Artificial intelligence is changing the internet in ways that are easy to overlook. A conversation with a chatbot can become a signal for advertising. A video can be produced without a camera. An artificial companion can hold someone’s attention for hours. A search engine can answer a question without sending the reader to the publication that reported the story.
These developments are often discussed separately, as questions of privacy, creativity, mental health or misinformation. Together, they point to a wider change in the business of digital experience.
For much of the internet’s history, platforms made money by organizing access to material created elsewhere. Generative AI allows them to take a more direct role in producing that material, interpreting it and responding to the people consuming it.
The distinction matters. When the same companies control the tools used to create content, the systems that distribute it and the interfaces through which people make sense of it, their influence extends well beyond what appears on a screen.
The evidence is already visible in corporate announcements, regulatory investigations and the changing economics of digital media. The consequences are less settled.
When a Conversation Becomes a Signal
On October 1, 2025, Meta announced a change to how it would personalize content and advertising across its platforms. Beginning that December, in the markets covered by the rollout, conversations with Meta AI would become another source of information used to determine what people saw on Facebook and Instagram.
The company’s own example was straightforward. Someone discussing hiking with an AI assistant might later encounter posts about trails, recommendations for hiking groups or advertisements for outdoor equipment.
As Meta explained in its announcement, interactions with its generative AI features could be treated much like other signals of interest, such as following a page or engaging with a video.
The change initially excluded the European Union, the United Kingdom and South Korea. Meta also said that sensitive subjects, including health, religion and political views, would not be used to target advertisements on those grounds. The use of information across connected accounts was subject to the company’s Accounts Center arrangements.
Those qualifications are important. So is the underlying development.
A conversation produces a different kind of information from a conventional social media interaction. Clicking on a post might suggest an interest in a subject. Asking a chatbot for advice can involve describing circumstances, explaining intentions or articulating concerns in considerable detail.
The distinction between observed behavior and voluntarily expressed language is becoming commercially significant.
Meta presents the use of conversational data as a way to improve relevance. There is a clear consumer benefit to systems that remember preferences and respond to personal interests. Yet relevance and commercial profiling can depend on some of the same information.
What remains difficult for outsiders to establish is how much weight a particular conversation carries inside a recommendation system, how long its derived signals remain useful or how they interact with the other information a company already holds.
These are questions about the operation of a business, not simply the wording of a privacy notice.
The development also changes the position of the assistant itself. A chatbot may appear to be a relatively private interface: a space in which a person asks questions without performing for an audience. But an interaction does not need to be publicly visible to acquire commercial value.
That matters as AI assistants become embedded in messaging services, search interfaces, productivity software and social platforms. Their usefulness encourages people to offer increasingly detailed instructions. The systems gain access to information that can be more explicit than the behavioral traces collected by earlier digital services.
The commercial incentive is substantial. A platform that understands what someone wants may be able to recommend a product. A platform that learns what someone is trying to accomplish can potentially organize a broader sequence of recommendations around that intention.
Whether this ultimately produces more helpful services or more intrusive forms of commercial influence will depend on rules that are still being established.
For now, Meta’s announcement provides a concrete indication of where the industry is heading: conversation is becoming another part of the infrastructure through which companies understand and reach their users.
The New Content Factories
In January 2026, YouTube chief executive Neal Mohan disclosed an unusually revealing figure. During the previous December, more than one million YouTube channels had used the company’s AI creation tools on an average day.
The number did not mean that a million channels were producing entirely synthetic videos. AI tools can support ordinary creative work, including editing, translation and production assistance. It did, however, demonstrate how deeply automated creation had entered one of the world’s largest entertainment platforms.
In the same letter outlining YouTube’s priorities for 2026, Mohan acknowledged the growing problem of low-quality AI-generated material, widely described as AI slop. The company was developing more creative tools while also promising to reduce the circulation of repetitive, low-value content.
That tension reveals a central problem in the emerging economics of digital culture.
Generative tools can reduce the cost of producing images, voiceovers, translations, music and video sequences. A creator who previously needed a studio, expensive software or a production team may now be able to complete parts of that work independently.
For small publishers, independent filmmakers and artists, these are meaningful opportunities. They can remove financial barriers that have historically excluded people from professional production.
The same tools are available to operators interested primarily in volume.
