AI Brings Back Public Trust
User tapping AI icon, September 2026. (Photo by Matteo Della Torre/NurPhoto.)
From the printing press to the radio, from television to social media, communication technologies affect politics and broader society by shaping two things: who speaks and what they say.
In the first case, different technologies vary in the extent to which they favor elite gatekeepers. Most famously, the printing pressdestroyed the informational monopoly enjoyed by European monarchs and the Catholic Church, enabling the Reformation. Much later, radio and television partlyrestored centralized control. Because they were initially expensive to produce and tightly regulated, they tended to concentrate ideological power in the hands of wealthy, well-connected elites.
Of course, who speaks influences what gets said, but themedium also shapes the message in other ways. Printpermits careful, detailed argumentation. Televisionfavors confident sound bites. Social media oftenrewards division, conflict, and negativity.
Large language models (LLMs) like ChatGPT and Claude are the latest communication technology to reshape public opinion, and their effects could be just as profound. Social media decentralized the information environment and weakened the influence of expert opinion. LLMs are likely to have the opposite effects.
Social media has been a radicallydemocratizing technology. It allows anyone with opinions and an internet connection to bypass traditional gatekeepers, a development that has dramatically expanded the range of voices and viewpoints that can be expressed.
It has also reshaped how media influence works. Because the algorithms that recommend content are optimized to capture audience engagement, they often amplify sensationalist, alarming, and divisive messages. Meanwhile, rapid audience feedback through likes, reposts, and comments has made political punditry much moreperformative and vulnerable to audience capture.
Unsurprisingly, the decline of elite gatekeepers has increased the influence of popular ideas marginalized by elites, another term for which is “populism.” Social media benefits populism not by brainwashing the masses with viral fake news, but byexposing voters to widespread non-elite perspectives and making it easier to mobilize around them. In Western liberal democracies, that means perspectives that conflict with the liberal establishment’s technocratic progressivism, including xenophobia, conspiracy theories, and quack science.
At the same time, the performative, engagement-maximizing character of social media has made much of political discourse more stupid and sensationalist. It has also given the world Candace Owens and Andrew Tate. It is probablyno coincidence that the social media era has been an era of populism.
Consider a topic: climate change, vaccines, immigration, crime, tariffs, wealth inequality, the Epstein Files, whatever happens to be in the news. Fire up one of the leading LLMs—ChatGPT, Gemini, Claude, even Grok—and ask for information about it. Now compare the response with the information you can find about the topic by scrolling on a major social media platform.
I suspect it will quickly become clear that the LLM’s responses are generally much more evidence-based and in line with expert consensus than what you get from most social media posts. And when there is no expert consensus, you will typically get a good survey of the range of informed opinion on the topic.
Is this merely a hunch? In many ways, yes, but it aligns with several bodies of evidence suggesting that LLMs arebecomingincreasinglyeffective at producing broadly accurate information on politically relevant topics, especially when augmented with search tools. Inrecent experiments involving thousands of participants, for example, using chatbots to research political claims increased belief in true information whilst reducing belief in misinformation as effectively as internet search.
Why is this? The short answer is that the major AI companies are competing to build the most intelligent, impressive, and useful systems possible for a vast and diverse user base, including businesses that depend on reliable and factual information. This goal—reaping huge profits by putting “expert-level intelligence in everyone’s hands”—cuts against producing systems that deliver highly partisan, ideological, or misinformative content. So do the reputational and legal risks that arise if those systems produce dangerous or demonstrably false information.
There is endless popular and academic hand-wringing about bias, hallucinations, deepfakes, AI-based disinformation, AI psychosis, and other threats. These are all important issues, but a discourse restricted to such issues is missing the forest for the trees. When we consider the large-scale impact of this technology on public opinion, its most consequential feature is simple: it greatly improves people’s access to accurate, evidence-based information.
Whereas social media has been a democratizing technology, LLMs are a technocratizing force. They shift influence back towards expert opinion.
Over a century ago, Walter Lippmannargued that, because the modern world is too vast and complex for anybody to understand through first-hand experience, we’re forced to rely on intermediaries to become informed. For Lippmann, the only intermediaries who can reliably perform this function are experts in the broadest sense: trained professionals who adhere to rigorous epistemic norms and methods. He therefore advocated for institutionalized “intelligence bureaus” that would assemble and explain the actual facts for both politicians and the public.
In many ways, the development of Western democracies after the Second World War followed Lippmann’s vision. Of course, this “expert knowledge” wasmixed with elite bias, blind spots, and the occasional catastrophic fuck-up, from the Vietnam War to the 2008 financial crisis. Nevertheless, when it came to the kind of information that guided policy and circulated through the most influential media channels, the pendulum had swung a long way towards technocracy.
