Growth & updates

Roblox Home recommendations: get discovered and sustain growth

Aim for suitable players who understand the game and return, rather than a single traffic spike. Work through three practical checks: the game is eligible for recommendations; its first users provide useful evidence; newly reached audiences still have a strong experience. There is no switch that turns on Home or a guaranteed concurrency threshold.

MARPLA team · Published 2026-09-18 · 14 min read

Full guide

Start with the essentialsRoblox Home recommendations: action plan Knowledge sectionGrowth & updates →

Original diagram of recommendation stages. Not a Roblox interface, a traffic promise or a fixed country sequence.

Home is not one traffic source

Roblox Home contains Recommended for You (RFY), Continue, friends, ads and other surfaces. This guide focuses on organic RFY. Do not attribute every visit from Home to that system or call a paid impression organic. First identify acquisition source, then inspect the specific Home signals.

Start in Creator Hub → Analytics → Acquisition → Home Recommendations. Compare the timing of impression changes with signals and player composition. The official Discovery documentation separates candidate retrieval from ranking; these are not guaranteed successive growth tiers.

How a game becomes a candidate and earns impressions

Stage	What is known	Creator action
Retrieval: candidate selection	The system considers games a particular user might enjoy. Evidence from multiple channels can help a game be considered.	Attract relevant initial players and measure their experience.
Ranking: personalized ordering	Organic RFY distribution relies on behavior from users organically acquired through RFY.	Improve those users’ first experience and repeat play.
Explore → expand	The system explores new audiences and may extend distribution when responses are strong.	Expect audience composition to change rather than demanding a constant average.

Ads, search, friends, external communities and other channels can help surface a game as a candidate. They do not buy a durable position. Even a small genuine audience can provide evidence; Roblox does not publish a universal entry requirement for CCU, ad spend or D1.

A test after an update is a possible scenario, not a promise. Reach also depends on competing games, seasonality, Home visitors and Roblox changes. Stable metrics in your own game do not guarantee the same impression volume.

Which signals Roblox evaluates

Use the current signal list rather than an old formula from a video. The 2026 RFY description emphasizes PTR, early bounce, play days and time per user. These are several interacting signals, not a competition on D1 or average session alone. See the June Roblox update and staff replies for the current list.

Signal	How to interpret it	What to inspect in the game
PTR	The share of RFY impressions followed by play. It is not advertising CTR.	A clear, accurate thumbnail promise.
First play bounce	Early departures, with separate under-60-second and 61–180-second windows. Higher is worse.	Loading, controls, the first goal and promise-to-game match.
Play days / playtime per user	Depth and repeat play across D1, D2–7 and D8–28. Playtime is capped at 60 minutes per user, per game, per day.	Progress, reasons to return and meaningful actions.
Intentional co-play days	Returning to play intentionally with friends.	Useful invitations and shared goals rather than pressure.
Qualified sessions, spend days, Robux per user	Additional engagement and value signals.	Core-loop quality and appropriate purchases.

Creator Hub: PTR and first-play bounce in Home recommendation signals — Official screenshot from Roblox Discovery. Values illustrate the interface; they are not targets for your game.

New games do not have to wait 28 days before ranking: early signals are available sooner, and longer windows add evidence as they mature. Do not compare yesterday’s immature cohort with a fully observed month. A dashboard benchmark is a reference point, not an algorithmic passing score.

What to prepare before a substantial test

Check game availability, eligible audiences, moderation and reduced-visibility warnings. If Experience Beta is enabled, check its Home participation conditions separately.
Test a first visit with a new account on a target phone: load, first action, understandable reward, saving and a return visit.
Prepare accurate names and thumbnails. Do not lead with Robux promises or imitate another game’s name and artwork to appear equivalent.
Instrument the first-step funnel and record the build. For multiplayer, test empty and underfilled servers.
Bring a small relevant audience through genre communities, useful content or bounded advertising research. Decide in advance which outcome calls for a fix.

The practical fast path is to shorten the distance between the thumbnail promise and the first interesting action. Buying more traffic before fixing a broken entry usually just exposes the same problem faster. See first-session retention and smart advertising.

How new countries enter the audience—and what is unknown

A personalized recommendation matches a game to a person rather than giving creators a mandatory country ladder. Roblox describes audience exploration and expansion but does not publish a universal country sequence, country-test size or fixed D1 requirement for unlocking another region. A traffic spike alone cannot reveal the model’s internal decision.

A new country’s growing share can coincide with changes in language, device, age, time of day and acquisition source. Country is not language; language is not spending capacity; lower-end devices exist everywhere. Inspect the observed combination before attributing everything to nationality.