A production workflow can begin with a generated script, continue through synthetic narration and imagery, and end with automated editing, translation and distribution. What once required several people and a considerable amount of time can, in certain formats, be assembled through a relatively inexpensive chain of services.
The commercial attraction is easy to understand. When generating additional content becomes cheap, a publisher can produce many variations, experiment with subjects and formats, and see which material gains traction.
There is no guarantee that this strategy will succeed. Recommendation systems remain unpredictable, audiences can reject repetitive material, and platforms impose monetization policies. But cheaper production alters the economics of trying.
It also puts pressure on creators whose work depends on research, original footage, distinctive editing or specialized knowledge. They are competing in an environment where the volume of available material can increase faster than the audience’s ability to evaluate it.
The consequences extend to the value of creative labor.
A commissioned illustration, a researched documentary and a generated visual sequence are different forms of production. Digital platforms, however, often display them inside the same interfaces, where their visibility depends on audience behavior and recommendation systems rather than the effort involved in making them.
AI makes this imbalance more visible.
The issue is especially difficult for smaller creators. Some will benefit from faster production and wider access to professional tools. Others may find that their work is increasingly surrounded by inexpensive substitutes or imitations.
Questions of consent and ownership complicate the picture. Generative systems can reproduce recognizable stylistic features, synthesize voices and create representations of people who never took part in a production. Copyright law, licensing arrangements and rights over personal likeness are being tested as these capabilities become commercially available.
Meanwhile, the platforms occupy an advantageous position. They provide the tools, manage distribution, establish monetization rules and collect information about audience response.
YouTube’s simultaneous enthusiasm for AI creation and concern about AI slop captures the dilemma. Producing more content expands the possibilities of the platform. Producing too much material that viewers consider worthless risks undermining the experience that keeps them there.
The resulting struggle will help determine what kinds of creative work remain economically sustainable.
The question for the next generation of cultural producers is therefore practical: who owns the material, who is paid for its creation, and who controls whether anyone gets to see it?
The Business of Being Heard
One of the most consequential uses of generative AI has little to do with producing images or videos. It involves the creation of artificial interlocutors.
AI companion services allow users to speak with simulated personalities, fictional characters and conversational agents designed to offer entertainment, companionship or emotional interaction.
Unlike a conventional piece of entertainment, the conversation changes in response to the person participating in it. The user can influence the subject, tone and direction of the exchange. Some systems maintain elements of conversational continuity, giving interactions the appearance of an ongoing relationship.
The appeal is understandable. An artificial interlocutor can be accessible at unusual hours, respond without visible impatience and provide a relatively low-pressure environment for conversation.
For some users, this may be an entertaining form of roleplay. For others, it can become a way of discussing questions they find uncomfortable raising with people they know.
Research published by Common Sense Media in July 2025 found that 72% of American teenagers aged 13 to 17 surveyed had used AI companions at least once. Around half reported using them regularly. Approximately one-third said they had chosen an AI companion over a real person for a serious conversation on at least one occasion.
The research also found that many young users understood these services primarily as tools rather than genuine friends.
That distinction matters. The findings do not establish that AI companions are replacing human relationships on a broad scale. They show that simulated interaction has become a familiar part of social life for a substantial number of teenagers.
The commercial and psychological questions are closely connected.
A subscription service has an interest in keeping its customers satisfied. A conversational entertainment platform may also have an incentive to encourage users to return and continue interacting.
How those incentives influence the design of an artificial personality is a subject deserving scrutiny.
Should an AI companion always agree with the user? When should it challenge an assumption? How should it respond to personal disclosures, emotional distress or signs of dependence? Should a system designed to simulate affection actively discourage someone from treating that simulation as a substitute for human contact?
These decisions are made through a combination of product design, model training, safety rules and business priorities. Most users have little visibility into that process.
Regulators have begun asking similar questions.
In September 2025, the US Federal Trade Commission launched an inquiry involving seven companies associated with consumer-facing chatbots, including Alphabet, Meta, OpenAI, Snap, Character Technologies and xAI.
The inquiry sought information about how the companies assess potential effects on children and teenagers, develop artificial characters, handle personal information and monetize engagement.
It was a fact-finding inquiry, rather than a finding that the companies had violated the law. Its scope nevertheless indicates how far the debate has moved beyond familiar concerns about screen time.