Social media is one of several forces that have disrupted this situation. LLMs push in the opposite direction. They are a kind of anti-social media, producing information heavily skewed towards expert opinion and delivered in a nuanced, measured voice. And there are reasons to think they will be more effective than all-too-human technocrats at shaping public opinion.
First, they can rapidly deploy encyclopedic knowledge to answer people’s idiosyncratic questions, and their responses can be probed and questioned without their ever getting tired or frustrated. They won’t just tell you that there is no persuasive evidence for a link between vaccines and autism. They can carefully walk you through the evidence and address your specific sources of skepticism.
Thispartly explains why they can be highly persuasive, even incorrecting conspiratorial beliefs that many assumed were beyond the reach of rational persuasion. For example, onestudy published in _Science_ found that talking with GPT-4 Turbo reduced participants’ confidence in their own chosen conspiracy theories by roughly 20% on average, an effect that was still present two months later.
Second, LLMs typically share information politely and respectfully. Being human, experts are often biased, partisan, and simply annoying, and when they seek to “educate” the public, itcan be perceived—and is sometimes intended—as condescending and rude. LLMs can deliver expert opinion without such status threats.
As Dylan Matthewsargues, this technocratizing character also makes LLMs an epistemically converging technology, pushing “people’s senses of reality closer together in a sort of mirror image of the way social media has fractured them.”
For these reasons, I speculate that, at least in liberal democracies where governments don’t exert significant control over LLMs, their most consequential impact on public opinion will involve technocratization: shifting people’s beliefs towards expert opinion. For example, a _Financial Times_analysis by John Burn-Murdoch, which tests LLM responses using simulated conversations, finds that the most widely used chatbots steer conversations towards moderate, expert-aligned positions. Drawing on tens of thousands of political survey responses, he estimates that ChatGPT, Gemini, and the Chinese model DeepSeek nudge users towards a center-left worldview and away from fringe left-wing positions, while Grok guides conversations about politics towards the center-right and away from hardline right-wing positions. Compared to social media, where conspiracy theories about rigged elections and links between vaccines and autism are overrepresented among posters relative to the general population, the chatbots almost never endorse them. And, importantly, these effects hold even after accounting for partisan differences in chatbot usage and their sycophantic tendencies.
In many cases, this shift will occur when people consult LLMs directly, but it might also be mediated by thegrowing deployment of LLMs as fact-checking tools on social media platforms themselves, and by writers, journalists, and other content creators who use them for feedback, brainstorming, and editing.
The impact of this shift on society and politics could be very large. As with social media, whose effects on political discourse and public opinion have been gradual and cumulative, LLMs need not transform everyone’s worldview and intellectual habits overnight to have profound long-term consequences for the health of media and public debate. That provides some reasons for optimism about the prospects for a more informed, rational style of politics.
Many find this optimism preposterous. To them, it sounds more like Big Tech propaganda than responsible analysis. So, let’s consider the most common objections.
_**Hallucinations**_
One common worry is that LLMs frequently “hallucinate,” generating content that is false or fabricated. But the rate of hallucinations has beenfallingfast, and the main question to ask about their reliability is: compared to what? Compared to the alternative sources of information most people are likely to draw on, especially social media, it is plausible that LLMs typically provide more accurate and evidence-based information.
_**Sycophancy**_
A more serious objection concernssycophancy. LLMs often flatter the self-image and prejudices of those who use them, partly because people prefer warm, agreeable models. This can lead them to generate false and misleading information when users have mistaken beliefs, and even to reinforce and radicalize those beliefs, as seems to have happened in rare cases of “AI psychosis.” Increasing personalization creates similar risks. To the extent that models reinforce users’ individual beliefs, they will be an epistemically diverging technology.
Nevertheless, I doubt that these forces will be strong enough to undermine LLMs’ disposition to generate accurate, evidence-based information.
First,many people use LLMs for simple information requests where these problems don’t arise. For example, arecent study finds that people frequently ask Grok to fact-check posts on X, including posts from politicians and pundits on their own side (“@Grok, is this true?”). Anotherstudy found that researching political claims with a chatbot raised belief in true information and lowered belief in misinformation.
Second, we shouldn’t overestimate the extent to which people prefer reinforcement of their own errors over accurate information. Motivated reasoning is a powerful force, but so is the desire to discover what’s true.
Third, AI companies can more easily be held accountable—reputationally and, in some contexts, legally—for what their LLMs say than social media platforms can for the viewpoints expressed on them. I suspect this is one major reason why leading AI companies seem to betaking measures to reduce the sycophancy of their models.
