AI companions can ease loneliness in the short term, but longer-term evidence points towards withdrawal from people, and which outcome wins, for whom, is not yet known.
- In the short term, they help. Experimental work found companions reduced reported loneliness, through the experience of feeling heard.
- Over months, something shifts. A quasi-experiment, not a causal trial, found users' posts showed rising signals of loneliness and low mood against matched groups.
- One mechanism may explain both. Something always available can make the effort of human relationships feel costlier, so people reach out less.
- Use is large and young, and the law is responding. California now sets specific rules for companion chatbots, including AI disclosure and break reminders for minors.
- The useful question is what it replaces. Using one is nothing to be ashamed of; the check is whether it is a place to recover before going back to people, or has become the destination.
It works, in the short term
- Experimental work found that interacting with an AI companion reduced reported loneliness, and identified the mechanism as the experience of feeling heard — the sense that something has understood you and received it with attention.De Freitas, Oğuz-Uğuralp, Uğuralp & Puntoni, AI Companions Reduce Loneliness, HBS Working Paper 24-078, since published in the Journal of Consumer Research — an earlier version of this page called it a working paper only. Read at source 11 Sep 2026: “feeling heard explains levels of loneliness reduction”, and Study 4 “uses a longitudinal design and finds that an AI companion consistently reduces loneliness over the course of a week”. Described by its authors as among the first causal tests rather than correlational ones.
- A survey of 14,721 Japanese adults found associations with wellbeing were strongest among lonelier and less socially connected people — the group with least to lose and most to gain.Not read at source: cross-sectional panel surveys, Dec 2024 and Jan 2025. The publishing organisation is not named here because this site could not re-identify it at source on 16 Sep 2026 — treat the 14,721-respondent figure as unverified until it is
- Interview research consistently finds people turning to companions during grief, isolation or relationship breakdown, using them to rehearse difficult conversations or process feelings without fear of judgement.
Over months, something shifts
- A 2026 study combining long-term analysis of public posts with interviews found that, compared with control groups, companion users' posts showed increasing signals of loneliness and low mood over the following year."Mental Health Impacts of AI Companions: Triangulating Social Media Quasi-Experiments, User Perspectives, and Relational Lens", CHI 2026, read at source 11 Sep 2026. It is a quasi-experiment, not a causal trial — longitudinal Reddit data with stratified propensity score matching and difference-in-differences, plus interviews with 18 active users. An earlier version of this page called it causal, which overstates the method. It found “increases in language about loneliness, depression, and suicidal ideation” against matched comparison groups.
- Its proposed mechanism is the important part: unconditional, effortless availability raises the perceived cost of human relationships, which are "messy, unpredictable, and require effort". In the lead author's words, over time people stop reaching out.
- Separate research on teenagers found that a substantial share were using companion apps for emotional support rather than entertainment, and that the teenagers themselves were describing their attachment as unhealthy and affecting life offline.Drexel University ETHOS lab, CHI 2026, read at source 11 Sep 2026 — more than 300 Reddit posts from users self-identifying as 13 to 17 years old. Note the source says teens began using it “for emotional and psychological support or entertainment” and that use then “evolved into dependency”; the two are not alternatives, which an earlier version of this page implied.
Both findings are real, and they are not actually contradictory: short-term comfort and long-term withdrawal can be produced by the same mechanism. What has not been established is the conditions under which one outcome dominates — for whom, at what intensity, for how long, alongside what else in a person's life. The large, long, randomised studies that would settle it do not exist yet and will not for some years. Anyone offering a confident verdict in either direction is ahead of the data.
Be wary of a confident verdict either way. The availability that comforts in the short term may pull people away from others over time, and the long randomised studies that would settle it do not exist yet.
How many people, and who
Adoption is large and young. Character.AI says it supports “over 20 million monthly active users”.Character.AI, Harnessing data at scale, 7 Feb 2025, read at source 16 Sep 2026: “Character.AI operates at scale, supporting over 20 million monthly active users across our services.” An earlier version of this paragraph added that more than half of those users are under 24, and that another platform had reached roughly 30 million users by 2024; both came from third-party trackers this page could not verify, and were removed. Estimates of US teenage use vary widely — a spread that comes from disagreement over what counts as a companion app and what counts as regular use. No survey figure is quoted here, because none was read at source for this page. A range is more honest than any single figure.
Read any single usage figure with care. Estimates of teenage use differ with what counts as a companion app and as regular use, so the range is the honest answer.
