Bluesky PBC has published detailed documents about how its recommendation algorithms work. Many users want to know why certain posts appear in their feeds and who controls the algorithm. Bluesky’s transparency reports explain the ranking signals, data sources, and user controls that shape the experience. This article covers what Bluesky has disclosed about its algorithmic system and how you can verify the claims.
Key Takeaways: Bluesky’s Algorithm Transparency Disclosures
- Open-source algorithm code on GitHub: Bluesky publishes the ranking logic for the Discover and Following feeds in a public repository.
- Settings > Moderation > Muted Words & Muted Users: Users can directly control which content the algorithm shows by hiding specific words or accounts.
- Custom feeds via the AT Protocol: Any user can build and share a feed using a different algorithm, bypassing the default ranking entirely.
What Bluesky PBC Has Published About Algorithm Transparency
Bluesky PBC operates as a public benefit corporation. Its transparency policy covers three main areas: the source code of its recommendation engine, the data used to train and tune the algorithm, and the user-facing controls that let people modify their feed. The company has released the code for the “Discover” feed and the “Following” feed under an MIT license on GitHub. The repository includes the ranking model, the scoring functions, and the integration with the AT Protocol’s relay network.
The algorithm uses signals such as recency, interaction count, and the user’s own interaction history. Bluesky does not use external data brokers or purchase demographic information. The company has also published a white paper titled “Algorithmic Choice in Social Media” that explains the design philosophy behind giving users control over their feed ranking. This white paper is available on the Bluesky blog and describes how the system can be replaced entirely by a third-party feed provider.
Bluesky’s transparency reports are updated quarterly. These reports list the number of moderation actions taken, the categories of content removed, and the appeals processed. The reports do not include raw training data sets because the algorithm is largely hand-tuned with rule-based scoring rather than machine learning. The company states that it does not use engagement-based optimization that maximizes time on site. Instead, the ranking aims to show content the user is likely to find relevant based on explicit signals like likes, reposts, and replies.
Where to Find the Published Materials
The main algorithm code lives in the bluesky-social/feed-generator repository on GitHub. The README file explains how to set up a local feed generator. The algorithm documentation is also linked from the Bluesky Help Center under the “Feeds and Algorithms” section. The quarterly transparency reports are posted on the official Bluesky blog at bsky.social/about/blog. Each report is a PDF with a summary of moderation data and algorithm changes.
Steps to Review the Algorithm Code and Controls
You can verify Bluesky’s algorithm transparency claims directly by accessing the code or by using the built-in controls. The steps below show how to inspect the default ranking logic and how to change your feed to use a custom algorithm.
View the Default Algorithm Code
- Open the feed-generator repository
Go to github.com/bluesky-social/feed-generator. The repository is public and does not require a login to view. - Locate the ranking logic
In the repository, open the src folder and then algos. You will see files for discover.ts and following.ts. These files contain the scoring functions that rank posts. - Read the scoring comments
Each function includes comments that explain what each signal does. For example, the Discover feed assigns a base score of 1.0 to a post and then adds 0.5 for each like from a user you follow. The code is written in TypeScript and can be run locally with Node.js.
Switch to a Custom Feed
- Open the Bluesky app
Launch Bluesky on your phone or desktop. Log in to your account. - Go to the Feeds tab
Tap the Feeds icon at the bottom of the screen. This shows your current feeds and available custom feeds. - Browse or search for a custom feed
Tap Find Feeds and search for a feed that uses a different algorithm. For example, the What’s Hot feed ranks by total interaction count, while the Skywatch feed uses a machine learning model. - Add the feed to your list
Tap the + button next to the feed name. The feed appears in your Feeds tab and becomes selectable as your primary feed.
Common Misunderstandings About Bluesky Algorithm Transparency
“The algorithm is fully open-source and anyone can audit it”
The code is open-source, but the production deployment uses configuration files and data sources that are not published. For example, the list of accounts that are globally muted or the blocklist used for spam detection is kept private to prevent abuse. The ranking logic itself is transparent, but the operational data is not fully public.
“Bluesky does not use any machine learning”
The default feeds use rule-based scoring, but third-party feeds can use machine learning. Bluesky itself does not train a central ML model for its own feeds. The AT Protocol allows any feed generator to use any algorithm, including neural networks. Some popular custom feeds do use ML, but those are not part of Bluesky PBC’s transparency disclosures.
“Transparency reports include all moderation actions”
The quarterly reports include aggregate numbers, not per-action details. For example, the report states how many posts were labeled as spam but does not list the specific posts or the accounts that were moderated. Bluesky says this protects user privacy and prevents gaming of the moderation system.
Bluesky Algorithm Transparency vs Other Platforms
| Item | Bluesky | X / Twitter |
|---|---|---|
| Algorithm source code | Public on GitHub for default feeds | Not public |
| User control over ranking | Full replacement via custom feeds | Limited to “For You” vs “Following” tabs |
| Quarterly transparency report | Published with moderation and algorithm changes | Published biannually with limited algorithm details |
| Data used for training | No external data; uses interaction signals only | Uses demographic and behavioral data from other services |
| Third-party feed creation | Supported via AT Protocol without approval | Not supported |
Bluesky’s transparency model gives users more direct access to the algorithm code and the ability to replace the feed entirely. X provides less visibility into its ranking logic and does not allow third-party feed generators. The trade-off is that Bluesky’s default feeds are simpler and may not surface as diverse content as X’s machine learning system.
Conclusion
You can now review Bluesky’s algorithm code on GitHub and switch to a custom feed that uses a different ranking method. The company publishes quarterly transparency reports that summarize moderation and algorithm updates. To go further, try building your own feed generator using the open-source repository and the AT Protocol SDK. This gives you complete control over what appears in your feed.