How businesses actually use Bluesky
We track a fixed panel of hand-checked business accounts on Bluesky and publish what it does. This issue covers 69 of them in the period from 2026-07-16 to 2026-08-15, 604 observed posts, and the single most useful number in them is that 41 percent of the accounts published nothing at all. The panel keeps growing between issues, so each issue states the panel it was measured on rather than today's size. The list of accounts is available on request.
Free to quote and to reproduce with a link. If you need the underlying figures broken down differently for a story, write to [email protected] and you get them.
What the panel shows
41 percent of business accounts posted nothing at all
Of 69 verified business accounts, only 41 published a single post in thirty days. The rest set up a profile and stopped. Whatever else is true about Bluesky, the competition for attention there is thinner than the follower counts suggest.
Those who post, post 2.6 times a week
Median across active accounts. Not daily, not once a month. Friday is the busiest day of the week and Sunday the quietest, which is the opposite of what a consumer platform usually looks like and fits a place where people post from work.
A picture more than doubles the response
Posts with an image or video reach a median of 9 likes, text-only posts 4. That is the single clearest lever in the data, and it is free.
Only 7 percent of posts get no reaction at all
44 of 604 posts received no like, no reply and no repost. For a platform this size that is a low silence rate: a business account is unlikely to be shouting into an empty room.
One post carried two fifths of everything
A single post took 43 percent of all likes in the period (16,859 likes). That is why every number on this page is a median. The mean would have been 65 likes per post, which would be true of the arithmetic and false about the platform.
When the panel posts
Both charts count the same 604 posts. Hover a bar for its exact value; the busiest and quietest day and the busiest hours are named in the table below. On a narrow screen each chart scrolls sideways.
604 posts, 2026-07-16 to 2026-08-15. Friday leads, Sunday is less than a third of it.
Two peaks: late afternoon and mid evening. The quiet hours are the small hours in Europe and the Americas alike.
The numbers
Period 2026-07-16 to 2026-08-15. Medians unless stated.
| Accounts in the panel | 69 |
| Accounts that published at least once | 41 (59 percent) |
| Posts observed in the period | 604 |
| Observations stored in total | 1721 |
| Posts per active account per week | 2.6 |
| Likes per post | 6 (mean 65) |
| 90th percentile of likes | 52 |
| Likes with media / text only | 9 / 4 |
| Posts with media / text only | 274 / 330 |
| Posts with no reaction at all | 44 (7 percent) |
| Busiest weekday | Friday |
| Quietest weekday | Sunday |
| Busiest publishing hours (UTC) | 17:00, 21:00, 16:00 |
| Posts declared English | 60 percent |
Method
What is measured
A fixed panel of business accounts, queried the same way every day. Per post we store the timestamp, likes, reposts, replies, quotes, whether media or a link is attached, the character count and the declared language. No post text is stored: a barometer is about when and how often people publish and what comes back, not about what anyone wrote.
Where the data comes from
The open AT protocol AppView only. No login, no private data, no circumvention of any restriction. These records are public by design of the protocol.
How an account gets into the panel
The search is automatic, the decision is not. A candidate must use its own domain as its handle, which means somebody owned that domain and set a DNS record, something businesses do and private accounts almost never do. It must carry a business marker in its profile, and it must sit between 50 and 50,000 followers. Then a human reads the profile and decides, because the last step cannot be automated: an illustrator with a shop link, a satirical magazine called The Clinic and the International Energy Agency all pass every keyword rule and none of them is a business account. Every member of the panel has been read and accepted by hand, and the reason is recorded with the entry.
Why the panel is small
Because the manual step is the expensive one, and it is the step that makes the number worth quoting. Of 71 machine-generated candidates in the last round, 28 survived the read-through. Earlier passes were worse: keywords alone produced 300 accounts of which about two thirds were private people who happen to mention a bakery, and a later pass matched 'spa' inside 'space' and pulled in a national space agency. A small panel that survives being clicked on beats a large one that does not.
What changed on 15 August 2026
The first version of this panel was assembled automatically, and it was not good enough. A review the same day removed a AAA games publisher, two platforms with millions of users, a labour union, two government agencies and about twenty individual artists with a shop link, and added 28 hand-checked businesses. Their posts no longer count, retroactively. That is why the figures here are lower than the ones published a few hours earlier: the median likes per post dropped from 12 to 6. The share of active accounts moved the other way, from 52 to 59 percent, because a good part of what was removed had never posted at all. The population was wrong, not the arithmetic. It is written down here rather than quietly corrected, because a source that revises silently is worth nothing.
What this measurement cannot tell you
- This is a sample, not a census. Statements hold for this panel, not for Bluesky as a whole.
- The own-domain rule selects for the technically more capable half of the platform. The panel likely overstates activity.
- English dominates: 60 percent of posts declared English, with Spanish second at 16 percent since several Spanish bookshops joined the panel, and Japanese at 5 percent. German-language search terms still find very few accounts using their own domain, so German business behaviour is underrepresented here.
- Public top-level posts only. Replies are filtered out so conversations do not inflate the frequency.
- Engagement is counted at fetch time, so a post from yesterday had less time to collect likes than one from three weeks ago. Future issues will correct this by comparing posts of equal age.
- This first reading is a look back, not a live measurement: the panel was built on 15 August and each account contributed its most recent 30 posts. For ten of the forty-one active accounts that ceiling was reached, so 604 is a floor rather than a count. Medians and the share of active accounts are unaffected, because only the top of the distribution is cut off. Daily collection started on 15 August, and from the next issue on the ceiling is gone.
Next reading
The panel is sampled every night, and the next issue compares the same accounts over the following month. Brandlix publishes and schedules to Bluesky along with nine other platforms, which is why we watch it closely enough to count.