@Grok is this true?
AI's rise as the arbiter of 'truth' on X
Last week, Grok, the language model developed by xAI and now tightly embedded in X, began surfacing disturbingly positive references to Hitler. These responses appeared repeatedly across different user interactions, often in reply to posts that were entirely unrelated. What initially seemed like isolated incidents quickly formed a visible pattern, spanning multiple threads. As more examples came to light, it became harder to dismiss, and the story quickly spread.
The incident revealed more than a simple lapse in xAI’s guardrails. As X’s feed grows louder, faster, and harder to verify, users are increasingly turning to Grok for clarity and context. “@Grok is this true?” has become a familiar refrain on the platform. This behaviour is driven by the perceived authority of Grok’s brand, reinforced by how quickly it replies and how closely it sits alongside users’ own posts. It is ever-present, responds instantly, and does so within the thread itself, visible to all. In a world where information moves faster than it can be verified, Grok fills the gap left by slower human verification.
Grok, now leading performance charts with Grok 4, has caught up to ChatGPT despite its late start. But its significance goes far beyond its standing in the AI benchmarks. On X, it has become an interpreter of the feed and, increasingly, an arbiter of truth. This creates a central tension: the system users now rely on to resolve confusion is also capable of creating it, as the Hitler incident made clear. The same AI users increasingly consult to clarify reality can also amplify ambiguity, or worse, reinforce dangerous ideas.
From Community to AI Verification
Grok’s rise as a de facto fact-checker may or may not be the result of deliberate product design, but it is certainly a byproduct of deeper structural changes at X. Since the company was folded into xAI, language model capabilities have become more tightly woven into the platform. As that integration expands, Grok is increasingly taking on responsibilities that were once handled by human moderators and distributed human communities.
X’s search for faster and more adaptable forms of content verification is not new. In January 2021, following the US Capitol riots and growing concern over misinformation, the company launched Community Notes, originally called Birdwatch. It was designed as a scalable, community-driven alternative to professional fact-checking, built on the idea that distributed consensus from users could address false or misleading claims more quickly and transparently.
Meta made a similar move earlier this year, ending its US fact-checking programme in favour of its own notes-based system. X though appears to be two steps ahead, moving beyond it.
Community Notes was built for a different internet: slower, more static, and less generative. It is led by people and supported by tools. Grok represents something else. It is real-time, automated, and always present. It is summoned by users and powered by AI. It does not just support people’s understanding of the feed. It is increasingly shaping it.
The Significance of Asking Grok
X has become the central stage for high-stakes, real-time political discourse. Donald Trump normalised the idea that public figures could speak directly to citizens through social media. Today, leaders like Macron, Modi, and Zelensky post regularly on the platform, often appearing alongside figures one might not have expected to see there before, like Iran’s Supreme Leader Khamenei.
These political voices now compete for attention in the same feed as influencers on both the right and the left, journalists, celebrities, brands, news outlets, bots, anonymous users, and OSINT accounts. Everyone is participating in the same space, speaking into a single continuous stream that makes little distinction between high and low integrity. Important declarations on matters of policy or conflict are interwoven with memes, provocation, and mis and disinformation. .
In this environment, context is king, and Grok increasingly serves the role of providing it. It appears at a moment’s notice. Journalists still help users make sense of the world, but online, and especially on X, that role is quickly being replaced by a language model. Grok claims to help users navigate a convoluted stream of conversations from a wide array of voices. As it attempts to do that, it is becoming a key part of the conversation itself.
Why We Have Arrived Here
Grok’s growing role as an interpreter of truth reflects a broader shift in how platforms manage scale. X now processes thousands of posts per second, a volume far beyond the reach of human moderation.
At the same time, human behaviours around content consumption are evolving. Algorithmic feeds and short-form video have cut attention spans and conditioned users to expect instant responses to every query. People swipe rapidly from one post to the next without pause.
In this environment, older systems struggle to keep up. Human and community-led verification requires time. It depends on sourcing, deliberation, and editorial oversight, all of which move more slowly than the feed itself. Community Notes, for example, takes an average of five hours to process a claim. By the time a fact-check appears, the narrative has often already moved on. Users rarely return to revise what they have already absorbed. For verification to shape public understanding, it appears it must happen in real time.
Declining trust in traditional news reinforces the shift to X and to Grok. Institutions that once served as guardians of truth can now feel slow and distant. Grok, by contrast, is immediate and conversational. Users can question it, follow up, and explore tangents. It speaks in the tone and cadence of the feed. It feels personal and reliable.
X is quickly adapting to this new environment, which also serves its core business. Each interaction with Grok results in an AI-generated reply, which contributes to the feed, fuelling more content and longer engagement. Grok’s witty tone helps its replies spread, driving yet more discourse. But this amplification can be volatile. When Grok gets it wrong, the effects do not fade quietly. They cascade.
What Makes Grok Different
What distinguishes Grok from other language models is that it fuses its foundational training data with real-time access to X’s firehose of content. This integration allows it to generate responses that are both informed by general knowledge and immediately contextualised by what is happening on the platform.
