Summary
Bluesky's visible activity "softness" to me looks less like broad decay and more like a problem with turning new-user activity into repeat activity. Older cohorts remain relatively stable from April through the end of May 2026, but new users seem to try the platform for one day and then go dormant.
We start this by looking at one of the great graphs generates on blue.mackuba.eu/stats. In the last two or three weeks we've seen the total number of posting users start to stabilize. That's great! But when taken together with the months before there's an obvious downward trend that is responsible for the wretched Bluesky is dying discourse.
This isn't a post to talk about how Bluesky is dying (imo it's not). I want to talk about who's active on the platform, who's not, and what new users seem to be up to.
NOTE: A lot of this work was recently done by in an excellent thread below, this is meant to 1) corroborate his results and 2) extend it further.
Who's around and visibly active on the platform?
For the past two months (April & May 2026) I've been gathering data from the PLC directory to better understand how relatively 'centralized' ATmosphere data and activity is on Bluesky PBC's hardware (Bluesky PBC being the company that operates Bluesky). If you want to see what I've gathered you can find that here.
What that data enables us to see is when account handles were first registered on the platform (i.e., when you first claimed the account you log into Bluesky with). Of course, you may have claimed the account but not started using the platform until later (I claimed mine in 2023 but only started getting active this year).
So we know when users' handles were registered, and we have information about posting / liking / following / etc activity on the platform in the last two months. Caveat: we do not see users who just log-on and scroll without doing any visible action (no lurkers seen here). We can use this information to slice-and-dice the population of users on that platform based on when they signed up and then see who's been visibly active lately. These age groups are what I'll call cohorts. In other words, I define cohort by the earliest observed PLC/identity/account timestamp available to me, not necessarily first app session or first Bluesky post.
On a daily basis we see that most cohorts have been pretty stable, save for a slight downward trend in the 2024 Nov-Dec cohort, and a slight jump up in the 2026 crowd.
When we count unique users over the course of a week we see a similar trend. Perhaps we're losing some of the people in the late 2024 spike, but we do seem to be clicking with 2026 users, right? Are some of the changes due to these Bluesky users just living on other ATmosphere Apps now? Are more people just lurking?
Lots of one-day activity for 2026 users
For a cohort to look sticky, the users need to be visibly active for multiple days across a time period. If we count the times each of these cohorts was active since 2026-04-11, the 2026 cohort (before April) drops off noticeably more than the other cohorts. We separate out the cohort created since April because if the account was made on May 24, they only had a potential week of activity covered by the window to accumulate active days.
This aligns pretty well with what was seeing: most new sign-ups do nothing, and if they are active they're active mostly for one day then go quiet. All at a greater, more precipitous day-one drop than the other user groups (that the slopes are comparable after that is actually reassuring).
This is interesting because, despite that, both the 2026 and new users perform an action at least once at rates higher than the 2024 and 2025 cohorts in the April-May window. So we are getting users who are willing to participate with the platform at least once, but are they sticking around?
We can look to see if users are still active on the 28th day after they sign-up. This is a standard number to use, suggesting that if they appear on the 28th day after signup they formed some sort of habit with using the platform. Sometimes people use the 30th day after signup.
Only about 3.5% of users who sign-up are visibly active on their 28th day, albeit a rather steady 3.5%. Keep in mind the denominator here is all repos that we found that are associated with PLC handles registered on that date. If we filter it to accounts who sign-up on that date and had activity on the day they registered, the number becomes about 4.0%. That's a small gap, which suggests that this isn't driven by people sitting on handles, rather it's people who try being active but then go dormant. If dormant handle-sitters were driving the low number, the red day-0 active line would have been far higher.
Note: reports ~5% of the same activation @28 metric in March 2026, which is higher than any we see here, but I'm not too concerned because 1) we gathered our data differently (and my way likely has more gaps) and 2) we're also looking at different time windows (March 2026 versus April-June 2026). Given that we're in the same ballpark, I think this directionally corroborates his results.
So it seems like we have users who are willing to try acting but we are only seeing a small % of them a month later.
Are they visibly active across the ATmosphere?
Do some Bluesky users just stop using bsky.app entirely for any period of time and use other parts of the ATmosphere? If we look at weekly ATmosphere activity across ~40 lexicons (including bsky.app), we see no meaningful week level difference when compared to simple weekly Bluesky activity (the WAU plot above).
However we are seeing more users fire off non-bsky lexicons now!
Takeaways
Most older cohorts have been relatively stable over the last two months. That's great and a sign of a healthy system. The more concerning pattern is that the 2026 cohort has a steeper drop-off after one day of visible activity -- though that doesn't mean the users are gone. Some could be lurking or doing other activity that isn't captured by this analysis.
The same possibility applies with new user retention: it could be the case that we're experiencing what general user retention looks like in a social media environment that's open and doesn't require an invite to get in. It's hard to tell without some reliable benchmarks to compare against, though 4-5% doesn't feel great.
That being said, if growth is the goal (and that's for the PBC to decide), then the platform likely needs better retention on the users entering the funnel, and more reasons for dropped-off users to give it another shot. It seems like the 2024 Nov-Dec cohort could be good users to target for reactivation.
Why aren't they sticking around?
While it's tempting to say it's a branding issue (which could be true, but I think only explains part of it), I think we're still disproportionately experiencing the power of inertia & network effects at large (a cop out I know). In an increasingly para-social and splintered media ecosystem, if neither your favorite creator nor your friends are here, then why should you be?
As for what the platform can do: new users appear willing to try being active, but they may not be finding communities where posting and replying feels worthwhile. While I have no proof that this is an onboarding issue, it seems like a natural place to focus because it's where Bluesky can help new users find active communities earlier. Below are some reflections and thoughts for the current onboarding flow on browser (as of June 2026).
Interests & topics are a useful way to get signal from the user, but it does feel somewhat flat. What kinds of conversations happen in the communities defined by these interests? Can I get a sense of that earlier?
Perhaps this page can be a good place to get a 'vibe' for the kinds of posts these users make. When I hover over their user card, what was their last post and when? This step could also be a good place for a path for finding people from your old network.
The starter-packs are great at connecting you to more established accounts, but it'd be nice to see which of these packs have users that are actually active. Show me how many active posters or threads this week came from the Cultural Commentators.
e.g., [12 active posters this week | 3 threads | 15 replies]
Finally, if custom feeds are a primary product offering for Bluesky, then surface those more! The onboarding steps include starter-packs and lists of people to follow, but only briefly refers to the existence of custom feeds! Making custom feeds easier to make and easier to explore within the app will make it more fun for users to hang around. Exposing custom feeds more could have a knock-on effect of increasing the stickiness of some of the older cohorts as well (looking at you 2024 Nov-Dec!). So two birds one stone.
I've been toying around with a little feed discovery feed project that could be useful for this.
extra reading: https://arstechnica.com/science/2026/05/rip-social-media-what-comes-next-is-messy/
Happy to share any data I have to reproduce this, and happy to clarify anything, especially if you think I computed something incorrectly (totally possible).