There is a particular kind of person the Reuters Institute has learned to count, and the count is growing. She is not angry at the media. She has not cancelled a subscription in protest or fired off a furious letter about bias. She is not part of any movement, and she would be surprised to learn that anyone was studying her at all. She checks her phone in the morning and sees, without asking for it, a tidy paragraph telling her what happened overnight — a summarised headline, an AI-generated notification, a single sentence at the top of a search result. She reads it, or half-reads it, and moves on. She feels no urge to click through, and increasingly there is nothing to click through to. She is not uninformed, exactly. She is not misinformed. She has simply, quietly, and without any dramatic decision, stopped participating in the news as an activity. And there are more of her every year.
The 2026 Digital News Report, published by the Reuters Institute for the Study of Journalism in June and drawn from surveys across 48 markets, put a number on this drift that deserves more attention than it received. A quarter of respondents — 25 per cent — now count as what the researchers call casual or passive news users: people who consume news at most once a week and say they have little or no interest in it. In 2021 that figure was 16 per cent. A rise of nearly ten percentage points in five years is not a blip. It is the emergence of a new modal relationship to the news, one defined less by consumption than by ambient exposure, and it is happening at the same moment that the tools mediating our access to information have become extraordinarily good at doing the reading for us.
Most of the coverage of the 2026 report fixed, understandably, on trust. Global trust in news has fallen to 37 per cent, the lowest level the institute has recorded since it began tracking the measure in 2015. Trust declined in 29 of the 48 markets surveyed and fell by five percentage points or more in 19 of them. In the United States, only a quarter of people say they trust the news most of the time. These are alarming figures, and they have generated the expected round of soul-searching about media credibility. But trust is a story about the relationship between the public and the press. The disengagement figures point to something structurally different and, I want to argue, more consequential: not a breakdown in the relationship between citizens and journalism, but a fraying of the relationship between citizens and a shared picture of reality itself.
The Disappearance That Nobody Announced
It is worth being precise about what the casual-user category measures, because it is easy to fold it into more familiar anxieties. This is not the same as news avoidance, the deliberate turning-away that the same report tracks separately and which has held roughly steady, hovering near 42 per cent globally with much higher figures in a handful of markets. Avoidance is an act. It requires enough engagement to decide that the engagement is not worth it. The casual user, by contrast, is not avoiding anything. She has not made a judgement about the news at all. She has simply allowed it to recede from the list of things she does deliberately, in the way that a person might stop cooking without ever deciding to stop cooking, because the food arrives anyway.
The report situates this within a broader collapse of interest. The proportion of people who describe themselves as extremely or very interested in the news has fallen sharply across the surveyed markets over the same period, and the population the institute once labelled “news lovers” — those for whom keeping up with events is a genuine appetite — has thinned considerably. The picture, then, is not of a public storming out of the room. It is of a public gradually drifting toward the exit, in ones and twos, without any of them noticing that the room is emptying.
The Reuters researchers attribute the shift to several overlapping pressures. There is the experience of news overload during a decade of successive crises — pandemic, war, economic shock, climate emergency — that has left many people feeling anxious, overwhelmed, or powerless in the face of events they can neither influence nor escape. There is the platform-driven fragmentation of news consumption, which has scattered the audience across social feeds, video networks, and messaging apps where news arrives incidentally, tangled up with everything else. And there is the phenomenon the report describes as news fatigue: the sense, reported by growing numbers of people, that keeping up simply costs more emotional energy than it returns. These are real and sympathetic reasons. Nobody drifting away is behaving irrationally. That is precisely the problem. The exit is not a failure of character. It is a rational response to an information environment that has made staying informed feel both exhausting and optional.
What a Nation Used to Read at Breakfast
To understand why this matters beyond the fortunes of newspapers, it helps to recall what the daily news was ever for, in the civic rather than the commercial sense. The most durable account belongs to the political scientist Benedict Anderson, whose 1983 book Imagined Communities argued that the modern nation is held together not by face-to-face acquaintance — you will never meet more than a vanishing fraction of your fellow citizens — but by a shared act of imagination, sustained by common texts read at common times. Anderson gave the newspaper a central role. The morning paper, consumed by millions of strangers simultaneously, produced what he called a sense of simultaneity: the quiet, structuring knowledge that unseen others were, at that very moment, absorbing the same events, forming the nation into a community that could be imagined precisely because it could be jointly informed. Anderson thought it hard to conceive a more vivid figure for the secular, historically clocked imagined community than the mass ceremony of the newspaper read in silence, in parallel, across a country.
