24 May 23
Twitter Inc.

Community Notes on Media

As Community Notes grows its user base, some patterns and behaviors emerge that present significant opportunities for impact. One clear pattern we noticed is that most Community Notes are added to Tweets that feature images or videos. With the advent of Generative AI and Tweets promoting misinformation by use of manipulated media, this is becoming more and more common.

notes on media animated

This trend was really easy to spot by looking at our "frequent words" dashboard. "Pictures", "Images", and "Videos" are the most common words used in notes that other people rate as "Helpful."

top words in helpful notes

We didn't anticipate this at the start of the project years ago, and our system was designed with a strict 1-to-1 relationship between notes and Tweets. This means every Community Note is attached to one Tweet, and can't be applied to other Tweets. When a misleading image or video goes viral, contributors have to manually add their note to multiple Tweets, and those notes have to separately be rated by readers.

The opportunity here was to allow contributors to add a note to an image or video, and have that note automatically be shown on other Tweets containing the same media. In order to implement this, we had to rearchitecture the system to allow for one note to be shown alongside multiple Tweets containing the same image.

notes on media

The UX was also challenging to figure out. The main goal was to make it clear to contributors that they were adding a note to the image, not to the Tweet. We also wanted to make it clear to readers that the note was about the image, and not about the Tweet.

Our update allows contributors to tag their notes as being "about the image" instead of "about the Tweet". When this is selected, a media matching service is triggered that tries to identify other Tweets sharing the same image and shows the same notea across all of them.

notes on media

This was a large engineering project, and even though the under-the-hood workings of media-matching are out of my depth, I'm proud to have contributed not only in design but also implementing the GraphQL and front-end changes required.

For the initial launch, we'll only support Tweets with exactly 1 image, but we'll be working on supporting Tweets with multi-media, videos, and GIFs in the future.

Other projects in Community Notes


This website was built using Obsidian, Eleventy and Vercel.
The text is set in Untitled by Klim Type Co.