Evidence Pack: Does Crowdsourced Fact-Checking Actually Reduce Misinformation?

Over the past three years, the architecture of truth on the internet has undergone a quiet revolution. Moving away from centralized teams of professional fact-checkers, major social platforms have increasingly outsourced reality-checking to their own users.[6]

The model, pioneered by X (formerly Twitter) as "Community Notes" and adopted by Meta in early 2025 for Facebook and Instagram, allows ordinary users to append contextual corrections to misleading posts. But as this decentralized approach becomes the default standard for billions of users, a critical question has emerged: does crowdsourced fact-checking actually work?[4]

According to a wave of recent empirical studies, the answer is a qualified yes. When crowdsourced notes successfully attach to a post, they are remarkably effective at halting the spread of misinformation.[2]

A comprehensive study published in the Proceedings of the National Academy of Sciences (PNAS) analyzed the diffusion of tens of thousands of posts. Researchers found that once a Community Note is appended, engagement plummets. Reposts drop by an average of 46%, and likes fall by 44%.[2][5]

The mechanism behind this success is not just the presence of a label, but the context it provides. A 2026 study in PLOS One demonstrated that crowdsourced fact-checks are just as effective as expert fact-checks at reducing a reader's confidence in false claims.[1]

Crucially, users across the political spectrum report higher trust in these community-driven notes than in traditional, top-down "misinformation" warning labels. Because the notes provide transparent explanations and source links, they mitigate the partisan reflex to dismiss fact-checks as platform bias.[1]

The engine driving this trust is known as a "bridging algorithm." Rather than simply publishing the notes that receive the most upvotes, the system requires cross-partisan consensus.[3]

The algorithm analyzes the past voting behavior of contributors to estimate their latent ideological positions. For a note to be published, it must be rated as "helpful" by users who typically disagree with one another. If a left-leaning user and a right-leaning user both agree that a note adds necessary context, the system deems it highly credible and attaches it to the post.[3]

However, this strict requirement for consensus is also the system's greatest vulnerability. While the notes are highly effective when they appear, they rarely do.[3]

Data reveals a severe bottleneck in the crowdsourcing pipeline. Across millions of proposed notes, only about 8.3% ever achieve the cross-partisan consensus required to become visible to the public. The vast majority of user-submitted corrections remain hidden in the backend, perpetually waiting for more ratings.[4][6]

Speed is another critical failure point. On average, it takes between 14 and 26 hours for a proposed note to gather enough diverse votes to be published. By the time the correction appears, the misleading post has typically already reached 80% of its total audience, passing its peak virality.[3][4]

Furthermore, the bridging algorithm systematically struggles with the most toxic and divisive content. A 2026 analysis found that highly polarizing claims—particularly those surrounding major elections—frequently fail to generate cross-partisan agreement. Because partisans are less likely to agree on the facts of a deeply ideological dispute, these posts remain under-moderated by design.[3]

These limitations were central to a March 2026 policy advisory opinion by the Meta Oversight Board, which evaluated Meta's plans to expand its own community notes program globally.[4]

The Board concluded that while the system enhances freedom of expression and works well in stable democracies, relying solely on crowdsourced consensus poses severe human rights risks in fast-moving crises or repressive regimes where coordinated networks can game the voting system.[4]

Ultimately, the evidence suggests that crowdsourced fact-checking is a powerful, trust-building tool that genuinely changes user behavior. But as a standalone defense against viral misinformation, its algorithmic demand for consensus makes it too slow and too sparse to replace comprehensive moderation.[6]