Whatever one makes of the details, the underlying insight is not nostalgic sentimentality about print. It is a claim about infrastructure. Democratic self-government presupposes not merely that citizens hold opinions but that they hold them about a broadly common reality — that when a policy is debated, the debaters are arguing about roughly the same set of facts. The philosopher Jürgen Habermas gave this its most influential theoretical form in his 1962 study of the public sphere, describing a domain mediating between private citizens and the state in which people could gather, at least in principle, to discuss matters of common concern and, through that discussion, form a public opinion capable of holding power to account. The public sphere was never a physical place. It was a shared communicative space, and its precondition was a population that showed up to it. In his 2023 reflections on a new structural transformation of that sphere, published three years before his death in March 2026 at the age of 96, Habermas warned that without appropriate regulation, digital media risked hollowing out the very institutions through which democracies shape their own social and economic life.
Both thinkers, in other words, were describing a commons — an epistemic commons, a shared pool of common knowledge that no individual owns but on which collective self-government depends. And commons are vulnerable to a specific failure mode. They do not usually collapse through malice. They collapse through the quiet, individually reasonable decisions of people who each conclude that their own participation is not really necessary, that the thing will hold together whether or not they personally show up. What the casual-user figure describes is exactly this dynamic applied to the common information environment: a slow, uncoordinated, entirely comprehensible withdrawal from the shared act of paying attention.
The Difference Between Not Knowing and Not Caring
There is a temptation to treat all of this as a subset of misinformation panic — to fold the disengaged into the vast literature on falsehood and manipulation. But the disengaged are a distinct problem, and in some respects a harder one. The person consuming conspiracy content is at least still engaged with a picture of the world, however distorted, and can in principle be reached, argued with, corrected. The disengaged citizen has no false picture to correct. She has an increasingly thin one, assembled from ambient fragments, held without much conviction and without the connective tissue that would let one fact bear on another.
The most careful recent study of who these people are and why they withdraw is Avoiding the News, the 2023 book by Benjamin Toff, Ruth Palmer, and Rasmus Kleis Nielsen, published by Columbia University Press and based on extensive survey data alongside more than a hundred interviews with consistent news avoiders across the United States, the United Kingdom, and Spain. Their findings resist easy moralising. The people who turn away from the news are disproportionately younger, more often women, and more likely to come from lower socioeconomic backgrounds — which is to say they are, on the whole, people who already feel that the news is not made for them, does not speak to their lives, and does not repay the effort it demands. The authors organise the causes under three headings: identities, ideologies, and infrastructures. People avoid the news partly because of who they understand themselves to be, partly because of what they believe about the media, and partly because of the practical architecture of their information environment — how easy or hard the surrounding technology makes it to encounter news in the first place.
That third category, infrastructure, is where the story turns, because infrastructure is exactly what is changing fastest. The identities and ideologies that drive news avoidance are old and slow. The infrastructure is being rebuilt, in real time, by a wave of AI systems whose entire design purpose is to reduce the effort involved in encountering information — and, in the process, to reduce the effort involved in not encountering it. The casual user of 2021 still had to do something to stay minimally informed: open an app, scan a homepage, choose a story. The casual user of 2026 increasingly does not. The information comes to her, pre-chewed, and the act of full engagement has become genuinely optional in a way it was not before.
When the Summary Becomes the Whole of It
Consider the actual texture of how an ordinarily busy person now brushes against the news. She types a question into Google and, before any list of sources appears, an AI Overview supplies an answer in a paragraph. A Pew Research Center study published in July 2025 tracked the browsing of 900 American adults across 68,879 Google searches. When an AI summary appeared, users clicked a traditional search result in 8 per cent of visits, against 15 per cent when there was none, and they clicked a link inside the summary itself in just 1 per cent of visits. They ended their browsing session altogether on 26 per cent of pages carrying a summary, compared with 16 per cent of pages without one. The summary, in other words, is not a doorway to the underlying reporting. For most people, most of the time, it is the terminus. The question is asked, the paragraph is read, the session ends. The researchers Paulo Couraceiro and Pedro Caldeira Pais have a name for this: news sufficiency, the sense that a condensed summary is already enough, which leaves the reader with little reason to open the article it was built from.