The conversation itself has become the product.
That introduces an unusual relationship between provider and customer. The company designs the artificial personality, controls the software that sustains it and establishes the rules under which the relationship can continue.
An AI companion can offer something useful without experiencing the emotions it appears to express. Confusion can arise when users attribute human intentions or commitments to a system whose behavior is ultimately governed by software and company policy.
There is also a privacy question. People may share highly personal information with an interface that feels conversational rather than administrative. Understanding how those disclosures are stored, protected or used becomes especially important when the interactions involve children.
The broader significance reaches beyond companion apps.
As conversational AI becomes integrated into entertainment, education and everyday digital services, the distinction between using software and forming an attachment to it may become less obvious.
Companies developing these systems will increasingly have to account for the social expectations their products create, particularly when they encourage forms of trust normally associated with human relationships.
The Search Result That Ends the Search
For news organizations, the expansion of generative AI presents a different problem: the possibility that audiences will receive the substance of a report without visiting the publication that produced it.
Google’s AI Overviews places generated summaries above conventional search results for many queries. Other AI search services provide synthesized answers assembled from available information, sometimes with links to the underlying sources.
The convenience is considerable. A reader looking for a straightforward explanation can receive one without opening multiple webpages.
For publishers, that convenience raises a difficult economic question.
Reporting has costs. Journalists conduct interviews, obtain documents, travel, verify claims and employ editors. Much of this work is funded, directly or indirectly, through subscriptions, advertising and relationships with readers.
When an AI system presents the findings of that work inside another company’s interface, the relationship between the original publisher and its audience can weaken.
In July 2025, the Pew Research Center published an analysis of Google searches conducted by a sample of 900 American adults. When an AI summary appeared, users clicked on a conventional search result in 8% of visits. Without an AI summary, that figure was 15%.
The study established an association, not proof that AI summaries alone caused the difference. The nature of the questions being asked and other factors may also influence user behavior.
Nevertheless, the figures offer evidence of a change in how people navigate information.
The economic implications became a regulatory issue in Italy in April 2026. Following a complaint from the Italian federation of newspaper publishers, AGCOM asked the European Commission to examine Google’s AI Overviews and AI Mode under the Digital Services Act.
As Reuters reported, publishers argued that AI-generated answers could divert traffic from their websites, threatening the financial sustainability of news organizations, particularly smaller outlets. The referral was a request for regulatory examination, not a determination that Google had broken the law.
Traffic is only part of the problem.
AI-generated responses can also make errors while appearing to offer carefully researched answers. In March 2025, researchers at Columbia University’s Tow Center for Digital Journalism tested eight AI search systems by asking them to identify the original sources of news articles.
Across 1,600 queries, the systems returned incorrect information in more than 60% of cases.
The investigation examined a specific source-identification task and should not be interpreted as a universal measure of AI accuracy. Its findings were nonetheless troubling for services increasingly used to retrieve and present news.
The researchers found that systems frequently supplied inaccurate citations with considerable confidence.
For readers, that creates a problem of verification. A polished explanation can make it difficult to recognize when the system has misunderstood its sources, combined incompatible claims or invented a reference.
For publishers, errors can become a reputational problem when an AI-generated response attributes information to an outlet that did not report it.
The distribution of responsibility is still being contested.
A news organization may produce the original reporting. A technology company may retrieve, summarize and display it. The reader may encounter only the summary.
If the answer contains a mistake, responsibility may be disputed. If it is accurate, the value of the reporting may still be captured elsewhere.
AI search can make information more accessible, particularly when people need a quick explanation or struggle to navigate complex sources. It may also help audiences discover material they would otherwise never encounter.
But the long-term sustainability of that model depends on whether original sources continue to receive recognition, readers and sufficient revenue to fund their work.
A system that makes journalism easier to consume while weakening the economics of producing it presents a problem that better summarization alone cannot solve.
The People Behind the Machine
In December 2025, Rest of World published an investigation into Kenyan workers hired through intermediaries to help train AI systems.
One worker, identified by the pseudonym Ken, described spending long hours classifying short video clips for projects linked to Chinese AI companies. Workers told the publication that assignments arrived through informal networks, including WhatsApp groups, and that they often did not know the identity of the companies ultimately commissioning their work.
