This repo contains code and data about the usage of "alt text" image descriptions on Bluesky.
See the blog post for more information on the project: Image descriptions on Bluesky: not bad, could be better. See also the web-based Bluesky Alt Text Stream for a streaming view, if you don't need to save the data.
The repo contains three parts:
- Actual data, collected in July 2025
- Data collection from the Bluesky Firehose (top-level +
src), in Node - Analysis of the resulting data, in Python
The files image_posts.csv and top_100_users.csv are data collected for this project.
image_posts.csv is a collection of 2.8 million Bluesky image posts and their corresponding image descriptions (alt text).
The image description is an empty string if it was not present.
This file was generated using index.js (see below, Data collection)
during the period from July 14 to July 21, 2025.
Note that this dataset contains sexually-explicit image descriptions.
top_100_users.csv is a hand-labeled collection of the top users (users with the most posts) in the dataset.
I used this to filter out the most frequently-posting bots from the dataset.
Requirements: node - I'm using v16.15.0 and it's untested on other versions.
To collect data, run the following:
node index.js
Note this will overwrite the existing image_posts.csv.
If you don't need to save the data but just want to view the live activity, check out the web-based Bluesky Alt Text Stream instead.
The analysis folder contains Python code for a basic analysis.
See the README in that folder for more info.
This project is inspired by the 2019 paper “It's almost like they're trying to hide it”: How User-Provided Image Descriptions Have Failed to Make Twitter Accessible, by Cole Gleason, Patrick Carrington, Cameron Cassidy, Meredith Ringel Morris, Kris M. Kitani, Jeffrey P. Bigham.
See my blog post for more information.