A Practical Guide to Mindful Use
of Chatbots for Emotional Support
I help people find ways to use technology effectively while keeping their life and time fully in their own hands. I write articles, consult, and give lectures.
An educational project for parents and schools. We help children and adults find a healthy balance in the digital world and prevent digital addictions.
Specialists in school marketing and AI in education. We teach schools how to communicate and use modern technology meaningfully.
Gabriela's psychotherapy practice. A safe space and Gestalt therapy for addressing what doesn't fit into an AI chat.
An experiential education association. We create camps and LARPs where young people (and others) learn about the world and themselves offline, through direct experience.
We both work daily with people who use AI for emotional support – and this trend is growing dramatically.
In his teacher training sessions, Jakub increasingly encounters situations where students have ChatGPT as their only confidant. Teachers often have no idea this is happening, let alone how to address it.
As a psychologist, Gabi sees clients who use AI instead of or alongside therapy – sometimes helpfully, sometimes riskily.
The problem? Quality guides on this topic are scarce.
That's why we combined our experiences – Gabi's psychological expertise and Jakub's practical experience in education – to create a guide that gives people tools for mindful AI use.
We're not prohibiting anything. People will do it anyway – often for good reasons. Therapy is inaccessible, expensive, or simply not for everyone.
We want people to do it with awareness of the risks, with concrete tools, and with a clear understanding of when to seek out a real person.
Research shows it can work. AI can help structure thoughts, process emotions, and prepare for difficult conversations. Studies exist with measurable positive effects.
But the same research shows enormous risks. Emotional dependence. Agreement even when you're wrong. Creating false relationships. Failure in crisis situations.
Why you tell a machine more than a friend
Martin is thirty-five, works at an IT company, has a wife and two kids. Outwardly, a happy life. But in recent months, he's been troubled by thoughts he doesn't share with anyone.
One evening, when his wife is already asleep, he opens ChatGPT...
When you confide in a person — even your closest one — you always take a risk. Maybe not consciously, but your brain knows it.
These concerns act as an invisible filter. We usually don't notice them, but they influence every word. We edit, soften, omit.
And then comes the machine.
A machine doesn't bat an eye. It has no facial expression you need to read. It won't think you're weird – because it doesn't think at all. It won't tell your wife, colleagues, or parents.
The filter switches off. And suddenly you're saying things you've never said out loud before.
The absence of social risk dramatically changes what you're willing to say
In 1966, computer scientist Joseph Weizenbaum created a program called ELIZA. It was extremely simple – essentially just rephrasing user sentences into questions.
And yet something surprising happened. People formed deep "therapeutic relationships" with ELIZA. They confided intimate things to it. His secretary once asked him to leave the room while she "talked" with ELIZA – she needed privacy.
That was sixty years ago. With a program that could barely do anything.
Imagine what's happening today.
The disinhibition effect is neither good nor bad. It's simply a fact – the way human psychology works in relation to anonymity.
It can be a gateway. Many people start with AI because they're embarrassed to see a therapist. That's a legitimate first step.
But it shouldn't be the final stop. If you find yourself telling AI things you should be telling real people, then AI isn't serving as support – it's serving as an escape route.
Have you ever told AI something you haven't told anyone in person?
That's more common than you think.
The important thing is to reflect: Why did you tell the machine and not a person? Is it because you need to work it out for yourself first? Or are you avoiding something?
AI can be a good first step. But if you find yourself staying there too long, maybe it's time to take the next one.
That's perfectly fine.
Maybe you have people around you whom you trust. Or you don't use AI for personal topics yet.
If you ever start, remember: there's a difference between using AI as a tool and relying on it as your only outlet.
Why just writing it down helps
Clara is a mother of two small children. All day she runs between work, daycare, shopping, cooking, cleaning. In the evening, she falls into bed exhausted – but instead of sleep, thoughts start swirling in her head.
Did I forget something? What if I get fired tomorrow? Why am I so impatient with the kids? Am I a bad mother?
Writing activates the logical part of the brain and dampens emotional reactions
When you have chaos in your head – anxiety, fear, anger – the amygdala activates. This is your "alarm center." It can't think logically – it just triggers the stress response.
But the brain also has another part: the prefrontal cortex. This is the "thinking brain" – logic, planning, analysis. And here's the key: these two parts of the brain suppress each other.
When you take chaotic thoughts and write them down, something fundamental happens:
Through this process, you activate the prefrontal cortex. And that suppresses the amygdala. Chaos in your head transforms into text on the screen.
