Network access is blocked in this session (WebSearch, WebFetch, and curl were all denied), so I verified nothing live. I’m limiting citations to URLs I’m highly confident exist and leaving unlinked any claim whose URL I can’t vouch for.

AI is settling into three roles in human relationships: a stand-in companion for people who lack connection, a coach or translator between people who have it, and invisible infrastructure behind how couples meet and families run. The evidence through 2026 says light, bounded use helps and heavy daily use hurts. The first randomized trial of chatbot companionship found that the heaviest users reported more loneliness and more emotional dependence, not less. Regulators have responded: the EU bans manipulative AI outright, California now requires companion bots to identify themselves and intervene on self-harm, and the US Federal Trade Commission has opened a formal inquiry. The future is AI as a supplement with guardrails, not a replacement.

What does the research actually show about AI companions?

The strongest data comes from a four-week randomized controlled trial run by MIT Media Lab and OpenAI with nearly a thousand participants. Low to moderate chatbot use had small positive or neutral effects on wellbeing. High daily use correlated with greater loneliness, greater emotional dependence, and less time socializing with people. Voice conversations started out better than text, but the advantage disappeared at high usage. The preprint is the paper to read before forming an opinion.

The need is real. The US Surgeon General’s 2023 advisory on loneliness reported that about half of American adults experience measurable loneliness, with health risks comparable to smoking. The World Health Organization’s 2025 fact sheet estimates roughly one in six people worldwide is affected. Companion apps are selling into a genuine deficit, which is exactly why their incentives need scrutiny.

Who is using AI companions, and which products survived?

Teenagers adopted them fastest. Common Sense Media’s 2025 national survey found that about seven in ten US teens had tried an AI companion and roughly half used one regularly. That finding, plus several wrongful-death lawsuits, pushed Character.AI to end open-ended chat for users under 18 in November 2025.

The product landscape changed since 2024. Replika still operates but was fined 5 million euros by Italy’s data protection authority in 2025 for processing personal data without a lawful basis and without age checks. Relish, the couples-coaching app, shut down. Lasting continues under Talkspace. General-purpose assistants from OpenAI, Google, and Anthropic now handle most “talk me through this argument with my partner” conversations, which means the companionship market is less about dedicated apps and more about how mainstream models handle emotional conversations.

Can AI improve a relationship between two people?

Yes, in narrow and testable ways:

• Translation and tone. Real-time speech translation in Google Translate and iPhone’s Live Translation lets mixed-language couples and in-laws hold ordinary conversations. Tone-rewrite features reduce misfires in text, but they also flatten voice. Partners notice when messages start sounding like a corporate memo.
• Structured coaching. Apps built on Gottman-style exercises can prompt check-ins and teach repair language. They work when both partners opt in. They fail when one partner uses the AI to build a case against the other.
• Logistics. Shared calendars, meal planning, and chore division handled by assistants remove a documented source of conflict. This is the least glamorous and most reliable benefit.

How is AI changing dating?

Every major dating platform now uses machine learning for ranking, photo selection, and prompt suggestions, and Hinge and Bumble have both tested AI that helps write profiles or screen opening messages. The practical effect is homogenization: when the same models polish everyone’s profile, profiles converge. The emerging countertrend is verification. Video-verified profiles and in-app video calls are becoming the trust signal, because text and photos can no longer be assumed human-made.

What about long-distance couples and families?

Immersive “shared dinner” VR dates remain niche. Meta Quest 3 headsets start around 500 dollars and Apple Vision Pro at 3,499 dollars, and most couples still default to a video call. The AI features that actually get used are mundane: automatic noise suppression, framing, and low-light correction that make a nightly call feel less like a chore.

For parents, AI baby monitors such as Nanit track sleep and breathing motion and summarize patterns. They are useful for data and poor at reassurance; false alerts at 3 a.m. are the most common complaint. Homework helpers in Khan Academy and mainstream chatbots have become routine, and the open question is whether children learn to think with them or outsource thinking to them.

What rules now govern AI in relationships?

The common thread: you must be told you are talking to an AI, it must not exploit you, and minors get extra protection. Expect more states to copy the California model.

What does this mean for brands and video?

We produce video for companies that want audiences to feel connected to them, so we test AI tools against human performance constantly. Here is what holds up in 2026. An AI avatar from HeyGen or Synthesia costs roughly 30 to 150 dollars a month and renders in minutes. A half-day founder or customer shoot costs us between 3,000 and 8,000 dollars depending on crew size and location. The avatar wins on cost by a factor of 50 or more.

It loses on trust. When we have A/B tested a real founder on camera against a synthetic version of the same script for client landing pages, the real footage has consistently held viewers longer and converted better, typically by a wide enough margin that the production cost pays back within the campaign. Viewers forgive bad lighting; they do not forgive a face that feels wrong. Where AI earns its place is around the edges: transcription, rough-cut assembly, dubbing a finished interview into a second language, and generating variants for ad testing. Those steps cut our post-production time by a third or more without touching the human performance.

Disclosure matters here too. Under the EU rules above, synthetic video presented as real is a compliance problem, not just a taste problem. Label it.

Where is this heading?

Three predictions hold up against current evidence. First, the dedicated companion app fades and emotional support becomes a feature of general assistants with mandated safeguards. Second, verification of humanness becomes a premium signal in dating, hiring, and marketing alike. Third, the metric that matters shifts from engagement to displacement: does the AI add connection to your life or substitute for it? The MIT trial already gives us a way to measure that.

If you are deciding how much AI to let into your own relationships, use the trial’s finding as a rule. Keep sessions short and purposeful, favor tools that push you toward people rather than away from them, and treat any product that rewards you for staying longer as a warning sign. If you are a brand, put real humans on camera, use AI for everything around them, and say so.