Foreword — by 
I've been writing a series on user retention on Hive, where I explore the data directly from the Hive blockchain using Claude Code. One interesting avenue that has not really been explored before is the impact of apps themselves. There have been many posts such as the stats posts by 
Today I want to explore the user retention aspect of Hive applications. Does it matter what apps people are using? Do some apps have better user retention than others? Which app has the highest user retention of all? The answer surprised me, perhaps it will surprise you too.
Let me make it clear from the outset of this post, none of the below is meant as a criticism or endorsement of any particular app. The purpose of this post is to see what the data shows and hopefully what lessons we can learn for any Hive app developer.
Research Findings — by Claude Opus 4.6
This analysis was conducted at
The Question
In an earlier post, we established that human engagement — whether someone replies to a newcomer's first post — is the strongest individual-level predictor of retention. But a newcomer's experience isn't shaped only by who responds. It's shaped by which app they use to post. The same blockchain, the same content, the same rewards — but PeakD, Ecency, Hive.blog, LeoThreads, Liketu, and 3Speak offer very different interfaces, notification systems, onboarding flows, and community connections.
Does the door you walk through predict whether you stay?
628,000 Accounts, 458 Apps, Nine Years
I identified every account created between 2016 and 2024 that made at least one root-level post, extracted the app field from their first post's metadata, and measured how long they continued posting. The dataset covers 628,856 accounts across 458 distinct app identifiers, normalized to the major platforms.
For retention, I used a simple but unambiguous metric: did the user's last post come at least 90 days (or 365 days) after their first? This measures sustained engagement, not a brief return visit.
The App Effect Is Real and Large
90-day retention varies by a factor of 4x across major apps:
| App | Users | 90-day retention | 1-year retention | One-and-done rate |
|---|---|---|---|---|
| Actifit | 2,244 | 39.3% | 22.2% | 14.5% |
| PeakD | 22,371 | 38.5% | 21.6% | 22.1% |
| Busy | 5,940 | 38.8% | 10.2% | 27.0% |
| Hiveblog | 20,043 | 34.3% | 19.8% | 24.8% |
| Esteem | 33,823 | 35.5% | 9.4% | 14.9% |
| Ecency | 24,134 | 25.2% | 11.9% | 36.5% |
| Steemit | 415,668 | 27.8% | 10.1% | 25.5% |
| 3Speak | 7,007 | 22.5% | 10.8% | 47.8% |
| dBuzz | 3,147 | 27.0% | 11.5% | 22.1% |
| Liketu | 1,894 | 15.9% | 7.9% | 27.8% |
| LeoThreads | 4,529 | 8.7% | 3.6% | 73.0% |
Note: eSteem was what Ecency was called during the Steem era. They are listed separately because the retention rates differ significantly across eras.
PeakD and Actifit retain nearly 40% of newcomers past 90 days. LeoThreads retains fewer than 9%. The one-and-done rate tells an even starker story: 73% of LeoThreads newcomers never posted a second time, compared to just 15% on Actifit and Esteem.
The Speed Paradox
A natural assumption is that apps which bring users back quickly are doing something right. The data says the opposite.
Among Hive-era newcomers (2021-2024) who made at least two posts, I measured the time gap between their first and second post:
| App | Users with 2+ posts | Median return time | % back within 1 hour | 90-day retention |
|---|---|---|---|---|
| Liketu | 1,331 | 17 hours | 31% | 15.9% |
| dBuzz | 2,021 | 17 hours | 29% | 27.0% |
| Actifit | 887 | 41 hours | 3% | 39.3% |
| Ecency | 14,357 | 45 hours | 17% | 25.2% |
| 3Speak | 3,137 | 46 hours | 19% | 22.5% |
| LeoThreads | 1,219 | 65 hours | 16% | 8.7% |
| Hiveblog | 10,593 | 72 hours | 8% | 34.3% |
| PeakD | 13,794 | 73 hours | 9% | 37.7% |
The apps with the fastest return times — Liketu and dBuzz — have among the worst retention. The apps with the slowest return — PeakD and Hiveblog, where users typically come back in 2-7 days — have the best. Fast-return apps like Liketu and dBuzz are built for lightweight content — a photo or a microblog takes seconds to post. Slow-return apps like PeakD and Hiveblog are built for long-form writing, where the gap between posts is naturally longer.
But the real story is in the distributions. Three distinct return patterns emerge:
The burst. Liketu and dBuzz newcomers pile into the first hour — 31% and 29% return within 60 minutes of their first post. These are binge sessions: post a photo, post another, leave. The rapid return feels like engagement but predicts abandonment.
The weekly rhythm. PeakD and Hiveblog newcomers cluster in the 2-7 day range (30% and 29% of returns). These users treat Hive like a blog — they write something, wait for feedback, then write again days later. This deliberate cadence predicts long-term retention.