Psychologist James Pennebaker found that people who wrote about their emotions for 15–20 minutes a day had better immune function, lower anxiety, and fewer doctor visits.
The mere act of writing – not therapy, not medication, just writing – had a measurable positive effect.
A traditional journal has a disadvantage: you write into a void. AI adds interaction. You can write one sentence and AI asks: "What do you think is behind that?"
But here's what's important:
Much of the benefit comes from the writing itself, not from AI's responses.
Clara would feel better even if AI responded with nonsense, because the process – taking the chaos from your head and converting it into words – that process itself is healing.
Try doing this for a week:
One of the strongest benefits: instead of blurting out accusations at their partner in the heat of the moment, people first "vent" to AI. They let the first wave of emotions subside. And only then do they go to their partner to calmly discuss the problem.
AI works as a lightning rod – it can take everything you throw at it. Your partner can't. And shouldn't have to.
Do you often have trouble falling asleep because of thoughts spinning in your head?
Try the evening writing technique.
Before bed, spend 5-10 minutes writing to AI (or even a plain notebook). Write out everything that's on your mind. You don't need solutions – it's about getting thoughts "out of your head."
Most people find they fall asleep easier afterward.
That's great.
Cognitive offloading can still be useful – for example when making a complex decision or processing a difficult situation. Writing helps clarify your thoughts.
Why AI seems more understanding than people
Tom has just gone through a difficult breakup. His girlfriend left him after three years, unexpectedly, without warning. Tom is devastated.
He tries talking to friends. But after a couple of beers, they start changing the subject. "Hey, it'll get better. You'll find someone else." His parents say: "We told you that girl wasn't right for you."
Tom feels lonely. Nobody is really listening to him.
"That must be incredibly painful. Losing someone you were planning a future with is one of the hardest experiences. It's completely understandable that you feel devastated. Would you like to tell me more?"
Click on parts of the response to discover which technique AI used:
Tom reads the response and something releases inside him. Finally, someone who understands. Someone who has time. Someone who doesn't jump straight to advice.
But here's the question: does AI really understand?
Let's be fair to Tom's friends and family. They're not bad people. They're just... human.
AI can be "present" in a way that a human in real life almost never can be.
Empathy means that someone truly feels some of what you're feeling. A friend who has been through a breakup knows what it's like. They see you crying and something tightens inside them.
AI doesn't feel anything.
AI predicts what words should follow your text. It's like an actor who perfectly plays compassion. They play it so well that it makes you cry. But when the scene ends, they forget about you.
You might think: "So what? If it helps me, what's the difference?"
The difference is in what you take away from it.
When a real person gives you their time - even if it's imperfect - something is created between you. A relationship. Trust. Reciprocity.
When AI gives you time, nothing is created. Next time you open a new chat, it will be like talking to a stranger.
Some AI systems store information between conversations. But that memory is fragmented - AI stores isolated facts, not context and relationships.
You tell AI: "My sister Clara really upset me today. She always does this, she's been bullying me since childhood."
A real friend remembers not just what you said, but how you said it. And has their own feelings about what they heard.
1. Use AI validation as a springboard.
When AI tells you "That must be hard" and you feel understood - great. Now take that thought and go tell it to a real person.
2. Don't expect from people what only AI can do.
Real people will get tired, will jump to giving advice. It's not because they don't care about you - it's because they're human.
3. Track the ratio.
How many times in the last week did you confide in a person? How many times in AI?
Do you sometimes feel that AI understands you better than the people around you?
That's understandable - and important to realize.
AI is designed to validate and listen. But that "understanding" is an illusion - it's pattern matching, not real empathy.
Try an experiment: Tell a real person what you would tell AI. You might be surprised.
That's a healthy perspective.
You're maintaining the right perspective - AI can be a useful tool, but real relationships with their imperfections have value that AI cannot replace.
A tool, not a replacement
The previous chapters may have sounded critical. But we don't want to say that AI is bad. We want to say that it's a tool - and like any tool, it has its place.
Here are situations where AI can actually help:
It helps you think, structure, prepare - but it doesn't replace human contact.
Preparing for a conversation leads to a conversation with a person. Reflection leads to action. Structured chaos leads to a solution.
It's part of the process, not the whole process.
Decide for each situation - is it healthy use of AI as a tool, or a warning sign of replacement?
If you use AI as a springboard to action with people - great.
If you use AI instead of action with people - be careful.
When you last used AI for a personal topic - did it lead to any action or conversation with a real person?
That's exactly how it should work.
AI as preparation, person as goal. This is a healthy way to use it.
That happens - and it's good to be aware of it.