The rules arriving
California's Senate Bill 243, approved by the Governor on 13 October 2025, sets specific requirements for companion chatbots: a clear notice that the chatbot is not human wherever a person could be misled, break reminders at least every three hours for users known to be minors, a protocol for suicide and self-harm content, and measures against sexually explicit material for known minors.California Legislature, SB-243 Companion chatbots, chaptered text, read at source 17 Sep 2026: “Approved by Governor October 13, 2025.” and, for a user the operator knows is a minor, “Provide by default a clear and conspicuous notification to the user at least every three hours for continuing companion chatbot interactions that reminds the user to take a break and that the companion chatbot is artificially generated and not human.” An earlier version called it the first US state law of its kind and mentioned a settled wrongful-death case; neither was sourced, and both were removed. European provisions covering companion-style products are phasing in.
California went further on 10 September 2026. SB 1119, called “Adam’s Law” by the Governor’s office, requires operators to carry out a child-safety risk assessment before offering a new or substantially modified companion chatbot from 1 July 2027, and to have an independent child-safety audit by 1 January 2029 or before launch, whichever is later. SB 867 bans the sale of toys that include a companion chatbot until 1 January 2031.California Legislature, SB-1119 Companion chatbots: children’s safety, chaptered text, read at source 22 Sep 2026: “Approved by Governor September 10, 2026”; operators must, “beginning July 1, 2027, before making a new or substantially modified companion chatbot available to users in the state”, carry out a risk assessment; and “On or before January 1, 2029, or before an operator first makes a companion chatbot publicly available, whichever is later, an operator shall ensure the performance of a child safety audit.” SB-867 Toys: companion chatbots, chaptered text, read at source 22 Sep 2026: “No person shall manufacture, sell, or exchange, possess with intent to sell or exchange, or expose or offer for sale or exchange to any retailer any toy that includes a companion chatbot.” and “This section shall remain in effect only until January 1, 2031, and as of that date is repealed.” The name comes from the Office of the Governor of California, 10 Sep 2026, read at source 22 Sep 2026: “Named after Adam Raine”. SB 1119 also deletes the SB 243 duties described above that apply only when an operator knows a user is a minor, including the three-hour break reminder, and puts its wider child-safety duties in their place.California Legislature, SB-1119, Legislative Counsel’s digest, read at source 22 Sep 2026: “This bill would delete those provisions imposing requirements on an operator if the operator knows the user is a minor.” The amended section 22602 in the bill keeps only the not-human notice and the suicide and self-harm protocol.
Federal action so far is a study and a privacy rule. The FTC’s inquiry into companion chatbots, opened in September 2025, has not published findings: its page lists only a resolution, a model order and a cover letter.Federal Trade Commission, 6(b) Orders to File Special Report … AI Companion Products or Services, read at source 22 Sep 2026. The page lists three documents, all from September 2025: the “Resolution Directing Use of Compulsory Process”, a “Model Order to File Special Report” and an “AI Companion Chatbot Cover Letter”. Separately, companies covered by the US children’s privacy rule (COPPA) had until 22 April 2026 to comply fully with its amendments, and in February 2026 the FTC said it would not bring COPPA cases against sites that collect data only to check a user’s age.Federal Trade Commission, FTC Finalizes Changes to Children’s Privacy Rule, 16 Jan 2025, read at source 22 Sep 2026: “Entities subject to the final rule will have one year from that publication date to come into full compliance with amendments that do not specify earlier compliance dates.” The FTC’s Federal Register notice page dates the publication “April 22, 2025”; one year on is 22 April 2026. FTC Issues COPPA Policy Statement to Incentivize the Use of Age Verification Technologies, 25 Feb 2026, read at source 22 Sep 2026: the Commission “will not bring an enforcement action under the Children’s Online Privacy Protection Rule (COPPA Rule) against certain website and online service operators that collect, use, and disclose personal information for the sole purpose of determining a user’s age via age verification technologies.”
What that tells you: this stopped being a fringe question. Legislatures and courts do not usually arrive early.
Treat this as a mainstream issue, not a fringe one. California already sets specific rules for companion chatbots.
There is nothing shameful in it, and this page is not an argument for stopping. The research suggests the useful question is not whether you use it but what it is replacing.
A reasonable check: is it a place to rehearse and recover before going back to people — or has it become the destination? Are you reaching out to friends and family as much as you were a year ago? The measured risk is not the conversation with the AI; it is the human conversations that quietly stop happening around it.
If you are talking to it about how you are really doing, that is worth taking seriously as a signal, not as an embarrassment. It usually means something is genuinely hard, and a person, a doctor or a helpline can offer things a companion app structurally cannot.
This is the one area on this site where the technology is not the main problem. People are not turning to machines because machines are irresistible; they are turning to them because being lonely is common and being heard is rare. A companion app is a response to that, and for some isolated people it plainly helps. The finding worth holding onto is the mechanism, not a verdict: something always available, never tired and never demanding changes what ordinary human effort feels like by comparison. Knowing that is most of the defence.