Organic Home countries cannot be configured like Sponsored Ads targeting. You can improve accessibility, translations, performance and audience fit, but an advertising country filter does not control organic RFY. Restricting a country simply to improve an average is a poor first diagnostic action.

Why averages can fall while the game grows

Consider a hypothetical example with disjoint new-player cohorts and equally mature D1. Initially country A supplied 1,000 players and 200 returned: D1 = 20%. Later A stayed unchanged, while country B added 1,000 players and 50 returns. Overall D1 = (200 + 50) / (1,000 + 1,000) = 12.5%. There are now 250 returns instead of 200, although the average rate fell.

Illustrative example: adding a country reduces blended D1 from 20% to 12.5% — Original explanatory diagram. Fictional values, not MARPLA observations or Roblox benchmarks.

This is a composition change, not proof that A’s experience deteriorated. But if B arrived through organic RFY, its behavior belongs to the Home audience: you cannot declare a weak first experience harmless. In contrast, first-time ad-acquired users are excluded from the RFY ranking signals Roblox describes.

Real D1 requires the sizes of the corresponding cohorts. Do not weight D1 with daily DAU, average country percentages without weights or add overlapping sources. If denominators are unavailable, show the separate series and leave the exact composition contribution unknown.

What Roblox does as expansion changes metrics

Roblox explains that wider reach leads to a new metric baseline: not every new user is an immediate fan. Personalization continues looking for suitable players. This explains possible variation; it does not promise restoration of your old PTR or preservation of all traffic. See Discovery.

Exact country weights, normalization and redistribution rules are not public. Do not claim Roblox necessarily shields an established country from a new one or automatically discards all weak audiences. Inspect actual distribution and your own errors; collect a reproducible report for unexplained anomalies.

What to do when a new country has weaker metrics

Observation	Check	Next action
Overall D1 falls while established countries stay stable	New-cohort shares, equal maturity and RFY source separately	Do not roll back the whole game on the average alone; investigate the new segment.
More first-minute exits in the new country	Phone performance, memory, loading, join errors and translation of the first goal	Repair the specific barrier and examine a new post-fix cohort.
Good entry but weak D2–7	Progress clarity, saving, event timing and a reason to return	Improve the repeatable core loop rather than masking it with rewards alone.
Established audiences also deteriorate	Build, server load, prices, new thumbnails and latency	Investigate a product regression rather than only composition.
Stable metrics but fewer impressions	Seasonality, competition, visibility restrictions and Home changes	Do not accept an unsupported shadow-ban explanation; keep a dated record.
Sparse or incomplete data	Cohort sizes, reporting delays, UTC dates and gaps	Record the next review checkpoint; claim neither victory nor disaster.

In practice: record the expansion date; compare complete matching weekdays; verify Home as the source; choose the country and an available joint device breakdown; locate the earliest changed funnel step; fix one cause; compare a new mature cohort. A single before/after chart cannot separate translation effects from an update and seasonality.

Localization and performance: repair the underlying cause

Start with the thumbnail promise, first goal, buttons, join errors and progress explanation. Have a fluent speaker check automatic translations on a real device for text length, wrapping, verbs and purchase names. Roblox supports automatic and manual localization; automatic translation does not establish that a particular scene makes sense.

Inspect time to the first controllable action, not just FPS after loading.
Check network conditions, heavy assets, memory, touch controls and the position of the key button.
For social play, verify that newcomers can find an understandable team and start without a full server.
Separate device and country effects: good independent percentages do not establish a good joint segment.

Thumbnails: serve the right segments

Home thumbnail personalization and game ranking are different mechanisms. With multiple active thumbnails, Roblox first gathers observations and then shifts impressions toward suitable variants for different groups while retaining some exploration. Do not treat the entire period as an equally randomized A/B test.

Keep multiple truthful alternatives rather than only the image with the highest overall rate. A new region may prefer something different. The current Roblox workflow activates 2–5 thumbnails; compare their promise with entry quality. QPTR remains a thumbnail-personalization metric, while the current RFY signal list separately emphasizes PTR and bounce.

In MARPLA, preserve set membership, dates and each image’s history. Do not automatically call winning segments countries: that depends on the detail Roblox actually returns. Rolling L7 values cannot be summed across days. See Roblox thumbnails.

New Home set settings in MARPLA — Actual MARPLA form in a local teaching example; nothing was published. The displayed time is an example schedule, not a recommended launch hour.