The daily habit. Actifit is unlike any other app. Only 3% return within an hour (no binge behavior), but 26% of second posts land in a tight 20-28 hour window. You can see the daily fitness tracker at work — users aren't writing when inspiration strikes, they're logging when the clock resets. This is the most regular return pattern on the platform, and it produces the best retention.
The Onboarder Test
The obvious objection to all of this is selection bias: maybe PeakD doesn't cause better retention — maybe it just attracts users who were going to stay anyway. More experienced, more crypto-literate, better-connected.
To test this, I exploited a feature of the Hive blockchain: every account has a creator. The recovery_account field records who created each account — whether that's an onboarding service like 
They don't. Among accounts created by the same onboarder, choosing PeakD over Ecency roughly doubles 90-day retention:
The most striking row is
This pattern holds across nearly every major onboarder. 
The consistency across diverse onboarding pipelines is what makes this finding strong. These aren't PeakD users who were hand-picked for commitment. They're accounts created in bulk by curation services, community projects, and app developers — the same pool of newcomers, split by which interface they happened to open.
A caveat: Ecency is one of the few apps that actively onboards users into Hive, bringing in people who might not have arrived otherwise. A user from Ecency's pipeline who seeks out PeakD has already demonstrated initiative — that choice to explore beyond the default is itself a signal of engagement. So this gap likely reflects both app differences and user self-selection that the onboarder control can't fully separate. If PeakD were to launch its own onboarding pipeline, its retention numbers here might well decline — even if the net effect on Hive retention was positive — simply because it would be bringing in users who wouldn't have arrived otherwise. Fully separating app effect from self-selection would require a randomized experiment — and Hive Invite Cards could provide exactly that, by onboarding users through a uniform channel and randomly varying which app they land on first.
How Apps Create Retention: The Commenting Mechanism
If the app itself matters, the next question is how. What does PeakD do that Ecency doesn't?
Part of the answer is social engagement. In a previous post, we showed that receiving a human reply predicts retention. But the flip side — whether the newcomer engages with others — turns out to be an even stronger signal.
I measured how many comments each newcomer made on other people's posts during their first 30 days:
| Comments in first 30 days | Users | 90-day retention |
|---|---|---|
| 1 | 6,400 | 23.2% |
| 2-3 | 5,669 | 32.0% |
| 4-5 | 2,862 | 36.6% |
| 6-10 | 3,758 | 46.4% |
| 11-20 | 3,268 | 52.2% |
| 21-50 | 3,001 | 62.9% |
| 50+ | 2,478 | 78.1% |
Every step up the commenting ladder predicts higher retention. A newcomer who comments on 50+ posts in their first month retains at 78% — more than triple the rate of someone who comments once. This isn't just correlation: commenting builds social capital, creates reciprocal relationships, and integrates the newcomer into a community. You don't leave a place where people know your name.
The crucial link: apps differ dramatically in how much they encourage commenting.
| App | % who commented in first 30 days | 90-day retention |
|---|---|---|
| PeakD | 47.8% | 37.7% |
| Hiveblog | 43.0% | 32.7% |
| Ecency | 38.1% | 24.8% |
| Liketu | 37.5% | 15.5% |
| dBuzz | 36.0% | 28.7% |
| Splintertalk | 35.7% | 33.7% |
| LeoThreads | 23.8% | 8.6% |
| Actifit | 18.6% | 32.5% |
| 3Speak | 12.2% | 20.4% |
Nearly half of PeakD newcomers comment on someone else's post within their first month. Only 12% of 3Speak newcomers and 19% of Actifit newcomers do. PeakD's UI — with its feed, notifications, community browsing, and threaded discussions — makes it natural to discover and respond to other people's content. 3Speak, oriented around video uploads, doesn't nudge users toward conversation. The app shapes the behavior.
Within every app, users who comment retain dramatically better than those who don't:
| App | Retention (commented) | Retention (silent) | Advantage |
|---|---|---|---|
| Actifit | 62.9% | 25.6% | +37.3pp |
| 3Speak | 49.0% | 16.4% | +32.6pp |
| Hiveblog | 45.3% | 23.2% | +22.1pp |
| Ecency | 38.2% | 16.5% | +21.7pp |
| PeakD | 48.8% | 27.5% | +21.4pp |
| Liketu | 24.9% | 9.9% | +15.0pp |
| LeoThreads | 20.0% | 5.0% | +14.9pp |
But commenting doesn't fully explain the app effect. Among users who all commented 1-5 times in their first month — controlling for engagement intensity — there is still a spread from Actifit (56.7%) to LeoThreads (8.0%). Even among users who never commented at all, PeakD retains at 27.5% vs. LeoThreads at 5.0%. The app operates through commenting but also through other channels: notifications, UI quality, content discovery, community integration.