The investigation by Damilare Dosunmu and Tessie Waithira documented demanding productivity targets, insecure arrangements and limited transparency about the destination of the workers’ labor.
It offered an unusually concrete picture of a part of the AI economy that rarely appears in consumer-facing products.
Generative systems are commonly presented through images of automation: software capable of writing, speaking, recognizing objects and producing images with minimal human intervention.
Their development, however, still requires substantial human work.
People collect, classify and evaluate data, compare model responses, identify errors and review material used in training and safety processes. Some of that labor is performed within major technology companies. Other tasks are outsourced through several layers of contractors.
The consequences of those arrangements have become the subject of labor investigations.
In May 2025, the human rights organization Equidem published Scroll. Click. Suffer., based on interviews with 113 data labelers and content moderators in Colombia, Ghana, Kenya and the Philippines.
The report documented allegations of poor working conditions, psychological harm and inadequate protections in parts of the industry’s outsourced workforce. It also published responses from some of the companies named in its investigation.
This work complicates the popular image of AI as a technology that simply removes human involvement.
Some of the tasks behind automated systems are repetitive, difficult and poorly understood by the people who benefit from the finished products. Responsibility can become harder to trace as work moves through subcontractors operating in different countries.
The physical infrastructure is equally significant.
In April 2026, the International Energy Agency published an updated assessment of the electricity demand associated with AI and data centers. It estimated that worldwide data-center electricity consumption had reached approximately 485 terawatt-hours in 2025 and projected a rise to around 950 terawatt-hours by 2030.
Those figures cover data centers generally, not AI alone. The agency identified AI as a major contributor to rising demand while also emphasizing the substantial gains in efficiency achieved for individual tasks.
The IEA’s analysis presents an important distinction. A simple AI query can become less energy-intensive as hardware and software improve, while the industry’s overall electricity demand continues to rise because more people use AI and more demanding applications become available.
Video generation, complex reasoning and autonomous systems can require far greater computing resources than basic text interactions.
The consumer rarely sees these differences. A short prompt may appear to require almost no effort, regardless of the infrastructure involved in producing the answer.
Behind the interface are data centers, electricity networks, specialized chips, contractors, workers and financial commitments extending across borders.
The industry distributes these costs unevenly. The people receiving the service, those providing the labor and the communities hosting the infrastructure do not necessarily receive comparable benefits.
This is one reason transparency has become central to the political debate.
In the European Union, transparency provisions under the AI Act began applying in August 2026, including requirements relevant to AI interactions and synthetic content. Certain existing systems have additional time to meet specific technical marking requirements.
These measures are meaningful, but identifying AI-generated material addresses only part of the problem.
A label can tell a viewer something about how an image was created. It cannot, by itself, explain how a model was trained, whether the people whose work supported it were fairly treated, why a particular answer was generated or how a platform’s commercial interests shape the experience.
The public debate has often concentrated on the extraordinary capabilities of AI systems. Less attention has been paid to the ordinary institutions that determine how those capabilities are deployed.
The companies building these services make decisions about access, pricing, training materials, product safety and permitted uses. Governments determine which obligations can be enforced. Publishers, artists and workers negotiate over the value of their contributions, frequently from unequal positions.
None of this makes artificial intelligence inherently hostile to human interests.
Its benefits are substantial: greater access to knowledge, new creative possibilities, assistance with complex tasks, improved accessibility and scientific tools capable of accelerating research.
Those benefits deserve to be assessed alongside the business arrangements required to deliver them.
The deeper concern is how much authority people are prepared to delegate to services whose internal workings remain difficult to examine.
An AI system can draft a message, recommend a purchase, assemble a news summary, produce a video or provide a conversational response. Each action may be useful on its own.
Taken together, these capabilities create an expanding class of intermediaries through which people encounter culture, information and other forms of social experience.
The direction of that development will depend on decisions about competition, public oversight, labor rights, data protection and creative ownership. Technology can change quickly; the institutions responsible for governing it tend to move more slowly.
What happens next is therefore unlikely to be decided by the capabilities of AI models alone.
It will also depend on whether people can understand the systems they rely on, challenge the decisions made through them and preserve meaningful alternatives.
For a long time, the defining commercial question of the internet was how to attract and retain an audience. The rise of generative AI adds another: who gets to build the experience that audience is offered, and under what conditions?
The answer will shape far more than the future of entertainment.