A caveat is owed here, and it sharpens the point rather than blunting it. Pew measured search behaviour in general, not news in particular, and the 2026 Digital News Report notes that Google mostly does not enable AI answers for hard-news searches. Yet the same report records that Google organic search traffic to more than 2,500 sites fell by a third globally, and that publishers expect search traffic to fall by a further 43 per cent over the next three years. Whatever share of that decline the summaries account for, the direction of travel is the one the Pew data predicts: fewer journeys from the question to the reporting.
She glances at her phone and sees a notification that is not the headline a journalist wrote but a machine's compression of it. Apple learned the hazards of this the hard way. In January 2025 the company was forced to suspend the news-summary feature of Apple Intelligence after its system generated notifications that were not merely clumsy but false: it summarised a BBC report to state that Luigi Mangione, the man charged in the killing of a healthcare executive, had shot himself, which he had not; it announced that the darts player Luke Littler had won a championship hours before the final began; it reported that the tennis player Rafael Nadal had come out as gay. The BBC lodged a formal complaint. Apple paused the feature and promised to make clearer, in future, which notifications were AI-generated. The episode is usually filed under accuracy, and the accuracy problems are real. But the deeper significance is structural. Here was a system positioned between the reader and the reporting, rewriting the reporting on the fly, and doing so at a layer of the interface — the lock-screen notification — that most users experience as simply true, because it looks exactly like every other notification the device has ever shown them. Apple's eventual answer made the point for it. When summaries returned for news apps in iOS 26, released in September 2025, they came back as a setting users must switch on, carrying the footer “Summarized by Apple Intelligence”. The fix was a label. The machine-rewritten layer between reader and reporting came back with it.
And she talks, increasingly, to a chatbot. The 2026 Digital News Report found that weekly use of AI chatbots such as ChatGPT, Perplexity, and Google Gemini for news rose from 7 per cent to 10 per cent globally year on year, and reached 16 per cent — double the global average — among people under 35. This is still a minority behaviour, but it is the fastest-growing route into news the institute tracks, and it represents something categorically new. A chatbot does not present a front page. It does not convey, in the way a newspaper's layout or a bulletin's running order does, any sense of proportion — what is important, what is merely interesting, what a serious person is expected to know. It answers the question you happen to ask, in the register you ask it, and volunteers nothing. To be fair, 42 per cent of people who use chatbots for news tell the Reuters Institute that they always or often click through to the original sources, though the report itself cautions that what people say they do and what they actually do can differ. Even taken at face value, the click leads only to the story she asked about. The old news experience, whatever its faults, imposed an editor's judgement about what mattered onto a reader who had not asked for it, and that imposition was doing civic work: it exposed people to things they would never have chosen. The chatbot removes the imposition. It gives you what is relevant to you, which is a genuine service, and in doing so it quietly dissolves the very thing — shared, unrequested, common exposure — that made the news a commons rather than a personal utility.
The Economics of the Frictionless Exit
To see why AI curation is likely to deepen disengagement rather than reverse it, it helps to think about cost. Every act of staying informed has always carried a cost — of time, of attention, of the effort required to seek out, select, and make sense of information. For most of the twentieth century that cost was borne willingly by a large public, partly out of habit, partly because the alternatives were scarce, and partly because the infrastructure of daily life — the paper on the doorstep, the bulletin at six — made the informed path the path of least resistance. What has changed is not that people have become lazier. It is that the cost structure has inverted. Staying deeply informed now requires swimming against a current specifically engineered to carry you somewhere easier.
The psychological literature has a name for the mechanism at work: cognitive offloading, the human tendency to delegate mental effort to external tools whenever a reliable tool is available. The foundational demonstration came in 2011, when Betsy Sparrow, Jenny Liu, and Daniel Wegner published a study in Science titled “Google Effects on Memory,” showing that when people expect to have future access to information, they remember the information itself less well and remember instead where to find it. The internet, they argued, had become a form of transactive memory — an external store we lean on in place of our own recall. The finding was once a curiosity about trivia and search engines. It now describes the central design principle of an entire generation of consumer technology. We are being offered, across every surface of digital life, the standing promise that we need not hold the information ourselves, because the system will hold it for us and hand it back on demand.
