We are continuously evaluating and refining the Recommended For You algorithm to improve discovery. This includes running experiments on limited segments of users. We will provide regular updates on algorithm changes, including these experiments, through announcements like this one. For updates that introduce new recommendation signals or significantly adjust existing ones for a larger audience, we will strive to give you as much advance notice as possible.

Thank you for everything you do to make Roblox great.

Roblox Discovery and Creator Success Teams


FAQs

Do I have to optimize for every recommendation signal in the Recommended For You algorithm?

  • While optimizing for all recommendation signals is beneficial, it’s most effective to focus on creating a high-quality, engaging experience. Prioritize core gameplay, user retention, and accurate metadata to improve overall user experience. To improve QPTR, focusing on your thumbnail personalization strategy is key. This includes showcasing new content or gameplay, keeping multiple variations active, avoiding overly similar thumbnails, and monitoring QPTR over time, especially after surges in Home recommendations.

    It’s important to note that you may observe a temporary drop in QPTR when your impressions increase – particularly during initial discovery phases. This is a normal occurrence as new users discover your experience, and QPTR should stabilize as traffic patterns normalize.

Will these recommendation signals favor larger experiences with more users?

  • No. These recommendation signals are calculated as averages per user, not total values. This ensures that smaller experiences with highly engaged users are not disadvantaged. We are focused on per user engagement, not total engagement. The Recommended For You algorithm is designed to match users with experiences they are most likely to enjoy. By prioritizing user satisfaction, you’ll create experiences that resonate with your audience and foster long-term engagement.

How quickly will changes in developer actions be updated in the analytics and the algorithm’s recommendations?

  • Most of the signals get updated within 24 hours.
93 Likes

This topic was automatically opened after 11 minutes.


Good to know that my Obby / Platformer is compared to these games! Why is the new genre system seemingly not used here? I can imagine that these completely different games being counted as similar experiences is causing my experience stats to drop significantly.

11 Likes

Where do we see how our game icon performs?

I noticed one of my game’s Qualified play through rate dropped a lot, with no change, my hypothesis is that the icon was getting a larger QPTR.

By checking personalized thumbnail recs in the Thumbnail Personalization tabs, I see a more similar number to the new QPTR stat to what was previously, I think, a combination of thumbnail & icon QPTR.

3 Likes

It would be great if we could get an example place that shows how to create and implement all of the analytical stuff at each step.

Have you checked out the Plant reference project?

As we add new analytics capabilities, we continue to update this reference project

3 Likes

Previously, creator analytics showed unique user level qualified play through rate. To better align with the RFY signals, we updated to session level qualified play through rate now: Qualified plays divided by impressions for Home recommendations. Usually session level QPTR is lower than unique user level QPTR.

1 Like

Could you further explain what “And, to foster sustainably fun sessions, we limit recommendation benefits to the first 60 minutes of daily playtime per user, per experience.” exactly means, like 60 minutes of daily playtime per user is the max we can get or?

W update. Love the transparency into how the algorithm works so we can better figure out areas of focus to help our games grow

Could you please give us analytics access to viewing co-op player percent. You have to understand for us to learn we need direct visual feedback. If we see adding a invite prompt in this area vs this area preforms better we will improve our experiences greatly.

While there is no dedicated graph, the icon really only appears in a few places now. What you can do is look at your impressions and plays on returning users for a source like Continue Play, and calculate your CTR from that. It’s obviously not as convenient as a dedicated graph would be, but if you really want that info, it’s there.

How exactly will “avoiding overly similar thumbnails” be prevented just by having QPTR as a metric that influences recommandation rates?

I did, it’s a pretty cool guide for those looking to create bigger projects. I was simply referring to something aimed entirely at analytics and the best practices for updating and tracking the data. I know there’s some custom end points or a way to track experience specific statistics.

I know there’s some custom end points or a way to track experience specific statistics.

the reference project does this

the docs also have this information in bite sized form:

there is also a YouTube video

3 Likes

Basically, all of those stats link directly to your retention and monetization. Your game will be compared to games with severe pay to win mechanics and if your game is PVP Strategy (aka: main point of the game is that everybody has the same chances to win) for example, you will never get as much attention. I think Roblox staff should also take note that popular games will get higher payer conversion just because it’s what people talk about in schools (peer pressure, kids wanting to fit in with the rest) and they think that this game will not die.

Do all recommendation signals have equal weight?

I assume the robux spent per user over 7 days is most important.

Where is this, I can’t seem to find this anywhere?

If you go to the Analytics > Acquisition page on Creator Dashboard, and select the Home Recommendations tab, you should see a 7 day intentional co-play days per user chart.

Let us know if you can’t see this chart or if you are looking for a different metric

2 Likes

Sounds nice, but I hope it’s really good or else I am not using it…

For ‘engaging’ plays does this mean the metric is not based only on people pressing the play button, but that our onboarding also has influence? For example, if a player joins the game then leaves quickly would it not count towards qualified plays?