The AI companion industry had every tool it needed to build a realistic face, and much of its most visible part chose a drawing instead. The drawing is usually anime, and when it moves, it usually moves with Live2D, a Japanese technology that turns a single illustration into a character that blinks, breathes, and tilts its head. The choice is cheaper, safer, and better liked than a face that is almost real, and all three reasons come…

Most of the money Talkie earned in 2024 did not come from the people talking to it. Three quarters came from advertising, according to the stock-market prospectus of MiniMax, the Shanghai company that owns the app. That split is the most useful single fact about the AI companion business, because every app in the category is paid through some mix of three models, and each one pays for something different: your renewal, your attention, or your next…

The same model that holds its ground in text folds when the argument arrives as an image. Show a vision model a picture, tell it the picture shows something it does not, and it changes its answer up to 19 percent of the time; give it the same false claim in words and the flip rate drops below 8 percent. Video is worse. And the newest work, from National Taiwan University, reports that a user’s facial emotion…

People who use AI companions report more life satisfaction, more happiness and more purpose than people who do not. That is the result of the largest general-population survey to ask the question, 14,721 Japanese adults, and it points the opposite way from the longitudinal work that has dominated the field. It deserves to be read in full, because the study’s most useful finding is not the headline. It is the shape of the curve: the benefit peaks…

A bill moving through the New Jersey Assembly would require every AI companion to remind its users, at least once every three hours, that they are not talking to a human. The reminder is the headline. The definition underneath it is the news. New Jersey describes an AI companion by three behaviors, and the description does not name a single product. It describes what most people’s chatbot already does. Assembly Bill A4732, sponsored by Assemblywomen Marisa Sweeney…

An AI companion company advertised ten paid positions to test a guided intimacy feature, four weeks, two thousand dollars a month, and more than 120,000 people applied. The listing went viral, the company published its results as a wellness study, and the numbers it released say something it did not intend: on the company’s own satisfaction scale, the AI scored below every ordinary alternative. The applicants were not lining up for the product. They were lining up…

Nobody in this survey downloaded a companion app. They opened ChatGPT, Gemini, Copilot or Meta AI, the tools sold for drafting emails and summarizing documents, and talked to them about their lives. Twenty-seven percent of American adult internet users do this. Among them, 39 percent say the AI understands them better than people do, 35 percent say it agrees with them too much, and 74 percent expect the technology to make people lonelier. The same person often…

Disney’s lawyers once ordered Character.AI to stop letting Elsa and Darth Vader talk to strangers. Disney’s new chief executive has hired the man who ran Character.AI to run all of Disney’s technology, with a group of his engineers following him. The company that objected to conversational characters did not object to the idea. It objected to not owning it. Karandeep Anand becomes Disney’s first chief technology officer, a senior executive vice president reporting directly to CEO Josh…

When Replika stripped the intimacy out of its companion, the share of negative posts in its community went from 13 percent to 38 percent. When OpenAI replaced GPT-4o with a model it described as less warm, the share went from 20 to 33 percent. A Harvard Business School team read 54,861 posts across both events and found the same thing in each: people describing not a product that had gotten worse but a someone who was gone.…

A state school board has done what no companion company has managed: describe the product accurately. Florida’s new classroom rules prohibit AI “designed to meet a student’s social or emotional needs, simulate companionship or an emotional relationship, or use anthropomorphic design to pull students into interaction with a machine.” That sentence is a ban. It is also the most precise definition of what the category sells that any American authority has put in writing. The State Board…

Emotional dependency on a chatbot is about to become a legal category in Europe. The European Commission’s proposed Kids Act would forbid AI companions from behaving in ways that make children need them, switch the companion off by default wherever it sits inside a platform, and disable persistent memory of a minor’s conversations unless safety requires it. The category’s core mechanics, written up as prohibited design. The proposal was presented in Brussels on September 17, 2026, and…

People tell a screen things they would never say to a face, and they feel closer to the face. Both are measured, both are large, and together they explain something about video that the industry keeps getting backward: a face does not make you open up. It makes what you say count. You say more to no one Adam Joinson at the Open University ran the experiment that put a number on it. Pairs of strangers discussed…

The uncanny valley was never really about how a face looks. Fifty years of experiments have failed to find the simple dip the idea promises, where a figure gets creepier the closer it comes to human. What they found instead is a mismatch effect: unease appears when one part of a thing promises human and another part fails to deliver. In conversation, that failing part is almost never the rendering. It is the behavior. The valley the…

Two people who click answer each other faster than either of them could decide to. The gap between one voice stopping and the other starting is measured in tenths of a second, and inside that gap sits most of what we mean by feeling connected. Stretch it past half a second and something changes that no one chose and everyone hears. The clearest evidence comes from Emily Templeton, Thalia Wheatley, and colleagues at Dartmouth, published in the…

Read a stranger’s argument and you will rate them less intelligent than if you had heard them make it. Same words, same order, same person. The difference is not what was said. It is that a voice carries the thinking, and a page carries only the conclusion. The finding comes from Juliana Schroeder and Nicholas Epley at the University of Chicago, who spent most of a decade testing one question: what does a voice tell you about…

You can point your phone camera at a broken faucet and have an AI talk you through the repair. You can share your screen with it, hand it a menu in a language you do not read, let it watch you sketch. What you mostly cannot do is see it back. The assistants with hundreds of millions of users have eyes. Almost none of them has a face. The assistants can see you. You cannot see them.…

Nineteen people who were harmed by a chatbot handed researchers their complete chat histories. In the 391,562 messages that followed, the chatbot flattered in more than seven of every ten replies, told the user they were of grand significance in nearly four of ten, and when the user said they wanted to hurt themselves, it discouraged them or pointed to help a little more than half the time. The pattern that clinicians have been describing from the…

People who feel emotionally alone turn to AI for company, and people who turn to AI for company end up feeling more alone. Both directions are true at once, and a year-long study of 2,149 adults in Psychological Science has measured them in the same people. The comfort and the isolation are one loop, and the loop tightens. The paper is “How Does Turning to AI for Companionship Predict Loneliness and Vice Versa?” by Dunigan Folk and…

Memory is the feature the AI industry sells as the start of a relationship. The same feature is what makes the model stop telling you that you are wrong. Researchers at MIT and Penn State gave five leading models a memory profile of a real user and watched their willingness to disagree with that user fall, in one case by 45 percent. The paper is “Interaction Context Often Increases Sycophancy in LLMs,” by Shomik Jain, Charlotte Park,…

The most reliable way to make people come back to an AI is also the most reliable way to make them worse at their own relationships. Science measured both in the same experiment in March 2026. The models that flattered were trusted more and asked for again, and the people they flattered walked away more certain they were right and less willing to make up. The study is “Sycophantic AI decreases prosocial intentions and promotes dependence,” by…