The Trust Crisis That Never Resolved

Every year since 2017, someone has published a report declaring that media trust is at an all-time low. The reports are technically accurate and practically useless, because “media trust” as a single aggregate number is a category error. People don’t trust or distrust “the media.” They trust or distrust specific sources, for specific reasons, in specific contexts — and the aggregate score is the average of a thousand different relationships that have essentially nothing to do with each other.

The Gallup numbers from September 2029 show 23% of Americans expressing “a great deal” or “quite a lot” of confidence in newspapers, and 18% for television news. Both figures are lower than 2024. If you stop at those numbers, the story is simple: trust collapsed further and didn’t recover. But if you look at what those numbers are actually measuring, the picture gets complicated in ways that are genuinely important.

What the aggregate numbers can’t capture is the emergence of what researchers at Reuters Institute started calling “micro-trust” around 2027 — high-confidence relationships between specific audiences and specific publications or journalists that coexist with general distrust of “the media” as an abstraction. The same person who tells a pollster they don’t trust journalism will pay forty dollars a month for a newsletter from a specific reporter covering their city’s school board. These two facts are not contradictory. They’re describing different things.

The crisis that followed the 2024-2025 AI content wave was real and it was serious, but it wasn’t primarily about AI-generated misinformation fooling readers. That was the story that dominated tech coverage, and it produced an entire industry of AI-detection tools that mostly didn’t work and a lot of editorial handwringing about disclosure standards. The actual crisis was subtler: AI-generated content flooded search results and social feeds to the point where readers lost confidence in the process of finding reliable information, even when the information itself was accurate.

This is a different kind of trust problem than the industry had dealt with before. Previous trust crises were about outlets making specific errors, or about perceived bias, or about scandals involving individual journalists. Those crises had names and faces and could, in theory, be addressed by corrections and apologies and personnel changes. The 2025-2026 crisis was structural — a crisis of epistemics rather than a crisis of ethics. Readers weren’t (mostly) asking “did this specific article lie to me?” They were asking “how do I know anything I read is worth reading?” That question doesn’t have a simple answer.

The industry’s response to the structural crisis was, characteristically, to reach for the tools that had worked on previous crises. Fact-checking organizations proliferated. Transparency initiatives multiplied. The News Provenance Project, backed by a coalition of publishers, rolled out blockchain-based content authentication that would let readers verify that an article was published by a credentialed outlet. All of these were reasonable responses to the wrong problem.

The right problem was that credentialing no longer meant what it used to mean. A credentialed outlet in 2025 might publish a mix of original reporting, AI-summarized wire copy, AI-generated explainers, and genuine investigative journalism — with varying levels of disclosure about which was which, and with no consistent standard for what disclosure meant. Readers had no framework for understanding this mixed ecosystem, because the mixed ecosystem was new and the outlets’ own explanations of it were often self-serving.

What eventually helped — and “helped” should be understood modestly here, not as “solved” — was granularity. Trust recovered at the specific level before it recovered at the institutional level. Readers who had been burned by AI-generated content from aggregator sites developed more specific habits: they followed individual journalists, not publications. They subscribed to newsletters by people they could evaluate on track record. They shared not articles but specific claims, with context about where the claim came from and why the sender trusted the source.

This is micro-trust in practice. It’s a slower, more effortful relationship with information than the one that existed when you could read the New York Times or watch the evening news and assume a shared editorial standard governed everything you encountered. It places more cognitive burden on readers. It scales poorly — the people who develop careful source-evaluation habits are mostly already educated and curious; the habits don’t spread easily to people without the time or inclination to maintain them.

The political dimension of the trust crisis deepened rather than resolved. In 2029, trust in journalism has a stronger partisan valence than at any previous measured point. Republicans trust mainstream outlets at rates that round to zero. Democrats trust them more than the aggregate but less than they did in 2020. The partisan trust differential, which was already wide in 2024, widened further after several high-profile AI-related editorial failures — cases where AI tools used in production produced errors that went uncorrected, or where the process of using AI was inadequately disclosed in ways that, when revealed, became fodder for exactly the “fake news” argument that the affected outlets had been fighting for years.

The failures that hurt trust the most weren’t the spectacular ones. They were the mundane ones. A regional outlet runs an AI-drafted summary of a city council meeting that misattributes a vote. A national outlet uses AI to generate background context for a breaking story and the context contains an outdated statistic. Neither error rises to the level of a scandal, but both erode, incrementally, the sense that reading this outlet is a reliable way to know what happened.

The outlets that maintained trust most successfully were the ones that made a public commitment to keeping AI out of their original reporting — not out of technophobia but out of a clear-eyed reading of where their value lay. The Wall Street Journal made this commitment explicit in 2026 and publicized it aggressively. So did the Atlantic. Both saw measurable trust gains from the announcement, independent of whether readers could verify the commitment was being honored. The announcement itself was signal — it told readers that the publication understood the trust problem and had a position on it.

That positions matter more than verification is both interesting and slightly troubling. It suggests that the resolution of the trust crisis (to the extent one is occurring) is being driven less by actual changes in journalistic practice than by the social dynamics of signaling. Publications that signal trustworthiness gain trust; the signal is partly but not entirely backed by practice. This is not new — it describes how institutional credibility has always worked. But it’s newly visible in the AI era because the gap between signaling and practice is larger and more consequential than it used to be.

There’s a version of optimism available here, if you look at it from a certain angle. The audience that remains is, by most measures, more engaged than the broad shallow audience of the peak-traffic era. Subscribers who pay for journalism read more, share more substantively, and are more likely to act on what they read. The micro-trust relationship, effortful as it is, produces better-informed readers than passive consumption of algorithmically curated feeds ever did.

But that optimistic reading depends on ignoring the vast majority of people — the ones who aren’t reading quality journalism in any form, who are getting their information from social media and from AI chatbots and from each other, in an ecosystem that has no quality controls and no accountability structure. The trust crisis didn’t resolve. It sorted. Some readers ended up with more reliable information relationships than they had before. Many more ended up with worse ones. Whether that’s a crisis or a natural distribution is probably a philosophical question as much as an empirical one.

The honest answer is that we don’t know how to measure what we actually want to measure, so we keep measuring trust in “the media” and reporting the number as if it tells us something useful. It doesn’t. The number is going down, and that’s bad, and beneath that bad number is a more complicated situation that is also, depending on where you stand, somewhere between bad and complicated and occasionally, narrowly, better.