Ways to be heard
Is it healthy to talk to AI about your problems?
What two large studies found about talking to AI about personal things, the signs it's standing in for people, and how to make it lead back to them.
By the Ember team · Published November 8, 2026 · 6 min read
Talking to AI about personal things is no longer unusual. In a May 2026 survey by Elon University's Imagining the Digital Future Center, 27% of US adults who use the internet said they'd had significant emotional or social interactions with AI.
So a lot of people are asking, often of the AI itself: is this good for me?
The short answer is that saying a hard thing out loud can help, and an AI is often the easiest place to say it first. The research so far points to a narrower risk: the AI becoming the place where the conversation ends, standing in for the people in your life instead of leading back to them.
What the research says so far
Two large studies are worth knowing about.
Stanford and Carnegie Mellon, 2026. Researchers surveyed 1,131 US adults who use Character.AI, and 237 of them also donated their chat histories. People who said companionship was their main reason for using the chatbot reported lower well-being, and that link was stronger when their use was intensive and when they disclosed a lot. People with smaller offline social networks were more likely to use the chatbot mainly for companionship. The study, published in Nature Human Behaviour in August 2026, shows an association, not cause and effect.
MIT Media Lab and OpenAI, 2025. In a four-week randomized study of 981 people who talked with a chatbot by text or voice, researchers measured loneliness, time spent with real people, emotional dependence on the AI, and problematic use. The mode people were assigned made little difference. How much they used it did: people who chose to use the chatbot more had consistently worse outcomes.
Neither study says that talking to an AI is harmful in itself. Both point the same way: the trouble starts when the AI takes up the space where people would otherwise be. In Stanford's write-up of their study, the researchers say they want to explore ways to direct users to real human support when their chats show they need it.
When it tends to help
- Putting it into words for the first time. Some things are hard to say to someone who knows everyone involved. Saying it privately first can show you what you actually feel.
- Getting ready to talk to a person. What you want to say to your sister, your partner, your grown son. You can try the words before the real conversation.
- The hours when everyone is asleep. At 3 a.m., saying the loop out loud can take the edge off until morning.
What these have in common: the AI is a step toward something, not the end of it.
Signs it's standing in for people
A few honest questions:
- Have I stopped telling anyone else about this, even a small piece of it?
- Do I keep having the same conversation, with nothing changing?
- Does it mostly agree with me? An AI that always takes your side can feel good and leave you more stuck.
- Am I choosing it over people who are actually there for me?
If several answers are yes, it doesn't mean you've done something wrong. It may mean the conversation needs somewhere to go next.
How to make it lead back to people
- Tell one person one piece. Not everything. Just enough that this isn't only between you and a screen.
- Rehearse, then have the real conversation. Let the AI help you find the words, then say them to the person.
- Look for people who've been through it. A support group, a forum, someone a step ahead of you. What they lived through is a kind of knowledge no model has on its own. Where to find them.
- Notice when it's more than talking can hold. If you're thinking about harming yourself, or someone is hurting you, the section below is for you.
How we built Ember around this
We take both sides of this seriously. People are already bringing their lives to AI, and we think those private conversations can become a new way into real human support, with the AI as a bridge rather than the destination.
You talk privately with Ember by voice or text: on the web, in the iOS and Android app, or by text message or phone call. Behind the conversation is what real people in similar situations went through, what they tried, and what happened next, so Ember can tell you how it went for people who've been there. It won't tell you what to do. When what someone told could help a person standing where they stood, Ember asks that teller once; only with their yes can that person read it, under a nickname, and leave a note back. And when you're ready, you might meet people who get you.
Nothing you tell becomes a public post. No one can read your Ember unless you separately say yes. Ember is an AI; it isn't therapy or an emergency service.
We measure ourselves by whether strangers who've been through something similar become real support for each other, not by how long anyone talks to the AI. More on that: Why we're building a community without posts or feeds.
When you need more than a place to think
An AI is not a counselor, a therapist, or an emergency service. If you're in despair or thinking about harming yourself, you don't have to be sure it's serious enough: you can call or text 988 any time to reach the 988 Suicide & Crisis Lifeline. If you are in immediate danger, call 911.
Frequently asked questions
Is it bad to talk to AI about your problems?
Not in itself. Saying something out loud can help you see it more clearly. Research so far links heavier use, and using a chatbot mainly for companionship, with worse outcomes. The useful question is whether the conversation leads you back toward people or replaces them.
Can talking to AI make loneliness worse?
It can go that way. In a four-week study by MIT Media Lab and OpenAI, people who chose to use a chatbot more had worse outcomes on loneliness and time with real people. A 2026 Stanford study found companionship-focused use associated with lower well-being, especially when use was intensive. Both found associations rather than proof of cause.
Is it OK to vent to AI before talking to someone?
Yes. Many people find it easier to say something to an AI first and then take part of it to a person. It becomes a problem when the second step never happens.
What is Ember?
Ember is an AI you can talk to about what's on your mind, any hour, by voice or text. It listens, remembers what you said, and knows how things went for people who stood where you're standing — what they tried, and what happened after. It won't tell you what to do.
Is Ember private, and does it cost anything?
Ember is private: nothing you tell becomes a public post, and no one can read your Ember unless you separately say yes. Ember is in beta, and we’re starting to test subscriptions with some new accounts. Most accounts aren’t asked to subscribe. If yours is, you’ll see it up front — $9.99 a month after a free trial, and nothing is charged unless you choose to subscribe. Ember is for adults 18+.
Say it out loud first. Then let it lead somewhere.
Talk it out on Ember
Pause or stop anytime · 18+ · Nothing you tell becomes a public post.