A paper posted to arXiv in May 2026 by the computer scientists Ilias Chalkidis and Anders Søgaard, provocatively titled “Brainrot: Deskilling and Addiction are Overlooked AI Risks” and accepted to the ACM FAccT conference, argues that this kind of harm has been systematically neglected by an AI-safety field preoccupied with more tangible threats such as discrimination, hate speech, and dangerous content. The authors focus on the deskilling and atrophy of critical faculties that can follow from over-reliance on generative systems, and on the dependence such systems can cultivate. Their broader point, and the one most relevant here, is that a class of slow, diffuse, hard-to-measure harms — the gradual dulling of capacities we do not exercise because a machine exercises them for us — falls precisely into the blind spot of a discipline that prefers risks it can quantify. Digital fatigue and disengagement, they suggest, are exactly this kind of under-recognised risk of an AI-mediated information environment.
Follow that logic to its conclusion. If cognitive offloading is the mechanism, and if AI systems increasingly curate, summarise, and pre-filter the news before it reaches us, then the cost of full disengagement from active news consumption falls toward zero. The casual user of a decade ago paid a real penalty for her disengagement: she was genuinely out of the loop, visibly and sometimes awkwardly uninformed, and the discomfort of that penalty was itself a mild incentive to re-engage. The casual user of 2026 pays almost nothing. She can maintain the surface of being informed — the summarised paragraph, the notification, the chatbot answer when a specific question arises — without any of the underlying labour, and without the social embarrassment that used to attend obvious ignorance. AI curation does not drag her back into engagement. It makes her disengagement comfortable, sustainable, and nearly invisible, to her and to everyone around her. It removes the friction that once served, however crudely, as a corrective.
This is the pivot on which the whole question turns, and it is worth stating plainly, because it cuts against the industry's own account of itself. The promise of AI curation is relevance and convenience: the right information, for you, with less effort. Taken on its own terms, that promise is real and often valuable. But relevance and convenience are precisely the qualities that lower the cost of leaving. A tool that gives you exactly what you need to know, when you ask, with no surplus and no friction, is also a tool that makes it effortless never to encounter anything you did not ask for — and it is the unrequested encounter, the thing you did not know you needed to know, that the shared information environment was for. The more perfectly the system serves the individual, the more efficiently it dismantles the commons.
The Quiet Secession From a Common Reality
What actually happens to a society in which a large and growing minority has decoupled, in this soft and frictionless way, from the common information environment? It is important to resist the melodramatic answers, because the melodramatic answers are wrong. There will be no sudden darkness, no moment at which the lights of shared understanding visibly go out. The disengaged are not stupid, they are not helpless, and in their daily lives they will manage perfectly well — the summarised world is, for most practical purposes, an adequate world. The consequences are not personal. They are collective and structural, which is exactly why they are so easy to miss.
The first consequence is the erosion of common knowledge in the technical sense that democratic theory cares about — not merely facts that many people happen to know, but facts that people know that others know, and know that others know they know. This recursive, shared quality is what allows a public to act as a public: to recognise the same event as significant, to hold power to account for the same failures, to argue productively because the argument rests on a common floor of accepted reality. The concerns raised decades ago by Cass Sunstein in Republic.com and by Eli Pariser in his 2011 book The Filter Bubble were early versions of this worry. Both argued that personalised information environments would fragment the shared world into millions of individually tailored ones, eroding the common ground on which deliberation depends. Their thesis was contested; subsequent research complicated the neat picture of sealed echo chambers, and many people's actual media diets turned out to be more varied than the filter-bubble alarm implied. But the disengagement data reframes the old debate in a way that sidesteps its most disputed empirical claims. The question is no longer only whether personalisation traps people in narrow bubbles of congenial content. It is whether a growing share of people are floating free of any substantial shared content at all — not enclosed in a small world, but only loosely attached to any common world.
