X’s community-led approach: tackling inaccurate & misleading information

There are meaningful conversations happening on X right now – with people in every part of the world posting hundreds of millions of times per day on every single topic, relying on X for critical connection, information and access. People want real-time information from voices around the world, and they want accurate information. It’s X’s ambition to make this a reality.

X has long had policies designed to safeguard this platform from bad actors, including teams that address coordinated accounts and similar problems. That remains true. In addition, we believe it’s important to add context to potentially misleading content.

A year ago, we widely launched a new approach to adding context — a community-driven approach, designed to be helpful and informative to people across viewpoints: @CommunityNotes. We want to share what we’ve observed, and what we’re working on.

In this first year — and thanks to the incredible contributor community — we’ve seen remarkable results:

Community Notes now address a far wider range of topics than our historic approaches to addressing misleading information. In the past two weeks alone, notes have been seen well over 100 million times, addressing an enormous range of topics from out-of-context videos to AI-generated media, to claims about specific events and more. When notes are added, they are effective — they’re helpful to people across the political spectrum, they measurably inform people’s understanding, and they lead people to Like and Repost potentially misleading content less. Often, post authors even choose to delete their posts after a note is added.

We believe innovation in this space has been important for the world. Here’s why, and why we are so optimistic about a community-driven approach:

• Trust. A key challenge with traditional fact-checking driven by companies or institutions (and the resultant enforcement actions) is that it can be highly divisive. Significant portions of the public often dismiss such actions as biased. Community Notes was created to deliver context and information in a way that people would trust and feel to be fair — we pursued it when we saw in research how positively people from different viewpoints respond to it. To build trust, Community Notes has two unique features that differentiate it from other approaches: (1) Bridging-based ranking. For a note to become visible to all users on X, it must be rated “helpful” by reviewers who’ve previously disagreed. (2) Open source. The code and all of the data that power Community Notes are open-source, so anyone can audit, analyze, critique or help improve it.
• Speed and scale. Traditional approaches also face challenges with speed and scale. It is challenging to cover the wide range of potentially misleading information out there, and it is common for context to take many hours or days to be added, particularly with new content and claims. We believe a community-led approach has the potential to scale much wider and respond more quickly, and we already see impressive results that continue to improve over time.
• Consistently helpful to people across political and ideological spectrums. Community Notes’ ability to produce visible notes that are broadly found helpful is remarkable given the strong opinions and divisions around the topics on which notes appear. We believe this is key to its effectiveness. To improve this even further, we regularly launch updates to Community Notes’ open-source algorithm, like yesterday’s note scoring improvement, and to the core product experience, like last week’s change that made sources required.

We put a high priority on note quality, as it’s critical to trust and effectiveness. At the same time, we’ve been driving up speed and scale:

• Notes are fast and getting faster, and are now seen 1.5 to 3.5 hrs more quickly than a month ago. Notes in the first few days of the Israel-Hamas conflict appeared at a median time of about 5 hours after post creation. And that doesn’t include the tens of thousands of notes that appear in just minutes on posts that reuse an image or video that has already received a note. We know some people will inevitably see posts before a note appears, so we also send notifications to people who previously engaged with a post that later receives a note. We continue to improve speed with infrastructure improvements, like a recent change to a real-time streaming architecture, and experience improvements, like note previews that help gather ratings on notes quickly.
• Notes are appearing on far more posts, powered by media matching and our expansion to well over 100K contributors in 44 countries and growing. We recently launched and have been enhancing “notes on media,” meaning that when a note is added to a photo or video, it automatically shows on other posts with matching media. In some cases, individual notes match hundreds or thousands of posts. 
• Notes often appear on viral and high-visibility content. This has the benefit of helping to inform a large number of people about a topic, and also allows people to become familiar with the ways in which content can be misleading, so they can better consume future information.

Finally, we listen closely to feedback on how Community Notes can improve. We’ve shipped more than one product update every week on average throughout 2023, and in the last two weeks alone, we’ve shipped eight. These improvements are often based on feedback from the public and contributors, and improve things like note quality, speed and experience. Community Notes regularly get better on a weekly basis.

• We are asked regularly how Community Notes prevent gaming and manipulation. We built Community Notes with the expectation that people would attempt to game it from day one, and designed it to deliver quality notes in the face of this. Notes takes multiple steps to achieve this — from account signup requirements, to writing ability and other capabilities needing to be earned through a track record of helpful contributions, and most importantly to the core algorithm that shows notes only when they’re found helpful by people who disagree. In addition, the open-source code and data make it possible for others to help identify weaknesses or make improvements. We will continue to be vigilant to ensure note quality stays high.
• We’ve heard people suggest there should be a mechanism to address cases where an account frequently receives Community Notes on their posts. We’re evaluating whether an option exists that people would find helpful and fair. It’s important to note that Community Notes sometimes just add informative context to posts, and do not always reflect a post being misleading. We welcome feedback and suggestions on this topic.
• We’ve heard that people want notes to be faster and cover more posts. A community-driven approach can maximize quality, speed and scale, and we’re always working to improve in those areas. Periodically, we’ve been challenged on these dimensions, but those challenges often do not quantitatively compare the performance, quality, visibility and reach of Community Notes to other products, publishers and news organizations. Being open-sourced, we welcome feedback and critiques. And we believe that when it’s done meaningfully and transparently, there can be a great deal of improvements to build better models to address misleading information globally.

We’re a year in, but we’re always on day one. We don’t expect perfection, but we have high ambitions for quality, speed and scale and helping people stay informed. As one person put it in an X post, “It’s not perfect, but it IS working pretty [darn] well.” With the help of our contributors, and feedback from the public, we’re going to keep making it work even more darn well. The world needs it. We’ll continue to innovate, and continue to openly share the data — along with the code — to show our progress.

We couldn’t do this without our contributors and all the people who provide feedback. Thank you, we are grateful, and please keep it coming.

Please follow us @CommunityNotes for more updates.