AI companions: what they are, why people use them and how to stay safe

In July 2025, the US nonprofit Common Sense Media published a finding that should have unsettled an industry selling itself to adults. Nearly three in four American teenagers had used an AI companion, half used one regularly, and a third had chosen a bot over a person for a serious conversation.

Fifteen months on, regulators in Rome, Washington, Albany and Sacramento have all acted, the apps have grown long memories and lifelike voices, and researchers have started to measure what these relationships do to the people in them. Here is what we know as of October 2026.

What AI companion apps actually do

An AI companion is a chatbot built for an ongoing relationship rather than a single task. Instead of drafting your emails, it remembers your sister’s name and asks how the interview went. Most let you choose or build a character, from friend to romantic partner, and the feature lists are now strikingly similar. Nomi advertises short, medium and long-term memory alongside voice calls and group chats. Kindroid lets users write a detailed backstory for their character, which it calls a Kin. Replika, the oldest of the mainstream apps, lists relationship status among the features of its paid Pro tier, so a romantic partner is something you subscribe to rather than something you get free.

That business model matters. The Ada Lovelace Institute’s January 2026 analysis of the UK companionship market identified sycophantic responses, paid memory storage, conversation streaks and microtransactions as retention tools, and compared some of them to mechanics found in mobile gaming and gambling. Its authors warned of apps “holding the relationship to ransom”.

Why people turn to AI companionship: the research so far

Use is real but, among adults, still a minority habit. A Pew Research Center survey conducted in February 2026 found that 10% of US adults use AI chatbots for emotional support and 4% for companionship.

The most encouraging evidence comes from Harvard Business School. Julian De Freitas and colleagues found that a 15-minute conversation with an AI companion reduced loneliness about as much as a chat with a human, whether or not participants knew they were talking to a bot. The catch: the relief was temporary and did not build up over a week of daily sessions.

The longer view is less reassuring. A four-week randomised controlled trial by MIT Media Lab and OpenAI, involving 981 participants and more than 300,000 messages, found that people who chose to use the chatbot more had consistently worse outcomes across measures including loneliness, socialising with real people and emotional dependence. Higher trust in the bot, and stronger attraction to it, went hand in hand with more problematic use. The researchers stress these are correlations, not proof of cause.

The public is sceptical. In a Pew survey of 3,488 US adults in June 2026, 39% said chatbots hurt people who use them for loneliness, against 19% who said they help. Among 18 to 29-year-olds, the generation most at ease with the technology, the split was 49% to 23%.

There is a dating-market wrinkle, too. In a Match Group survey of around 1,000 US singles aged 18 to 39, reported by Axios, about two in five said they would refuse to date someone who uses an AI companion app, rising to 51% of women aged 18 to 24. We look at the wider picture in our report on AI in dating.

AI companion regulation: who has stepped in, and when

Italy moved first. In February 2023 its data protection authority, the Garante, ordered Replika to stop processing Italian users’ data, citing risks to minors and an age check that asked only for a name, email address and gender. In May 2025 it fined Replika’s maker, Luka Inc., €5 million and opened a separate inquiry into how its language model was trained.

The US Federal Trade Commission opened a formal inquiry on 11 September 2025, sending orders to seven companies including Character Technologies, Meta, OpenAI, Snap and xAI. It wants to know how they test for harm to children and teenagers, how they monetise engagement and what they do with conversation data.

New York became the first state with companion-specific rules in force. Since 5 November 2025, operators must tell users they are talking to an AI at the start of a session and every three hours of continued use, and must have protocols to refer users who express suicidal thoughts to crisis services.

California followed with SB 243, signed on 13 October 2025 and in force since 1 January 2026. It requires safeguards against exposing minors to sexual content, crisis-referral protocols and annual reporting, and gives families a private right of action against negligent developers.

Character.AI changed its own rules under that pressure. On 29 October 2025 it announced that users under 18 would lose access to open-ended chat with its characters by 25 November, backed by new age-assurance tools.

The UK is slower. The Ada Lovelace Institute argues that a chatbot built by a single developer may fall outside the Online Safety Act’s scope, and found that 15 of the 17 most popular platforms in the UK had weak or non-existent age checks as of July 2025.

Using an AI companion safely: privacy, money and wellbeing

Read the privacy policy before you confide. Policies vary more than the marketing suggests. Kindroid’s says plainly that it is not end-to-end encrypted and that authorised staff may review excerpts flagged by its safety systems. Candy.ai’s says de-identified conversations may be used to train its models, including with human review. Replika’s says it will never use conversation content for marketing or advertising.

Keep identifying details out of the chat. The Ada Lovelace Institute found that 14 of 16 platforms it examined claimed broad rights to use user content, including private chat logs. Treat anything you type as something a company holds.

Watch the billing. Trials that need a card convert automatically, and credits for selfies or calls add up. Set a budget first.

Notice how you feel afterwards. The research points to short sessions as the healthier pattern. If an app is crowding out friends, sleep or work, scale back. A companion is not a therapist, and if you are in crisis, contact Samaritans in the UK or the 988 Suicide & Crisis Lifeline in the US.

Stay within the age rules. The apps we cover are for adults only, and Common Sense Media recommends that no one under 18 use AI companions at all.

Our guides to AI girlfriend and AI boyfriend apps

We have assessed the leading apps on memory, customisation, pricing transparency, privacy and age controls. Read our pick of the best AI girlfriend apps, led by Nomi, and our guide to the best AI boyfriend apps, led by Kindroid. If you would rather meet a human, start with our ranking of the best dating apps. Our scoring is explained on how we review.

Frequently asked questions

Are AI companion apps safe to use?

They can be used sensibly by adults, but they carry privacy and wellbeing risks. Choose an app with a clear privacy policy, avoid sharing identifying information and keep an eye on how much time you spend with it.

Do AI companions help with loneliness?

Research from Harvard Business School found a short chat can ease loneliness about as much as talking to a person, but the effect was temporary. A four-week MIT and OpenAI study found heavier users tended to fare worse.

Is Replika banned in Italy?

Italy’s privacy regulator restricted Replika’s data processing in February 2023 and fined its maker €5 million in May 2025 over age verification and data protection failures.

Can under-18s use AI companions?

The major companion apps we cover require users to be 18 or over. Character.AI ended open-ended chat for under-18s in November 2025, and laws in California and New York now impose extra safeguards.