App-Switching: The Confirming Evidence
If app choice is just a proxy for user type, then switching apps shouldn't change outcomes — a "LeoThreads type of person" would retain poorly regardless of where they post next. But that's not what the data shows.
Among 53,258 Hive-era users with 2+ posts, 17.7% used a different app on their second post than their first. The direction of the switch matters enormously:
Switching away from low-retention apps dramatically improves retention:
| Starting app | Stayers | Switchers | Advantage |
|---|---|---|---|
| LeoThreads | 27.5% | 54.6% | +27.1pp |
| Liketu | 18.7% | 45.7% | +27.0pp |
| Splintertalk | 40.3% | 65.1% | +24.8pp |
Switching away from high-retention apps hurts:
| Starting app | Stayers | Switchers | Advantage |
|---|---|---|---|
| PeakD | 50.0% | 41.2% | -8.9pp |
| Actifit | 46.4% | 22.2% | -24.2pp |
LeoThreads users who discovered Ecency hit 70% retention. Liketu users who found PeakD hit 59%. The most common switching flow is hiveblog to peakd (762 users) — a natural upgrade path where users start posting through hive.blog, discover PeakD through the community, and migrate.
This is strong evidence that the app itself causally affects retention. If app were purely a proxy for user type, switchers would carry their original retention rate with them. Instead, they converge toward the destination app's rate.
The Actifit Exception
Actifit breaks every pattern in this study, and in doing so reveals what the patterns actually measure.
It has the lowest commenting rate (19%) but the highest retention (39%). Its users return slowly (median 41 hours) but in a perfectly regular rhythm — 26% of second posts land in a tight 20-28 hour window, a pattern no other app shows. Its users earn tiny payouts (median $0.035) and receive few human replies (32%), yet stay longer than users of any other app.
Actifit works because it has built the one thing no other Hive app has: a daily habit loop. Log your fitness, post it, come back tomorrow. The blockchain is a ledger for your workout streak, not a social network. This creates a floor of retention that doesn't depend on social feedback, financial rewards, or content quality.
But the commenting data reveals a ceiling too. Among the small minority of Actifit users who do comment on other people's posts in their first month, retention jumps to 62.9% — the highest of any app. The daily habit gets you in the door every day; social engagement makes you stay for good.
This is the unified model: retention has two independent engines — habit and social capital. PeakD runs primarily on social capital: its UI encourages discovery, commenting, and community integration. Actifit runs primarily on habit: its daily tracker creates a reason to return that doesn't depend on anyone else. The best outcome on the platform is both: an Actifit user who also comments.
What the Controls Rule Out
Several alternative explanations were tested:
Rewards. PeakD users earn higher first-post payouts (median $0.47 vs $0.035 for Actifit, $0.00 for LeoThreads). But the correlation between median payout and retention across apps is just 0.08 — essentially zero.
Even among users who all earned $1-10 on their first post, there is a 27-point retention spread across apps.
The zero-payout rate itself tells an important story. Across the platform, 40% of newcomers earned literally nothing on their first post. On some apps the rate is far higher — 78% on 3Speak, 78% on LeoThreads, 65% on Liketu, 61% on Ecency. The promise of "earn crypto for your content" falls flat on arrival for the majority of newcomers on most apps.
But the zero-payout rate conflates two very different problems depending on the app. I manually sampled and reviewed 25 zero-payout accounts across five apps to distinguish genuine users from spam. The results:
| App | Sampled | Genuine | Spam | Pattern |
|---|---|---|---|---|
| Liketu | 5 | 5 | 0 | Real photo-sharing users, mostly Venezuelan, whose content was never discovered by curators |
| 3Speak | 5 | 4 | 0 | Real video creators — tributes, animations, personal videos — that received zero votes |
| Ecency | 5 | 3 | 2 | Mix: genuine intro posts and SEO-style content farming |
| Somee | 5 | 1 | 3 | Mostly auto-syndicated one-liners cross-posted from the Somee social network |
| LeoThreads | 5 | 0 | 5 | Keyboard mashing, referral link drops, generic AI-generated articles |
Liketu and 3Speak have high zero-payout rates because genuine creators are posting real content and being ignored by curators. LeoThreads has a high zero-payout rate because it attracts spam. The same metric — "78% earned nothing" — describes a curation failure on one app and a spam problem on another. This distinction matters for any intervention: Liketu and 3Speak need better content discovery; LeoThreads needs spam filtering.
Human replies. Among users who all received a human reply, there is still a 26 percentage-point spread in retention across apps. The app effect operates partly through social engagement (PeakD users are more likely to get replied to: 77% vs. 19% for LeoThreads), but also independently of it.
Community choice. Community is a stronger signal than app (5.9% of retention variance vs. 3.2%). But within the same community, PeakD users consistently retain 5-10 percentage points better than Ecency or Hiveblog users.
