Bluesky uses a labeling system that allows third-party services to tag posts and accounts with warnings or flags. Harassment often follows repeated patterns such as coordinated replies, slurs, or dogpiling from multiple accounts. Labelers can automatically detect these patterns and apply visible labels to offending content. This article explains how labelers work, how to subscribe to them, and how to use them to spot harassment on Bluesky.
Key Takeaways: Using Bluesky Labelers for Harassment Detection
- Bluesky Labelers: Third-party services that scan posts and accounts for harassment patterns and apply warning labels automatically.
- Settings > Moderation > Moderation Tools: Where you subscribe to labeler services by entering a labeler handle.
- Label categories: Harassment, spam, impersonation, and other flags that appear on posts and profiles after a labeler detects a pattern.
How Bluesky Labelers Detect Harassment Patterns
A Bluesky labeler is a service that runs its own moderation algorithm. It reads public posts and profiles, then applies labels to content that matches its rules. For harassment detection, a labeler looks for signals such as repeated use of slurs, mentions of the same target by multiple accounts in a short time, or reply chains that contain threatening language. The labeler does not delete content. It only adds a visible tag, such as Harassment or Spam, to the post or account. You as the user choose whether to hide or show labeled content in your settings.
What Labels a Harassment-Focused Labeler Applies
Labelers that specialize in harassment use categories defined by the Bluesky moderation system. Common labels include:
- Harassment: Applied to posts that contain direct threats, slurs, or targeted abuse.
- Spam: Used for accounts that mass-mention users or post repetitive links.
- Impersonation: Flags accounts pretending to be someone else, often used in coordinated harassment campaigns.
- NSFW: Sometimes applied to harassment that includes unsolicited explicit images.
Each labeler can define its own thresholds. For example, one labeler might flag a post that contains two slurs, while another waits for five. You can subscribe to multiple labelers to get broader coverage.
Prerequisites for Using a Labeler
To use a labeler, you need a Bluesky account. You do not need any special permissions. The labeler must be a public service with a Bluesky handle, such as @harassment-detector.bsky.social. You also need to know the labeler handle or find it through a directory. Some labelers are run by independent developers, while others are maintained by community groups.
Steps to Subscribe to a Harassment Detection Labeler
- Open Bluesky Settings
Click your profile picture in the top right corner of the Bluesky app or website. Select Settings from the dropdown menu. - Go to Moderation Tools
In the Settings menu, click Moderation then Moderation Tools. This section lists all labelers you have subscribed to and lets you add new ones. - Enter the Labeler Handle
In the Add a labeler field, type the handle of the harassment detection labeler. For example, type @harassment-detector.bsky.social or the handle provided by the labeler service. Click Add. - Configure Label Visibility
After adding the labeler, you see a list of labels it uses. For each label, choose what happens when content is flagged: Hide, Warn, or Show. To automatically hide harassment posts, set the Harassment label to Hide. - Save and Test
Click Save changes. To test, find a post that the labeler has already flagged. It will show a colored banner with the label name. If you set it to Hide, the post will be blurred or hidden entirely.
If the Labeler Does Not Detect Harassment Correctly
False Positives: Legitimate Posts Flagged as Harassment
A labeler may flag a post that is not harassment. This happens when the labeler uses broad keyword matching. For example, a discussion about harassment might use the same words as actual harassment. To fix this, you can change the label visibility to Warn instead of Hide. That way you see a warning but can still click to view the post. You can also report a false label to the labeler operator through their support channel.
False Negatives: Harassment Posts Not Flagged
No labeler catches every harassment pattern. If you see a post that looks like harassment but has no label, the labeler may not have a rule for that pattern. You can subscribe to a second labeler that uses different detection rules. For example, one labeler might focus on slurs, while another looks for dogpiling patterns. Combining two labelers increases your coverage.
Labeler Goes Offline or Stops Updating
A labeler is a third-party service. It may stop working if the operator shuts it down or if Bluesky changes its labeling API. Check the labeler handle page for status updates. If the labeler is offline, remove it from your Moderation Tools and find an alternative. The Bluesky community often announces new labelers on feeds like Bridgy Fed or community forums.
Bluesky Labeler vs Manual Moderation for Harassment Detection
| Item | Bluesky Labeler | Manual Moderation |
|---|---|---|
| Detection method | Automated algorithm scanning public posts | Human reading and reporting each post |
| Speed | Near-instant after a post is created | Minutes to hours depending on response time |
| Coverage | All public posts from subscribed accounts | Only posts you or your team see |
| False positive risk | Higher due to pattern matching | Lower because humans interpret context |
| User control | Choose Hide, Warn, or Show per label | Full control but requires active effort |
Using a labeler is faster than manual moderation but requires you to trust the labeler operator. For best results, use a labeler as a first filter and manually review any posts marked Warn.
You can now subscribe to a Bluesky labeler to automatically detect harassment patterns. Start by searching for a labeler handle on Bluesky social or through community directories. For advanced protection, subscribe to two labelers with different detection rules and set the most restrictive label to Hide. This creates a layered moderation system that catches more patterns with fewer false positives.