A deepfake offers a convenient target. It has a face, a voice, and often a jaw that moves with the delay of a digital ventriloquist: the falsehood still appears as an object separate from the world, and therefore as something a technician can detect, a newspaper can debunk, and a platform can label. Someone manufactures a lie, someone else recognises it, and society resumes operations.
The next information environment promises a different mechanism. Generative artificial intelligence steps between the event and the citizen as an interpreter. It summarises a speech, organises the news, explains a war, selects the context, removes one detail and adds another. Every step involves a political choice, because context determines how blame and causes are distributed.
Old propaganda searched for the message capable of reaching millions of people. New propaganda can search for millions of messages, one for each person. The anxious citizen receives an explanation that promises order. The furious citizen receives a narrative that supplies an enemy. The educated citizen receives sources, caveats, and subordinate clauses, the packaging prejudice prefers when it holds a university degree. Persuasion adopts the tone of personal care.
This capability already belongs to the present. A study published in Nature Human Behaviour compared debates between people with conversations involving GPT-4. Given a small amount of sociodemographic information, the model produced an 81.2 per cent increase in the likelihood of achieving greater agreement compared with a discussion between two people. The system had access only to age, gender, education, occupation, and political orientation. Platforms often possess a far richer biography, written through clicks, pauses, purchases, and searches.
The issue concerns the form of mediation. A newspaper publishes the same article for everyone. A television news programme delivers the same edit to millions of viewers. Every reader interprets that material through their own history, yet the text remains available as a shared object. Two opponents can point to the same sentence and spend three hours distorting it. Even that gesture contains a small democratic infrastructure: the disputed object remains visible to both sides.
A generative assistant changes this geometry. Its answer emerges from the relationship between the model, the question, the user’s history, their profile, and the service’s invisible rules. Another person receives a different text, priority, or fact. Debate loses the document and retains only the conclusions. Everyone arrives in the public square carrying a private receipt issued by a machine that alters its version with every request.
The Reuters Institute’s Digital News Report 2026 found that 10 per cent of the global public uses generative chatbots for news every week, up from 7 per cent the previous year. Only 1 per cent considers these tools its main news source. The shift matters because the chatbot gathers, synthesises, and presents information, while the original source recedes into the background. The citizen receives an answer instead of the process that produced it.
Summarisation enjoys an undeserved reputation for neutrality. Every summary decides what survives. Every hierarchy decides what carries weight. Every explanation establishes which doubts appear reasonable. A generative system performs these operations at a speed that makes selection almost invisible. The user encounters a smooth surface, like a hotel reception desk where someone has already hidden the cables, keys, and debts behind a cream-coloured wall.
Full Fact’s 2026 report describes the information environment as a network composed of media organisations, platforms, search engines, recommendation systems, AI tools, political actors, and closed groups. The report links design, economic incentives, automation, and distrust to the growth of public uncertainty. The harm includes confusion, withdrawal from participation, social tensions, and the loss of a shared factual understanding. A democracy can withstand incompatible opinions. It begins to collapse when it loses the floor on which those opinions confront one another.
The European Commission’s Joint Research Centre uses a sharper expression: “fractured realities.” Recommendation systems mix true, misleading, and false material, while economic models reward whatever retains attention. Users develop the impression that the news finds them, and mistake one portion of the world for the entire world. Every group preserves its own evidence and its own moral weather system.
Individual generation brings this fracture to a more intimate level. The feed displayed different content within a recognisable space. The chatbot constructs the space itself. The platform used to arrange the shop window; the assistant now produces the shop, the salesperson, the catalogue, and the ideal customer. Personalisation thus reaches the very description of reality.
The Carnegie Endowment for International Peace has called for the debate to be broadened, because AI affects democracy through elections, public services, deliberation, participation, and social cohesion. A deepfake can alter a vote. A system that mediates politics every day can alter the very concept of voting, transforming the citizen from a member of a public into the user of a cognitive service. The choice comes only after a sequence of private explanations, calibrated by companies that measure retention and return.
Convenience plays a decisive role here. Public reality requires effort. A newspaper contains irrelevant pages. A debate exposes us to irritating voices. An official document possesses the prose of a piece of ministerial furniture. An assistant removes friction, translates, shortens, and responds in the tone selected by the user. The service improves access while reducing encounters with anything that exceeds the question. Personalisation constructs a room in which every window displays the requested view.
That room can offer genuine benefits. A model can explain a law, translate a political programme, make a public budget accessible, and support a consultation. The Council of Europe recognises these possibilities and identifies AI as a tool for participation, inclusion, and public services. The same technology can expand citizenship or replace the citizen with a profile. The difference lies in the rules, the incentives, and public oversight.
Serious policy must shift its attention from individual pieces of content to the environment itself. Labels on synthetic material matter, as do verifiable provenance and archives. The public sphere also requires transparency over rankings, access to sources, the ability to reconstruct how an answer was produced, independent audits, control over personalisation criteria, and digital spaces founded on the public interest. A recommendation adopted by the Council of Europe in 2026 calls for greater transparency, democratic accountability, and independent oversight of platform algorithms and interfaces.
The central issue is the right to encounter the same reality. This differs from the right to truth, an enormous phrase that invites philosophers, priests, and chief executives to take their seats at the table. The right to a common reality means access to traceable facts, stable texts, citable sources, and shared procedures for correcting errors. It also means exposure to events that a personal profile would have excluded because of limited interest, low affinity, or insufficient profitability.
Democracy lives on this quota of imposed experience. A crisis, a war, a ruling, and a law enter the common space and demand attention even from those who would prefer something else. Public reality contains a useful form of rudeness: it interrupts, contradicts, and takes up time. The tailored system promises instead a world that is polite, relevant, and compliant. That world closely resembles the user, and for precisely that reason teaches them less and less.
A public sphere can exist within an individually generated environment only when personalisation encounters political limits. The public emerges from what remains the same while interpretations diverge. Technology can assist that confrontation, translate it, and expand it. Technology can also dissolve it into billions of flawless, considerate, and mutually irreconcilable conversations.
The end of public reality will arrive through a pleasant interface. Every answer will appear certain. Every user will feel understood. Every version of the world will bear their name. Democracy will then discover that consensus was a luxury, while disagreement required at least one simple and increasingly rare thing: the same room.
Niccolò Carradori