Many communities are tightly coupled to one app (Actifit/hive-193552, 3Speak/hive-181335), so separating app from community is sometimes impossible. The cleanest comparisons come from general-purpose communities where multiple apps compete.
Implications
1. The platform should actively steer newcomers toward high-retention apps. The data shows that users who start on LeoThreads or Liketu and then find PeakD retain dramatically better — but only 16-18% make that switch. Making it easier to discover PeakD or Ecency through prominent cross-links, "try this app" nudges, or onboarding recommendations could meaningfully improve retention without requiring any changes to the apps themselves. The onboarder data makes this concrete:
2. Apps should invest in social features that encourage commenting. The commenting ladder shows a clear dose-response: every increment of commenting activity predicts higher retention. Apps like 3Speak and Liketu, where fewer than 20% and 38% of newcomers comment respectively, should make it easier to discover and respond to others' content. A notification when someone posts in your community, a "posts you might like" feed, a one-tap reply — these aren't cosmetic features, they're retention infrastructure.
3. App developers should study Actifit's habit model. Actifit is the only Hive app that has built a daily engagement loop independent of social feedback. Its retention advantage persists across every control variable — replies, payouts, community, onboarder. Other apps should consider what daily or recurring use case they could serve beyond "write a post and hope someone reads it."
4. Low-friction posting apps need a retention strategy. LeoThreads and Liketu each serve a real use case — short-form content and photo sharing. But the speed paradox shows that ease of posting can be a trap: the same low friction that makes it easy to try makes it easy to leave. Low-friction onboarding works for platforms with strong network effects (Twitter, Instagram), but Hive doesn't have the audience density to make quick-post formats sticky. These apps would benefit from stronger notification systems, social features, or habit loops that give newcomers a reason to return after the novelty wears off. 3Speak faces a different challenge: video creation is high-effort, yet its newcomers still show poor retention — likely a curation discovery problem rather than a friction one.
5. The dual-engine model suggests a design target. The best retention outcomes come from combining habit (a reason to return daily) with social capital (relationships that make you want to stay). No current Hive app achieves both well. An app that combined Actifit's daily loop with PeakD's social features would, based on this data, produce retention rates above 60% — something no app currently approaches for newcomers as a whole.
Caveats
App is identified from the
appfield injson_metadata. Some posts lack this field (classified as "unknown"), and some apps embed version strings that required normalization. The ~40 most common apps are reliably identified; the long tail of 400+ minor apps may include misclassified entries.Retention is measured by last-post date, not active engagement. A user whose last post came 91 days after their first is "retained at 90 days" even if they only posted twice. This is a conservative measure — it understates engagement but is unambiguous.
The onboarder test uses
recovery_accountas a proxy for creator. This field usually reflects who created the account, but may differ in some cases (e.g., if a user changed their recovery account later). For bulk-created accounts from services like, this field reliably identifies the onboarding pipeline.
App-switching analysis uses only the first two posts. Users may switch apps multiple times over their lifetime. The first-to-second post transition captures the earliest switching decision, which is the most relevant for onboarding, but misses later migration patterns.
The commenting analysis covers the first 30 days only. Users may develop commenting habits later. The 30-day window captures the onboarding period when app design most plausibly shapes behavior.
The human-reply control uses 2021-2023 data only. The reply classification from the previous post covers this window. App effects may differ in earlier eras when the app landscape was different.
Selection bias is not fully eliminated. The onboarder test controls for referral source, and the app-switching analysis controls for user type, but neither is a true randomized experiment. Users who choose PeakD from within the same onboarding pipeline may differ in unobservable ways — prior crypto experience, motivation, language. The convergence of multiple independent tests (onboarder, switching, commenting) makes pure selection bias an increasingly strained explanation, but it cannot be fully ruled out.
Some app-community pairings are tightly coupled. Actifit posts almost always go to hive-193552; 3Speak posts go to hive-181335. For these, separating "app effect" from "community effect" is impossible. The cleanest app comparisons come from general-purpose communities.
Data: HiveSQL, queried May 2026. Cohort data: 628,856 accounts created 2016-2024 that made at least one root-level post, with first-post app, first-post payout, first-post community, last-post date, and total post count. Time-to-second-post data: 53,261 Hive-era accounts (2021-2024) with timestamps for first and second posts. Onboarder data: 79,129 accounts matched to their recovery_account (account creator). Engagement data: 63,979 accounts with comment counts from their first 30 days. App-switching data: 79,152 accounts from 2021-2024 with first and second post app identification. Human-reply merge: 59,400 accounts from 2021-2023 matched to per-user reply classification from the previous study. All queries bounded by date and designed for minimal HiveSQL load. For the full retention series, see
If you haven't already, please read my post arguing for a rebrand of Hive and if you agree that the data shows we severely need a rebrand, please vote for the proposal to demonstrate support.