Every network gets the spam it deserves, and Bluesky’s flavour is cheap: the protocol is open, accounts are free, and a script can create a thousand of them in an afternoon. What follows is how to tell those apart from the very large number of real people who simply behave in ways that look suspicious at a glance.
A signal to review, not a verdict
An account following thousands with very few followers may be worth reviewing. That pattern is not definitive: a new user, an avid reader and an automated account can have similar counts. Open the profile and consider the behaviour before deciding.
Everything else on this page is weaker than that, and the difference matters.
The signals that are not enough on their own
These get quoted constantly as bot tells. Each of them describes an enormous number of ordinary accounts:
- No avatar. Plenty of people never get round to it. On a network where the default avatar is a plain coloured circle rather than a shaming grey silhouette, there is even less pressure to.
- No posts. Lurkers are the majority of any social network. Someone who reads for a year before posting is the normal case, not the anomaly.
- No display name. The handle is often enough, and some people prefer it.
- A recent join date. Bluesky is still adding people in waves. A month-old account in a growth wave is unremarkable.
- A generic handle.
name1234.bsky.socialis what you get when the handle you wanted was taken.
Any one of these on its own tells you almost nothing. Two or three together start to mean something. An account with no avatar, no display name and no posts that followed you the day it was created is a different proposition from an account with no avatar and four years of posts.
This is exactly why Blue Horizons stopped flagging on a single signal. The “likely bots” filter needs either the mass-follow pattern on its own, or at least two of the weak signals at once. It used to accept any one of them, and it swept up a great many real people.
It also ignores accounts whose profile has not been read yet. Two of the three weak signals are absences, no avatar and no display name, and a row the app has seen only as a follow record is nothing but absences, so counting it would be a judgement about what we have not looked at rather than about the account.
The ones the data cannot see
Some of the worst accounts look perfect on paper: a stolen avatar, a plausible bio, a normal ratio, and a feed full of scraped posts. No counter will catch those. What catches them is reading two or three of their posts, which is why every account this tool flags is a suggestion and never an automatic action.
Worth checking by hand when something feels off:
- The posts are all replies, all short, all in the same shape.
- The bio links somewhere odd, or a link-in-bio service you do not recognise.
- The account posts at inhuman intervals, exactly on the hour, all day.
- Several accounts share the same bio, word for word. Searching a distinctive phrase from it usually surfaces the whole cluster.
What to actually do with them
A bot following you is not a problem in itself. It costs you nothing, does not appear in your timeline, and inflates a number nobody should be reading anyway. The reasons to act are narrower than people assume:
- Block, do not just remove. For an account that is harassing, scamming or impersonating, blocking is the tool. It is the only action that stops them interacting with you.
- Report the ones that are clearly coordinated. Bluesky’s moderation acts on patterns, and a report on one member of a spam cluster is worth more than silently blocking it.
- Unfollow the ones you are following. This is the part that affects you: a bot you follow is a hole in your timeline. A bot that follows you is not.
- Leave the rest. Chasing every fake follower is a hobby, not audience management, and blocking in bulk is how people end up mis-blocking real readers.
The number you are tempted to fix
There is a strong pull to “clean” a follower count so it reflects only real people. Resist it, for two reasons. Your follower count is not a metric anyone is scoring you on, and removing followers cannot be undone from your side: if you block to remove and later unblock, they are not following you again.
The list that repays attention is the one you control, which is the accounts you follow. That is what decides what you read tomorrow. There is a practical order of operations for pruning it, and a separate guide on why the follow graph is public and what that lets anyone compute about you.
See where you stand: analyze any handle, free and without signing in, and review your one-way relationships. Check activity inside the app before interpreting a missing post date.
Keep the result private and proportionate
A warning filter is a way to prioritise your own review, not evidence for publicly accusing a person of being fake. Missing information may simply mean the profile has not been read. Some automated accounts are useful and clearly describe what they do. Review concrete behaviour and use Bluesky’s reporting tools for actual abuse. Unfollowing changes your outgoing relationship; it neither removes a follower nor guarantees that they cannot read public content.