While artificial intelligence and crowdsourced platforms have changed how we verify information—offering remarkable speed—they still fall short when it comes to reliability and transparency. Until automated systems can successfully balance speed, trust, and clear explanations, human verification remains indispensable.
A functioning society relies heavily on shared facts and transparent public discourse. Public frustration often stems from the sense that these foundational systems are breaking down.
Although modern fact-checking gained momentum roughly a quarter-century ago, the landscape shifted dramatically in early 2025. Meta scaled back its third-party fact-checking programs, with leadership claiming independent reviewers had become overly politicized, though economic and political motivations also played a role. Following Meta’s lead, other major platforms reduced support, leading to a contraction of the global fact-checking community as voluntary and financial backing dried up.
In response to a flood of AI-generated media and fast-moving breaking news, traditional media outlets have evolved their approach. Broadcasters like the BBC and Australia’s public network replaced older, partnered fact-checking units with specialized verification desks focused on authenticating digital imagery and video.
Today, three distinct verification models are competing: expert verification, AI-driven tools, and crowdsourced context. Each method has notable limitations regarding trust, clarity, speed, or cultural context.
Grok is the warning sign
X’s push toward its AI assistant, Grok, as an in-platform fact checker has yielded concerning results. Research analyzing tens of thousands of posts during major geopolitical conflicts revealed that Grok frequently failed to distinguish genuine footage from fabricated media, occasionally misidentifying fake videos of damaged infrastructure as real. Because large language models generate confident responses without built-in caution, users are often misled.
Similarly, automated detection tools designed to catch AI-generated text, audio, and images face severe limitations. Whether relying on statistical writing patterns, metadata analysis, or watermarks like Google’s SynthID, these tools remain narrow in scope and prone to circumvention.
AI verification struggles fundamentally with explainability. When an algorithm outputs a simple confidence score, users cannot cross-examine its reasoning the way they can review a human fact-checker’s sources. When automated detectors mislabel authentic material as synthetic, public trust is inevitably undermined.
The power of crowds
Crowdsourced alternatives, such as community-driven context notes and online forums, offer a different approach. Studies show that these contributors frequently rely on established journalism, professional fact-checks, and encyclopedias. When context notes are successfully attached to misleading posts, they significantly reduce likes and reposts, often maintaining high standards of accuracy for major public health claims. However, this process takes time, and viral misinformation often spreads much faster than crowdsourced corrections.
On the positive side, crowdsourced verification excels at local context—what is accepted as authentic in one digital community may be rejected in another. Peer review rooted in community awareness also faces fewer criticisms of top-down censorship.
Ultimately, a critical question remains: Who decides what is true, how are those determinations audited, and what follows once a decision is made?
In the past, slow and transparent verification processes were acceptable because the methodology was visible. Today, however, those methodical institutions have been eclipsed by a handful of tech platforms operating behind opaque, algorithmic systems that often reward outrage.
While initiatives like national media literacy strategies are positive steps forward, education regarding AI content must look beyond simply spotting fakes.
If automated tools and crowdsourced notes are replacing traditional fact-checkers, media literacy must expand to include governance literacy. Users need to understand who holds decision-making power, what processes are used, and what financial or social incentives drive those choices. Crucially, future AI solutions must be designed to reflect the diverse, real-world contexts in which people actually experience technology.
The current rush toward automated detection tools and AI chatbots highlights a societal desire for speed and simplicity. While human fact-checkers may operate more slowly and face skepticism, automated systems have yet to match their level of trust and explainability—leaving human verification as an essential safeguard.
Ned Watt is a post-doctoral research fellow at the QUT GenAI Lab in Brisbane. Michelle Riedlinger is associate professor in the School of Communication at QUT.