How to investigate Home and countries in MARPLA

Connect your own game with analytics permissions. Public CCU does not replace private Home signals. Connection instructions.
In Audience quality, first inspect the full audience by clearing an unwanted saved filter. Choose countries and a metric, then separately inspect acquisition source and an available joint breakdown.
Verify that Roblox actually supplied the combination. If source × country is unsupported or unavailable, do not manufacture it from independent totals.
Enable the required country lines; they may all start disabled. Inspect the top ten and Other, UTC dates, gaps and retention maturity.
The table compares the latest 30 complete days with the preceding 30; each chart has its own range. For a short event, inspect the chart rather than attributing a monthly result to one date.
In Content → Home, inspect sets, membership, history and impressions. An image’s saved QPTR does not replace the full game-ranking signal set.
Add a milestone or note with the expansion date, build, translation, creatives, hypothesis and next review window. Export CSV when a precise combined investigation is needed.

Use the Home and audience-expansion workflow, audience quality, Home set history and milestones. MARPLA’s internal scores help interpret observations; they do not expose Roblox’s model weights.

Audience quality: countries and calculation limits — Actual MARPLA interface with fictional teaching data. Charts are hidden for this example. Dashes mean insufficient evidence, not zero retention.

What to take from developers’ experience

Case	Author’s observation	Useful lesson
RFY countries, September 2025	The author received players from Brazil, Vietnam and Indonesia and compared weak purchases with previous advertising.	Inspect source and payer mix separately. This is an unverified self-report without a Roblox reply; suggested region restrictions do not establish a way to redirect RFY to the US.
Bamboo Tycoon, October 2025	Retention and funnel completion improved while qPTR declined. The author suspected a mismatched audience.	Inspect composition rather than only averages and benchmark lines. The cause was not confirmed, and the case belongs to the older qPTR system.
Decline after a million impressions, June 2025	The author reported losing roughly 85% of impressions after a peak. Roblox staff explained exploration and stabilization alongside platform demand and competition.	A peak is not a promised new baseline. A reply about one game does not establish a universal timeline or acceptable decline.
BODYCAM regional availability, 2026	The game was unavailable in some countries despite Region: All; the author reported resolution after metadata changes. Roblox considered public-facing game material too.	For an abrupt country-specific stop, inspect availability and notices separately from retention. This is a regional-restriction case, not evidence of RFY expansion mechanics.

The useful production discipline is to preserve context, check source and availability, and choose one next test. Someone else’s successful launch does not reveal Roblox’s hidden formula. Do not promote a creator’s opinion into a confirmed platform rule.

Techniques that shorten the path to a useful test

Technique	Why it helps	How to check
One clear gameplay moment in the thumbnail and first visit	Reduces promise-to-experience mismatch	PTR together with bounce and first action.
A new-account test on a lower-end target device	Exposes barriers hidden from the team	Load time, errors and the funnel.
A small relevant starting audience	Provides initial evidence and feedback	Source, mature cohorts and a spending cap.
Several distinct truthful thumbnails	Gives personalization meaningful alternatives	Set results, exposure and play quality.
Localize and test FTUE before broad reach	Reduces unclear first-action risk	Language/country and device cohorts where available.
A substantial update with a reason to return	Supports repeat play rather than a one-off visit	Mature D2–7/D8–28 and play days.

These improve the product and the team’s learning speed; they do not guarantee promotion. Fabricated activity, empty paid sessions, AFK time-padding, misleading rewards and endless thumbnail changes do not create durable value. Do not change gameplay, prices, targeting and the Home set at once: you will lose the ability to interpret the cause.

A decision card for the next spike

Source and event: when did growth start, and is it genuinely organic RFY?
Volume: impressions, players, new-cohort sizes and country shares.
Quality: PTR, bounce, the first step and mature returns, split by available segments.
Context: device, translation, update, creatives, servers and seasonality.
Decision: observe, repair a specific issue, test a new hypothesis or report an anomaly.
Next review: an exact date and required cohort maturity; define success before changing anything.

Download the Home review card. For a suspected Roblox issue, include the Universe ID, dates and time zone, source, screenshots showing filters and a change log. Timing alone does not establish an algorithmic bug.

Primary sources

Put this into practice in MARPLA

MARPLA tool diagram: Investigate Home and new audiences

Separate organic Home from other sources, inspect countries and supported joint breakdowns, then relate audience changes to Home set history and record a hypothesis in a note.

Open the tool →Select your connected game. Sign-in and access to its data are required.

Where this deep dive applies

Discussion0

Loading comments…