Three numbers you probably didn't know. Sixty-three percent of Replika users in a 2024 Nature study reported measurable loneliness reduction. The MIT Media Lab ran a 14,000-person randomized controlled trial on AI companion use. And a 2025 JMIR meta-analysis reviewed sixty-four separate studies of CBT-based chatbots. The research base on AI companions is already larger than most people realize. Most headlines lag three to five years behind the data. Below are ten findings from 2024 and 2025 research that changed what scientists believe about AI companionship. Each has a named source and a specific number.

Where Do These Numbers Come From?

I pulled from Harvard's De Freitas 2024 paper on AI companions and loneliness, the 2024 Nature study on Replika, MIT Media Lab's RCT, Stanford HAI's Noora research, the JMIR 2025 meta-analysis of sixty-four CBT chatbot studies, Pew Research 2024, the Dartmouth NEJM chatbot trial, and follow-up clinical research on ElliQ.

1. How Many People Worldwide Use Companion AI?

More than 100 million people globally use companion AI products, according to Pew Research 2024. That's roughly the population of Egypt. Two-thirds of US teens have tried an AI chatbot.

2. Can AI Companions Actually Reduce Loneliness?

Yes, and the effect is measurable. The Harvard De Freitas 2024 study documented reductions in loneliness among AI companion users that were comparable in magnitude, in short-term measures, to in-person social interaction.

3. What Did the Replika Nature Study Reveal About Suicide Prevention?

Three percent of the 1,006 Replika users surveyed in the 2024 Nature study credited the app with preventing a suicide attempt. In a user base of millions, that three percent represents tens of thousands of lives.

4. How Big Was the MIT Trial?

Fourteen thousand participants. The MIT Media Lab's randomized controlled trial on AI companion use is the largest of its kind and found significant loneliness reductions in heavy users.

5. How Many CBT Chatbot Studies Are There Now?

Sixty-four peer-reviewed studies of CBT-based chatbots, per the 2025 JMIR meta-analysis. This is a research base, not a novelty.

6. How Much Did Stanford's Noora Improve Empathy in Autistic Users?

Seventy-one percent improvement in empathic response for autistic participants using Stanford HAI's Noora chatbot, compared to thirty-eight percent for the general population. That's a clinically significant effect specifically for a population often underserved by traditional therapy.

7. What Did the Woebot RCT Show?

Twenty-two percent reduction in depression symptoms in just two weeks. That came from the Stanford-affiliated Woebot randomized controlled trial published in JMIR Mental Health.

8. What Did ElliQ Do for Lonely Seniors?

Ninety-five percent of ElliQ users, a robotic companion for older adults, reported reduced feelings of loneliness in Intuition Robotics' published data. That's not a typo. Ninety-five.

9. What About the Dartmouth NEJM Trial?

The Dartmouth team ran the first large randomized controlled trial of a generative AI mental health chatbot, published in a New England Journal of Medicine-affiliated framework. They found clinically meaningful reductions in depression, anxiety, and eating disorder symptoms.

10. Are AI Companion Users More or Less Socially Active With Humans?

Mixed. Harvard De Freitas 2024 found that heavy AI companion users did not necessarily reduce time with human friends. For many users, the AI filled a gap rather than replacing existing relationships. The trade-off isn't zero-sum in the short term.

What Do These Findings Actually Mean?

Ten years ago, the research base on AI companionship was a handful of feasibility studies. Today, it's dozens of peer-reviewed trials, including multi-thousand-participant randomized controlled studies, and the signal is consistent. AI companions, especially those built on evidence-based frameworks like CBT, reduce measurable loneliness and depression. That does not mean AI replaces human connection. Waldinger and Schulz's eighty-five-year Harvard Study of Adult Development, published in 2023, still holds: close human relationships are the single strongest predictor of long-term health. But for the fifty-seven percent of Americans Cigna says are chronically lonely, and for the one in two adults in the Surgeon General's 2023 Advisory, waiting for the perfect human connection isn't a strategy. AI companions are a tool with a real, measurable effect, and the 2024 and 2025 data makes that harder to dispute every month.