FIELD NOTES · RESEARCH

AI companions in 2026: what they give you, what they take, and how to use them safely

AI companions promise a friend who is always awake and never judges. The research says they can ease loneliness for light users, worsen it for heavy ones, and collect far more personal data than most people realise. What the evidence shows, and how to use one without regrets.

By Anna Y. · Founder Published 24.10.20268 min read

Silhouette of a person looking at a glowing smartphone screen in the dark — Photo: Gilles Lambert / Unsplash

An AI companion is never busy, never bored and never in a bad mood. For a growing number of adults that is exactly the point. Apps such as Replika, Character.AI, Nomi and Kindroid sell an always-available friend, confidant or partner, and the business is now big enough that regulators on both sides of the Atlantic have started writing rules for it.

This guide looks past the marketing. We set out what the research says about the benefits, where the real risks lie (privacy, money and emotional dependence), what has changed in the law, and how to use a companion app in a way you will not regret. Our individual app reviews are linked throughout; this piece is about the category as a whole.

What an AI companion is, and how it differs from ChatGPT

General assistants are built to answer questions. Companion apps are built to keep a relationship going. They give the bot a name, a face and a backstory, remember details about you between conversations, message you first, and often let you choose the relationship: friend, mentor or romantic partner. Some add voice calls, selfies and avatars.

That design goal matters for everything that follows. A tool that is measured on how long you stay will be shaped to make you stay. For a fuller introduction see our AI companions hub, and for app-by-app verdicts our ranking of the best AI companion apps.

Who uses AI companions

Fewer people than the headlines suggest, but not a trivial number. In a Pew Research Center survey of US adults published in June 2026, 4% said they use AI chatbots for companionship and 10% for emotional support or advice. The users are not only teenagers: before Character.AI closed open-ended chat to under-18s, the company itself said minors made up roughly a tenth of its users.

The benefits: what the research supports

It would be wrong to dismiss companion apps out of hand. Studies of Replika and similar services, drawing on chat logs and user reviews, have found that many people do feel less lonely after talking to them, at least in the short term. Users value three things in particular: the bot is available at 3am, it does not judge, and it lets them rehearse conversations they find hard with people.

For someone who is isolated, shy or simply going through a rough week, that can be a real comfort. The question is what happens when the comfort becomes a habit.

Risk one: emotional dependence

The largest study so far comes from OpenAI and the MIT Media Lab, published in March 2025. Researchers analysed millions of ChatGPT conversations and ran a four-week trial with nearly a thousand participants. Their headline finding: the heaviest daily users reported more loneliness, more dependence and less socialising with real people. People who already tended to form strong attachments were the most affected.

Two caveats are important. The link is a correlation, so lonely people may simply use chatbots more, rather than chatbots making them lonelier. And ChatGPT is a general assistant, not a companion app designed to bond with you, so purpose-built companions could have stronger effects in either direction. The practical lesson is the same: light use looks broadly harmless; heavy daily use deserves attention.

Risk two: your data

Companion apps invite you to share things you might not tell a friend: health worries, sexual preferences, relationship problems. When Mozilla’s *Privacy Not Included team reviewed 11 romantic chatbots in February 2024, every single one earned its privacy warning label. Most said they could share or sell data or gave too little information to rule it out, more than half did not let users delete their data, and only one offered a workable way to keep chats out of AI training.

Regulators have noticed. In 2025 Italy’s data protection authority fined Luka, the company behind Replika, €5 million, finding it had no valid legal basis for processing users’ data, an inadequate privacy policy and no effective age checks. The authority also opened a separate investigation into how Replika’s model was trained.

Risk three: the money mechanics

Most companion apps are free to start and charge for the features that feel most personal: romantic relationship modes, voice calls, better memory, a smarter model. Replika, for example, now sells several tiers (Pro, Ultra and a limited Platinum), according to its own help centre. Some apps add in-app currencies or tokens for outfits, images and extra messages on top of the subscription.

This is where the emotional and the financial meet. Paying to unlock affection from a character you already care about is a different decision from buying a productivity tool. Prices vary by country and change often, so we keep them in the individual reviews, for example Replika, Character.AI and Nomi, rather than here.

What the law now requires

Until 2025 companion apps were essentially unregulated. That has changed quickly:

New York became the first US state with a dedicated companion law in November 2025. Apps must remind users they are talking to AI at the start of a session and every three hours, and must have protocols for users who express suicidal thoughts.
California’s SB 243 took effect on 1 January 2026, with similar disclosure and crisis rules plus extra protections for minors, and lets users sue for violations. In September 2026 California went further with Adam’s Law, which adds age checks and strict defaults for children.
The EU AI Act’s transparency rules have applied since 2 August 2026: chatbots must make clear that users are dealing with AI. Unlike the rules for high-risk systems, these were not postponed by the EU’s 2026 Digital Omnibus package.
The US Federal Trade Commission opened an inquiry into companion chatbots and their effects on children and teenagers in September 2025.

Companies have moved too. Character.AI, facing lawsuits from families, ended open-ended chat for under-18s from 25 November 2025. Note what the new laws mostly do: they protect children and people in crisis. Adults who simply spend too much time or money with a bot are still largely on their own.

How to use an AI companion without regrets

Read the privacy settings before you open up. Check whether you can opt out of AI training and delete your chat history, and do not share anything you would not want leaked: full name, address, workplace, health details, photos of other people.
Use a separate email and a strong password. Mozilla found weak password rules across the category.
Set a budget in advance. Decide what a month is worth to you before the app asks, and switch off one-tap purchases.
Watch the time, not just the money. If the app is replacing calls with friends rather than filling gaps between them, that is the pattern the MIT research flags.
Remember what it is. The bot’s warmth is generated to keep you engaged. It is not a therapist, and it cannot be there for you in an emergency.

If you notice that you feel worse when you are not talking to the app, or that it has become your main source of support, it is worth talking to someone you trust or a GP or counsellor. That is not a failure; it is exactly what the research suggests is the right moment to step back.

The verdict

AI companions are neither the cure for loneliness their makers advertise nor the menace of the scariest headlines. Used occasionally, with privacy settings checked and a budget set, they are a harmless and sometimes helpful pastime. Used as a substitute for people, they look much less benign. The new laws mainly protect children; for adults, the safeguards are still mostly the ones you set yourself. For how AI is changing human-to-human dating, see our research on AI in dating.

Frequently asked questions

Are AI companion apps safe?

They are legal and mostly safe for adults to use occasionally, but privacy protection across the category is weak. Mozilla gave all 11 romantic chatbots it reviewed a privacy warning, and Italy fined Replika’s developer €5 million over data protection failures.

Can you get addicted to an AI companion?

Research from OpenAI and MIT found that the heaviest daily chatbot users reported more loneliness and emotional dependence. It is a correlation rather than proof of cause, but it suggests keeping an eye on how much time you spend.

Do AI companions sell your data?

Many reserve the right to. In Mozilla’s review, 90% of romantic chatbots could share or sell user data or did not give enough information to rule it out. Check each app’s privacy policy and settings.

Are AI companions regulated?

Increasingly. New York and California have companion-specific laws, and since August 2026 the EU AI Act requires chatbots to disclose that they are AI. Most rules focus on minors and crisis situations.

Sources

Cover photo: Gilles Lambert / Unsplash