The second consequence is subtler and concerns the composition of the remaining public. As the casually disengaged drift away, the population that stays actively, intensely engaged with the news does not become a representative sample of everyone else. The Reuters and Avoiding the News findings suggest the opposite: those who remain most engaged skew toward the already-committed, the politically motivated, the demographically advantaged. A public sphere that loses its casual majority and retains chiefly its zealous minority is a public sphere with a distorted centre of gravity — louder, more partisan, more polarised, less representative of the quiet middle whose disengagement removes precisely the ballast that once moderated public argument. The people leaving are, disproportionately, the very people whose presence made the conversation broad rather than shrill. Their exit does not leave the room half full. It changes what kind of room it is.
The third consequence is a hollowing of accountability. Journalism's civic function has never depended only on producing accurate reporting; it has depended on that reporting reaching a public large enough, and attentive enough, to convert information into consequence. Investigations matter because they are read, discussed, and acted upon. A revelation that lands in an environment of general inattention — summarised into a sentence, glanced at, and forgotten before the session ends — has been technically delivered and civically wasted. The machinery of scrutiny can keep running at full capacity while the audience that gave it its force quietly disperses, and the powerful, who are acute readers of public attention, will notice the dispersal long before the rest of us do.
Something Other Than Nostalgia
None of this is an argument that the past was better in the round, and it is worth guarding against the reflex that says it was. The twentieth-century news commons was narrower than its admirers remember: it was shaped by a small number of gatekeepers, it excluded many voices, and its apparent consensus often rested on whose reality got to count as the shared one. The AI systems now mediating our information carry real benefits that the honest analyst cannot wave away — genuine gains in access for people the old media never served, real reductions in the drudgery of finding things out, real help for the overwhelmed. A person who reads a competent AI summary of a policy debate she would otherwise have skipped entirely is, in that instance, better informed than she would have been, not worse. The technology is not the villain of this story, and framing it as one merely repeats the doom-mongering that has made so much writing about AI useless.
The structural worry is more specific and more difficult than villainy. It is that a set of tools optimised, quite reasonably, to serve individuals — to give each of us relevance, convenience, and relief from overload — may in aggregate corrode a public good that no individual has any private incentive to protect. Each person's frictionless exit from active news consumption is individually harmless and collectively expensive, and there is no market signal that will register the collective cost, because the cost falls on the commons and the commons has no accountant. This is the classic shape of a collective-action problem, and collective-action problems are not solved by exhortation. Telling people to care more about the news, or to read past the summary, is like telling them to drive less to fight congestion: sensible, unobjectionable, and almost entirely without effect, because it asks individuals to bear a private cost for a benefit they cannot capture alone.
If there is a way through, it is unlikely to run through appeals to individual virtue and much more likely to run through design and institutional choice — through the way the systems themselves are built. The AI intermediaries that increasingly stand between citizens and reality are not forces of nature; they are products, shaped by decisions about what to optimise for. A summary layer could be built to widen exposure rather than narrow it, to surface the unrequested and the important alongside the merely relevant, to treat the click-through to original reporting as a feature to protect rather than a leak to plug. Whether the companies that build these systems have any incentive to do so is a separate and less encouraging question, and it is the question that the accuracy-focused coverage of Apple's stumble and Google's traffic-draining Overviews mostly failed to ask. The Pew finding that people clicked a link inside an AI summary in just 1 per cent of visits is not merely a business problem for publishers. It is a description of an information architecture that has been optimised, deliberately, to be a terminus rather than a threshold — to end the encounter rather than begin it.
The woman with whom this essay began will be fine. That is the hardest part to hold in view, and the most important. She is not a cautionary figure; she is not in distress; she has made no error that anyone could point to and correct. She has simply accepted, as almost all of us increasingly do, an offer that is genuinely attractive: to be spared the effort of the shared world and handed, instead, a private and sufficient version of it. The 2026 Digital News Report counted 25 per cent of us in her position and rising, and the machines that summarise our mornings are getting better at their work every quarter. The commons does not end with a crisis or a collapse. It ends the way commons always end — not with anyone deciding to abandon it, but with a great many people, each for perfectly good reasons, quietly ceasing to show up, until one day there is no longer a there for anyone to show up to, and no announcement is ever made, because there is no longer any shared place in which to make it.