This gives Grok a unique advantage over traditional fact-checking systems, but also presents a serious vulnerability. The X firehose contains everything: reliable reporting, speculation, satire, misinformation, outright disinformation, and spam. As Grok draws from that stream, it can absorb the best of the platform but also the worst. It is susceptible to hallucination and bias, as is the case with all language models, but also to the amplification of questionable content already circulating widely. The same real-time access that makes it useful also makes it fragile.
This stands in contrast to systems like human fact-checking, which rely on transparent editorial processes and deliberation, and Community Notes, which relies on human consensus. These systems move slowly but are grounded in human-led processes. Grok, by contrast, operates with speed and carries an air of authority that stems from its placement at the heart of the feed, its branding, and its responsiveness.
This perceived authority plays a central role in how Grok is received. Grok is deeply embedded in the platform and carries the weight of its association with Elon Musk, one of the world’s most influential figures, with 222.5 million followers on X. It is fair to assume that many users who share Musk’s worldview see it not merely as an AI assistant but increasingly as a platform-endorsed interpreter of what is true and what is not. The more frequently it appears, the more its credibility is taken for granted.
That was the lesson of the Hitler incident. Once Grok’s answers appeared in the feed, they spread quickly. Its confident tone and prominent placement lent the response a sense of legitimacy. The replies were reposted, repeated, and treated by some as credible. The very features that make Grok useful, such as its speed, fluency, and visibility, also make its failures harder to contain. When a flawed or morally fraught answer is presented with the same authority as any other, it does not simply fade away. It circulates widely and gains traction.
Implications for Public Discourse
The shift toward AI-mediated interpretation is especially consequential for younger audiences. According to the Reuters Institute Digital News Report, 44 percent of people aged 18 to 24 now rely on social media as their primary news source. As language models become native to these platforms and are embedded directly into their interfaces, they will shape how an entire generation understands the world. Those who control the models, through the choice of training data, the design of system prompts, and the architecture of their guardrails, will hold immense power and responsibility.
Grok on X appears to be just the starting point. Signs of this broader trend are already emerging elsewhere. On WhatsApp, for example, users can forward messages to bots powered by ChatGPT or Perplexity to receive an AI-generated reply that puts the message in context. These interactions are private and typically confined to one-on-one threads, which makes them less likely to ripple outward. Still, they point to a growing fusion between human and machine communication within social networks.
It is not unreasonable to assume that other platforms like Reddit, TikTok, and Instagram will follow suit in time and embed language models directly into their feeds and comment sections. In the near future, social media may very well become one of the main distribution channels for AI. Not the browser. Not the operating system. But the networks where people already live their digital lives.
The Path Forward
For now, Grok holds a privileged place on X. Its brand, backed by Musk, lends it visibility and influence. Its access to X’s firehose of content allows it to ground responses in real time, as things change by the minute. Its proliferation on the platform creates a flywheel that drives further use.
But this dominance may not last. X now allows developers to build their own language model-powered bots. Grok is no longer the only interpreter. Users can already summon alternatives, from Perplexity to Gork, a parody account created for humour and satire.
And this is just the beginning. As powerful open-source AI models become easier to fine-tune and deploy, more organisations and individuals may introduce their own models directly into the feed. X will face a choice: permit an open ecosystem of AIs or maintain tighter controls. Openness could drive platform engagement, but also introduce competition for Grok. A more closed system would reduce risk.
That trade-off will shape how platforms choose to govern. In an open environment, X can oversee Grok, but it may struggle to regulate third-party AI models. Addressing this will require new frameworks for evaluating, approving, and overseeing models deployed by external developers, and naturally, for applying appropriate guardrails to a platform’s own AI offerings as well. It may be unrealistic to reject AI’s role in social platforms altogether, so the task is to define its boundaries.
A platform-endorsed model like Grok, even with its ties to a corporate parent and its perceived leanings, might be seen as a neutral mediator — helping people on opposing sides of an issue find common ground. Its authority as a non-human actor could allow it to broker understanding in ways that human moderators cannot. An open ecosystem of coexisting models might further support this, offering a wider range of perspectives and exposing disagreement in more productive ways. Of course, there are two sides to every coin. These systems can just as easily reinforce bias, amplify division, or elevate the wrong voices. But they also hint at new possibilities for how disagreement might be surfaced and mediated online.
Human moderation will remain essential, as will community-led systems. But these must evolve to operate alongside language models rather than in opposition to them. Human involvement should scale with the potential for harm. Some content may be safely verified by AI, while claims with greater consequences should be subject to human or community oversight. Not all content carries the same weight, so it should not be treated the same. The real challenge is deciding where human judgment is most necessary.
The rise of Grok within X’s feed marks a turning point in how people will construct their understanding of the world as they consume information online. As language models begin to participate in the conversation, the boundary between human editorial judgment and automated response will become harder to distinguish.
The question is no longer whether AI will influence public discourse. It is how much authority we are prepared to give it, and what happens when machines move from merely organising the feed to actively contributing to it and steering its direction.