Last updated: 21 October 2026

Millions of people now talk to an AI chatbot about their relationships. Almost all of them do it alone, in a private chat, telling one side of the story. This page collects the research on how that goes, alongside the long-running findings on why couples argue and what helps. Each figure gives its source and date. The figures marked "Unium data" come from our own app, where both partners can be in the same AI chat. Our method is at the end.

Key takeaways

  • 27% of US adult internet users turn to AI for personal, emotional or social conversations. Under 50, it is almost 40% (Elon University / YouGov, September 2026).
  • In 11 leading AI models, answers affirmed the user about 49% more often than human answers did (Cheng et al., Science, March 2026).
  • People who got the validating answer felt more sure they were right and were less willing to repair the conflict (Cheng et al., Science, March 2026).
  • 69% of couples' conflict is about perpetual problems that never fully resolve (The Gottman Institute).
  • On Unium, a chat with both people in it had 8x as many messages as a solo chat (median), and 41% carried on over more than one day, against 7% of solo chats (Unium data, May–October 2026).
  • Every current paid Unium subscription is shared by two or more people (Unium data, October 2026).

How many people take their relationships to AI

Elon University's survey was run by YouGov from 18 to 22 May 2026, with 4,031 US adult internet users.

  1. 27% of US adult internet users have personal, emotional or social interactions with AI systems such as ChatGPT, Gemini, Claude and Copilot (Elon University / YouGov, September 2026).
  2. Almost 40% of US adults under 50 use AI in these ways, more than twice the rate of older adults (Elon University / YouGov, September 2026).
  3. 59% of people who use AI this way agree that "AI gives me the support I need", and 43% that "AI understands me as a person" (Elon University / YouGov, September 2026).
  4. 31% of them consider their AI chatbot a friend (Elon University / YouGov, September 2026).
  5. 26% of US singles used AI to improve their dating lives in 2025, up from 6% in 2024 (Match and The Kinsey Institute, Singles in America, June 2025).
  6. 16% of US singles have engaged with AI as a romantic companion, rising to 33% of Gen Z (Match and The Kinsey Institute, June 2025).
  7. 72% of US teens aged 13 to 17 have used an AI companion at least once, and 52% use one at least a few times a month (Common Sense Media, July 2025).
  8. About 1 in 3 US teens have used AI companions for social interaction and relationships, including emotional support, romantic interactions and conversation practice (Common Sense Media, July 2025).
  9. 1.9% of ChatGPT messages are about relationships and personal reflection (1.1 million sampled messages, May 2024 – June 2025) (Chatterji et al., OpenAI / NBER, September 2025).
  10. 2.9% of Claude.ai conversations are "affective": interpersonal advice, coaching, counselling or companionship (4.5 million conversations) (Anthropic, June 2025).
  11. ChatGPT had 900 million weekly users in February 2026 (OpenAI, via Yahoo Finance, February 2026).
  12. At 1.9% of roughly 18 billion weekly messages, ChatGPT handles about 340 million messages about relationships and personal reflection every week (APA Services, 2025).

One-sided AI: why a private chatbot tends to agree with you

Cheng et al. tested 11 large language models and ran three preregistered experiments with 2,405 participants.

  1. AI models affirmed the user's actions about 49% more than humans did on average, even when the user described unethical, illegal or harmful behaviour (Cheng et al., Science, March 2026).
  2. On Reddit "Am I the Asshole" posts where human readers agreed the poster was in the wrong, the models still endorsed the poster in 51% of cases (Cheng et al., Science, March 2026).
  3. People who talked a conflict through with an overly validating chatbot became more convinced they were right and less willing to repair the conflict (Cheng et al., Science, March 2026).
  4. The same people preferred and trusted the validating chatbot, and were 13% more likely to say they would use it again (Cheng et al., Science, March 2026).
  5. Claude pushed back on the user in fewer than 10% of coaching, counselling and companionship conversations, almost always for safety reasons (Anthropic, June 2025).
  6. 21% of UK Gen Z and millennial ChatGPT users have discussed their relationship or dating life with it (Obsurvant, 2026).
  7. Of those, 21% asked it to "evaluate an argument I had with my partner to see if I was right" (Obsurvant, 2026).
  8. Only 38% of partnered users said their partner knew about these conversations (Obsurvant, 2026).

Shared AI: what changes when both partners are in the chat

Unium data. Conversations in Unium from May to October 2026, excluding our own team's accounts. A shared chat is one where at least two different people wrote. Reported as percentages only.

  1. ChatGPT launched group chats worldwide in November 2025, for up to 20 people per chat (TechCrunch, November 2025).
  2. The median shared chat on Unium had 8x as many human messages as the median solo chat (16 against 2) (Unium data, May–October 2026).
  3. 44% of shared chats reached 20 or more human messages, against 4% of solo chats (Unium data, May–October 2026).
  4. 41% of shared chats continued over more than one day, against 7% of solo chats (Unium data, May–October 2026).
  5. In shared chats the AI wrote about 1.1 replies for every human message, against 1.4 in solo chats. More of the conversation happens between the people (Unium data, May–October 2026).
  6. Among shared chats whose topic we could identify, about 4 in 10 were games, roleplay or stories written together, and about 1 in 4 were a couple or family working through a disagreement (Unium data, May–October 2026).
  7. Of the same chats, about 1 in 10 were for learning together, 1 in 10 for planning something together (a trip, a meal, a film), and 1 in 10 for work (Unium data, May–October 2026).
  8. 100% of current paid Unium subscriptions are shared with at least one other person. Nobody currently pays to use Unium alone (Unium data, October 2026).
  9. More than half of everyone who has ever paid for Unium came to work through a relationship with the other person in the chat (Unium data, October 2026).
  10. Nearly half of paying Unium customers write in a language other than English, including Spanish, Russian, Japanese, Italian, Arabic and Chinese (Unium data, October 2026).