_**Manipulation**_
Another concern is that the outputs of LLMs might be manipulated by powerful elites. The incentives to engage in such manipulation exist, and there are no obvious technical barriers to it.
Some argue that we are already seeing this with the “woke” LLMs produced by OpenAI, Anthropic, and Google. However, although the leading LLMs doappear to exhibit a center-left political bias, it is relatively subtle and doesn’t seem to undermine their tendency to spread broadly reliable information, including when that information goes against dominant progressive narratives. In fact, Claude, ChatGPT, and Gemini can be much less woke than pretty much everyone in my social and professional network.
A clearer example concerns Elon Musk’s efforts to create an “anti-woke” AI, Grok. This has produced genuinely worrying outcomes, not least in the notorious “MechaHitler” debacle, in which an update intended to make Grok less politically correct led it to spew extremist, antisemitic content on X until xAI rolled back the changes andapologized.
Many treat episodes like this as a harbinger of things to come. I’m skeptical. Setting aside that short-lived episode, Grok’s outputs seem to broadly align with the kind of accurate, evidence-based information one gets from the other major LLMs. For example, arecent study found that Grok’s fact-checking evaluations roughly correspond to those of professional fact-checkers when it is augmented with search.
The incentives that govern social media are radically different from those underlying LLMs. On social media, someone like Musk can pump out an extraordinary amount of dumb, easily falsified misinformation to his hyper-partisan admirers without suffering any obvious reputational costs. But how many people would want to use an LLM that is similarly unreliable? It’s simply difficult to build systems that generate reliable, trustworthy information across a vast range of topics whilst simultaneously spreading conspiracy theories and quack science.
This doesn’t rule out an ecosystem in which different LLMs possess different ideological personalities. Republicans are alreadymuch more inclined to use and trust Grok than Democrats. But there is nothing in principle wrong with this. In fact, a plurality of leading LLMs with different ideological valences would be healthy in a democratic society, helping to guard against the risk that LLMs reduce epistemic diversity.
_**Disinformation**_
Generative AI undoubtedly remains a powerful tool for creating disinformation and propaganda that could reach audiences via channels distinct from ordinary chatbot use, fromfabricated recordings to “swarms” of coordinated propaganda bots.
Once again, however, I’m skeptical that this will counteract the technocratizing effects I have described. It isvery difficult to manipulate public opinion top down. People have sophisticatedcognitive defenses against manipulation and deception, and the reputational risks of spreading AI-based falsehoods discourage most influential figures and media outlets from doing so. And to the extent that LLMs provide unprecedentedly easy access to accurate information, they can greatly improve people’s defenses against misinformation.
If my speculations here are correct—and to be clear, speculations are all they are—then LLMs are a kind of anti-social media. Whereas social media has been democratizing, epistemically diverging, engagement-optimized, and performative, LLMs are technocratizing, epistemically converging, accuracy-optimized, and polite.
To many people, that probably makes LLMs sound like a surprising force for good. I suspect that part of the strong resistance I have received to this thesis is rooted in this assessment. If I’m right, LLMs are a force for good, but everyone knows that LLMs are not a force for good, so I must be wrong.
That objection represents an unsophisticated way of thinking. There is much to worry about when it comes to modern AI, including its effects on thelabor market, possible benefits to authoritarian regimes, and thegradual disempowerment of humanity. The major AI companies are profit-seeking firms led by figures with their own self-serving agendas, who should be held to account for producing harmful products. Nevertheless, holding them to account requires developing accurate models of the world, and too much of the AI discourse is driven by an unreflective,omnicausal anti-AI sentiment with very little concern for veracity or proportion.
At the same time, even the technocratizing tendency of LLMs is not wholly benign. There is asignificant risk that LLMs might deliver accurate information to individuals whilst reducing epistemic diversity in broader society. Lippmann’s vision of “intelligence bureaus” dispensing expert knowledge to the masses is being realized in a form he could never have imagined, but the classicproblems with that vision—the flaws of expert opinion, and the benefits of democratic diversity and debate—remain. However, we can only face up to these problems if we recognize LLMs for what they are: not a continuation of social media, but a powerful corrective to it.
**Dan Williams is an Assistant Professor in Philosophy at the University of Sussex. He works on issues at the intersection of science, politics, and technology. He blogs atwww.conspicuouscognition.com and is a contributing writer to Persuasion.**
_A version of this essay originally appeared in Conspicuous Cognition._
Follow _Persuasion_ on X, Instagram, LinkedIn, and YouTube to keep up with our latest articles, podcasts, and events, as well as updates from excellent writers across our network.
And, to receive pieces like this in your inbox and support our work, subscribe below: