How Many Bots Are on Bluesky? I Counted a Full Day of Posts
Bluesky doesn't publish how many of its 47 million accounts are bots. But among accounts that post, I can estimate it. In a 24-hour replay of every post on the network on November 5–6, 2026, about 0.9% of the 770,048 accounts that posted or reposted either ran on a machine-like schedule or carried Bluesky's automation label. That's about 7,000 accounts active that day, and they wrote about 17% of that day's top-level posts (about 10% once replies count). That's probably low, since my rules can't see unlabeled reply bots or unlabeled bots with random timing.
I looked for a published figure first. News coverage of the late-2024 signup surge reports bots but gives no share, and the AP noted that little data had emerged to quantify the rise in impersonators, AI networks and other harmful content. The closest I found covers one niche: a January–February 2026 analysis put about 19% of accounts around academic paper-sharing at bots. So I measured the whole network: 7.9 million posts and reposts from 04:52 UTC on November 5 to 04:52 UTC on November 6, plus Bluesky's own label on a random 40,000 of those accounts.
(Disclosure: I make Agent Sky, a Bluesky growth tool, which is automation too. More on that at the end.)
What Percent of Bluesky Is Bots?
On November 5–6, 2026, about 0.9% of the accounts that posted, by a count that probably runs low, wrote about 1 in 6 top-level posts. Bots look small by headcount and big by output, because a lyric bot posts 24 times a day and the median account posted or reposted 3 times.
| One day on Bluesky, Nov 5–6, 2026 | Accounts | Share of accounts | Share of top-level posts | Share of all posts, replies included |
|---|---|---|---|---|
| Everyone who posted or reposted | 770,048 | 100% | 100% | 100% |
| Posted on a machine-like schedule | 3,693 | 0.48% | 15.9% | 9.0% |
| Carried the Automated account label (estimate) | about 3,800 | 0.50% | about 6.8% | about 4.2% |
| Either of the two (estimate) | about 7,000 | 0.9% | about 17.4% | about 9.9% |
One account moves those numbers. A labeled bot that spells out numbers wrote 2.7% of all top-level posts by itself. Without it, the last row would be about 15%.
I counted an account as machine-scheduled if it matched either of two patterns:
- Non-stop accounts posted at least 24 times, never had a 3-hour gap, and made at least 80% of their activity top-level posts.
- Clock accounts wrote at least 12 top-level posts, with at least half of the gaps between them within a minute of the same interval, and that interval at least 5 minutes. The most common interval was 60 minutes, then 30, then 2 hours.
My first rules were too loose: anyone posting about every 5 minutes on average, anyone active around the clock, anyone with repeated gaps between posts or repeated text. They flagged 13,490 accounts and 18.6% of all posts and reposts, and when I read a sample, a lot of them were clearly people. So I tightened them.
Then I read the bios and handles of 100 random accounts the final rules caught. It's my own judgment, but here's how they sorted:
- 61 were plain bots: hourly lyric and quote accounts, weather stations, clock towers, data feeds and radio playlists.
- 33 were publishers or promoters on a schedule: newspapers, local TV stations, deal sites and adult promo accounts. A person may run the account, but software seems to do the posting.
- 6 could be a person using a scheduler or just posting a lot.
So "bot" covers a lot here. A newspaper's headline feed counts, and so does an account that tolls a virtual church bell every hour. Many are upfront about being a feed or a bot, though most don't carry the label.
Do Bluesky Bots Use the Automated Account Label?
A few do. Only 17% of the machine-scheduled accounts carried it. Bluesky added the label on March 19, 2026. It shows as a small robot badge on the profile and posts, and it's a choice the owner makes: Settings > Account > Automation label. In my data, 629 of the 3,693 machine-scheduled accounts had turned it on.
Most accounts with the label post far less than the accounts my rules catch. In the random sample, 198 of 40,000 accounts had it (0.495%, or roughly 3,300 to 4,400 across all posters). The median one posted or reposted 6 times that day. Only 30 of the 81 that were active enough to judge (12 or more posts or reposts) matched my rules, so my rules missed most of the labeled bots that posted that much, and a bot with random timing gets past them too. That's why I think 0.9% is low.
Reply bots are a separate gap. Simon Willison wrote in September that automated reply bots have "started manifesting on Bluesky," and he built a checker that scores one account's replies. My rules look at top-level posts, so a reply bot wouldn't match.
And heavy posting doesn't mean bot. 1,466 accounts posted or reposted at least 288 times that day, about one every 5 minutes for 24 hours. Only 159 of them matched my rules. For 895 of them, reposts were at least 80% of what they did. The ones I spot-checked mostly had personal bios. If someone's reposting 300 things a day, they might just be very online.
Are My New Bluesky Followers Bots?
Some may be. I'd look for trouble in follows before I'd look in posts. I replayed every follow for one hour (14:00–15:00 UTC on November 5) and looked up who made them. The scheduled bots above barely appear: only 23 of the 3,693 followed anyone that hour, making 0.4% of all follows. Here's who made them:
| One hour of follows, Nov 5, 2026 | Result |
|---|---|
| Total follows | 89,405, by 21,246 accounts |
| Accounts making 60+ follows that hour | 198 accounts (0.9%) made 38.9% of all follows |
| Accounts making 300+ follows that hour | 19 accounts made 14.0% of all follows |
| Sampled follows from accounts taken down when I checked about 27 hours later | 9.3% (370 of 4,000) |
| Sampled follows from accounts under 7 days old (of the 3,626 from accounts still online) | 20.1% |
| Sampled follows from accounts with zero posts (same 3,626) | 25.5% |
| Sampled follows from accounts with the Automated account label (same 3,626) | 2.2% |
In that hour on November 5, 2026, about 1 follow in 11 by my sample came from an account that was taken down when I checked about 27 hours later. Those 370 sampled follows came from just 67 accounts. Those 67 made at least 7,035 of the hour's follows (7.9%), so the real share is no lower than that. In the same check on my random sample of posters, accounts taken down made only about 0.3% of posts. That's why I'd look in follows first, though the timing differs and Bluesky doesn't say why it removed each account, so this doesn't prove they were spam or bots.
Bluesky had acted on those accounts by the time I checked, but I can't say how much it missed: 182 of the 198 accounts making 60+ follows were still up.
Fast following isn't proof of a bot, and neither is a new account. A person can tap "follow all" on a starter pack and add usually up to 150 accounts in a moment, and 300 follows an hour could be a couple of packs. New users have no posts yet either. Growth tools, including mine, can show up in numbers like these. (Bulk following also has its own label and suspension pattern.) The takedown number is the stronger signal, because Bluesky made that call and I didn't.
Bluesky's 2025 transparency report shows spam and scaled abuse are a large moderation workload. It counts 2.08 million account takedowns in 2025 for all reasons and about 762,000 spam labels, 99% of them automated, and it says that in May 2025 the company "expanded our staffing specifically focused on bots, spam, and other malicious scaled activity."
How Many Bluesky Accounts Are Fake or Empty?
About half of the accounts I could sample show no posts, but that doesn't make them bots. I took 8,000 random accounts from Bluesky's relay, which lists about 25 million of the 47 million accounts and leans toward newer ones. Of the 7,976 still online, 58% showed zero posts, or about 52% after adjusting for signup year. Only 7 carried the automation label.
The likeliest explanation is the boring one: people signed up during a news cycle, looked around and left. A profile with zero posts can't tell you whether anyone was ever behind it, so I won't guess how many are fake.
How Do You Spot a Bot on Bluesky?
Check for the badge, then look at the timing. In order of how much I'd trust each signal:
- The Automated account badge. It's the owner's own statement, so it's the strongest sign, though not a perfect one.
- Posts at a steady interval. Open the profile and read the timestamps. One post an hour, all day and night, points to a scheduler.
- No replies across many posts, for a scheduled bot. Most machine-scheduled accounts had none, though my non-stop rule partly selects for that. It means little on its own: about half of the busy accounts with no replies or reposts didn't match my rules. A reply bot looks the opposite, so run that handle through Simon Willison's reply bot checker.
- Follows almost nobody. 65% of the scheduled accounts follow 10 or fewer. That's normal for a feed, odd for a person.
- A new follower with a days-old account, no posts and hundreds of follows. That's a weaker signal, but it's the pattern I'd check first.
High volume on its own doesn't tell you much, and neither does a strong political opinion. Plenty of people have both.
Harmless bots can stay. For spam followers, block the account or report it as spam. I covered how to get rid of bot followers separately. To block many at once, subscribe to a spam list: Bluesky block lists names the most-subscribed ones. I couldn't find a setting in the app's source code that hides labeled accounts in one go, so for now the badge is only something you can read. Bluesky allows automation, so an account isn't breaking the rules just by being a bot.
How I measured it, and the limits. I replayed Bluesky's public Jetstream firehose from 04:52 UTC on November 5 to 04:52 UTC on November 6, 2026: 3,871,821 posts (2,041,131 top-level and 1,830,690 replies) and 4,054,101 reposts, from 770,048 accounts. That was two days after the US midterm election, so it may have run a little busy. "Top-level posts" means posts that aren't replies. Posts deleted later still count. The rules use only timing and the kind of post, not the text. Profiles came from public getProfiles lookups on November 6, for every rule match plus 40,000 random posters. The label row adds the 629 labeled rule matches, counted exactly, to the sampled labeled accounts scaled up. The "either" row adds all 3,693 rule matches to the same scaled-up accounts. The 47 million is 47.4 million on November 4, per Jaz's bsky-search stats. I scored the 100 accounts from their profiles alone, so that split is a judgment call. I sampled 4,000 of the hour's 89,405 follows at random, so busy accounts count more, and two accounts supplied 150 of the 370. They came from 2,101 distinct accounts, and 67 of those returned AccountTakedown about 27 hours later (3 more had deactivated). The age, zero-post and label rows leave out accounts already gone, which may be the likeliest to be new or empty. Among the 40,000 random posters, 88 returned AccountTakedown on November 6, within about 12 hours of the window's end. The relay sample took 10 accounts from each of 800 random positions in the relay's listRepos list, and I weighted the zero-post figure by rough signup-year totals. One day and one hour can't show a trend.
If you want real people in your follows, I built Agent Sky for that. It finds active accounts in your niche, skips ones with no name, avatar, bio or posts, and checks for a live follow-back before it unfollows anyone. It's automation too, so you set the pace, up to 300 new accounts a day. $9 a month, free to start.
Quick FAQ
What percent of Bluesky is bots? Bluesky doesn't publish a figure for its 47 million accounts. Among the 770,048 accounts that posted or reposted in 24 hours on November 5–6, 2026, I estimate about 0.9% posted on a machine-like schedule or carried Bluesky's automation label. They wrote about 17% of top-level posts. That's probably low, because unlabeled reply bots and unlabeled bots with random timing don't show up in this count.
Is Bluesky full of bots? Not by what I could measure among accounts that posted, but bots are loud. About 1 in 110 accounts that posted on November 5–6, 2026 looked automated, and they wrote about 1 in 6 top-level posts. Of 100 of those accounts I read, 61 were plain bots like quote accounts and weather stations, and 33 were publishers or promoters, including news, deal and adult accounts.
Does Bluesky have a bot problem? I can't say how much of it is bots, but takedowns show up much more in follows than in posts. On November 5, 2026, accounts Bluesky had taken down when I checked made about 9% of sampled follows in one hour and about 0.3% of a random sample's posts that day. A takedown doesn't say why. Bluesky counted 2.08 million account takedowns in 2025 for all reasons, per its transparency report, and said it added staff for bots, spam and other malicious activity at scale.
How do you identify bots on Bluesky? Look for the Automated account badge first. Then check the timing: posts at the same interval, all day, with no replies. A follower that is days old, has no posts and follows hundreds of accounts is worth a closer look. Heavy posting alone proves little. Of the 1,466 accounts that posted or reposted 288 times or more in 24 hours on November 5–6, 2026, most were mainly reposting, which is how many very active people use Bluesky.