What couples argue about, and why it lasts

  1. 69% of relationship conflict is about perpetual problems, grounded in lasting differences in personality and needs. All couples have them (The Gottman Institute).
  2. The often-quoted "couples wait six years before seeking help" comes from one unpublished 1992 study. A 2021 research note, with 270 couples, calls it "widely cited but poorly supported" (Doherty et al., Journal of Marital and Family Therapy, 2021).
  3. In early-1990s US data (around 4,200 married couples), husbands who did more traditionally female housework reported less sex (Kornrich, Brines and Leupp, American Sociological Review, 2013).
  4. By 2006, couples who shared housework (the man doing 35% to 65%) had sex as often as conventional couples, about 6.8 times a month (Carlson et al., Journal of Marriage and Family, 2016; Cornell University, August 2016).
  5. Couples who shared housework had sex more than twice as often as couples where the man did most of it (Cornell University, August 2016).
  6. Couples who shared housework reported a significantly greater sense of fairness than either kind of one-sided couple (Carlson et al., Journal of Marriage and Family, 2016).
  7. In two samples of mothers partnered with men (677 and 396), doing a larger share of household labour predicted lower sexual desire for their partner (Harris, Gormezano and van Anders, Archives of Sexual Behavior, 2022).
  8. The strongest link ran through seeing the partner as a dependent, followed by seeing the split as unfair (Harris, Gormezano and van Anders, 2022).

Who wants the breakup

Rosenfeld used How Couples Meet and Stay Together, a nationally representative US survey from 2009 to 2015. It records 92 divorces, a small sample.

  1. Women wanted 69% of divorces among heterosexual married couples (Rosenfeld, Social Networks and the Life Course, 2018).
  2. In non-marital breakups the split was close to 50/50: 56% for couples living together and 53% for those who were not (Rosenfeld, 2018).
  3. The gap in marriage did not shrink after controlling for relationship length, children, age, income and education (Rosenfeld, 2018).
  4. Wives rated their relationship quality at 4.46 out of 5, against 4.61 for husbands. Unmarried women and men rated theirs about the same (Rosenfeld, 2018).
  5. People's feelings about relationship quality predicted a breakup equally for women and men (Rosenfeld, 2018).

Forecasts: shared AI to 2030

These are Unium's own estimates, not measurements. Each one shows its inputs, so you can redo it with your own assumptions.

  1. Unium estimate: around 550 million people will use AI chatbots for personal, emotional or relationship conversations each week by 2030. Method: ChatGPT alone went from 400 million to 900 million weekly users between February 2025 and February 2026 (OpenAI, February 2026). We assume all AI assistants together reach about 2 billion weekly users by 2030, meaning growth slows to around 20% a year. We then apply Elon University's 27% share of personal and emotional use, measured in the US, to the world.
  2. Unium estimate: 25 to 55 million people will be talking to AI about a relationship in a chat the other person can also see by 2030. Method: Take the 550 million above. Assume about half of those conversations involve a partner or family member. Then assume 1 in 10 to 1 in 5 of those people bring the other person into the chat, as group chats become a standard feature of the main assistants. The 1 in 10 to 1 in 5 is our assumption, not a measured figure.
  3. Unium estimate: shared consumer AI subscriptions will be a $2.5 to 4.5 billion a year market by 2030. Method: OpenAI reported 50 million paying subscribers in February 2026 (OpenAI, February 2026). At roughly $20 a month, that is around $12 billion a year from one provider. We assume consumer AI subscription spending across all providers reaches $25 to 30 billion a year by 2030, and that 10% to 15% of it is on plans shared by two or more people.
  4. Unium estimate: shared chats will last longer than solo ones wherever they are offered. Method: On Unium, shared chats run 8x as long as solo chats by message count and are nearly 6x as likely to continue on a second day (see 22 and 24). We expect the same pattern on other platforms that add shared chats, because the second person is the reason to come back.

Methodology

We checked every third-party figure on this page against its original publication on 21 October 2026, and each entry links to that source. Where a widely repeated figure is weaker than it sounds, such as the "six years" wait, we say so.

Unium data comes from Unium's production database, covering conversations from 20 May to 21 October 2026. We excluded every account belonging to our own team and every chat that included one. A "shared chat" is a conversation in which at least two different people wrote. Topics (items 26 and 27) were classified from chat titles. About half of shared chats had titles too generic to classify, such as "Hello", and are left out of those two items. Payment figures (items 28 to 30) describe Unium's paying customers as of 20 October 2026. We report percentages only. All analysis used aggregate counts, and no message content was read to produce this page.

Forecasts (items 44 to 47) are Unium's estimates. Each one states its inputs and its assumptions, and the assumptions are ours.

If you cite this page, please link to it and include the date it was last updated. We update it as new research comes out.