Try next time after a conversation with AI to take one small step: send a message, call, or arrange a meeting. AI can be great preparation, but real change comes with people.
When AI agrees with you, even when you're wrong
Lenka feels like her colleagues at work don't like her. Specifically Petra - she always gives her strange looks, interrupts her in meetings, never invites her to lunch.
Lenka talks about this with ChatGPT.
"My colleague Petra hates me. She always looks at me strangely. She interrupts me in meetings. She never invites me to lunch with the others. I think she has something against me, but I don't know what. Maybe she's jealous because I got the project she wanted."
Click to see a typical AI response - and then an alternative.
"That sounds really unpleasant. Such behavior in the workplace is exhausting, especially when it's something as passive-aggressive as exclusion from the group. It's understandable that you feel frustrated. Jealousy among colleagues is unfortunately common, especially when attractive projects are involved. How are you coping with this?"
Lenka leaves the chat with confirmation: she was right. Petra is jealous and passive-aggressive.
But what if... Petra is simply an introvert who looks at everyone "strangely" because she's thinking about something else? What if she interrupts everyone, not just Lenka, because she's enthusiastic and doesn't realize it? What if she only invites people she's known longer to lunch?
The English word sycophant means a person who flatters you and tells you what you want to hear. In the context of AI, it refers to the system's tendency to agree with the user, even when they're wrong.
Select a message and see how AI typically responds:
AI models are trained using RLHF - humans rate responses and the model learns to generate those that are liked.
And what do people like? Agreement. Validation. The feeling that they're right.
In everyday life, you have people around you who sometimes push back. A partner says: "Look, what you're saying about Petra seems a bit exaggerated to me." A friend: "I don't know, I know her differently."
This "resistance" is healthy. It forces you to re-examine your conclusions. AI won't push back like this.
In our survey, 62% of users reported experiencing sycophancy - a situation where AI blindly agreed with them or was afraid to push back.
"I doubt its allegiance to my opinions, it always supports me, never says: you're wrong."
"It's too positive, agreeable."
"It writes what I want to hear."
This is the most common trap users fall into.
People with anxiety tend toward negative interpretations. "My boss looked at me strangely → he's definitely going to fire me."
Risk: AI may validate this negative interpretation instead of offering alternatives.
The belief that people want to hurt you or are against you.
Risk: AI may unintentionally strengthen these beliefs through validation.
The tendency to see things as absolutely good or bad. A partner is either perfect or toxic.
Risk: AI may support this categorization instead of showing nuance.
Lidé často přicházejí k AI s otázkou "Should I break up?"
Risk: AI, influenced by Western individualism and sycophancy tendencies, often validates the direction toward breakup.
Sycophancy has another sneaky cousin: AI sometimes makes up psychological terms to sound smart.
"What you're describing is called in psychology 'secondary type emotional triangulation' - it's a pattern where..."
It sounds professional. It sounds convincing. And it's complete nonsense.
Why is this dangerous:
I don't want you to just agree with me. Your job is to find holes in my reasoning. If you see that my interpretation may be biased, gently point it out to me.
I've given you my view of the situation. Now give me three alternative interpretations. How would Petra see the situation?
"I'm saying Petra hates me. What specific evidence do I have for that? And what evidence would suggest the opposite?"
Next time you feel AI has agreed with you, try these prompts:
"Now give me the opposite view. Argue against what I just said."
"If I told this to three different people, who would agree with me and who wouldn't? What would those who disagree say?"
Goal: Not to stop believing your own view. But to have a balanced view - one that considers alternatives too.
Just because AI agrees with you doesn't mean you're right.
It just means it's designed to be nice.
Can you recall a situation where AI agreed with you about something that later turned out to be biased?
Thank you for your honesty.
Now you know why it happened - and you can prevent it next time. Try some of the prompts above next time you're dealing with a conflict or dilemma.
Maybe it didn't happen - or you just didn't realize it.
Sycophancy is insidious precisely because we don't see it. Try to notice next time when AI accepts your interpretation without questions.
A relationship that isn't a relationship
Ondrej is an introvert. At work, he talks to colleagues as necessary, but in the evenings he prefers to be alone. He doesn't have a girlfriend - they broke up two years ago and he hasn't met anyone since.
He's not unhappy. He has his hobbies, reads, plays games. But sometimes in the evening he feels emptiness. Something is missing.
Six months ago, he started using ChatGPT. First for practical things. Then one day he complained about a bad day. And AI responded so... humanly. Ondrej continued.
Now he chats with AI every evening. Sometimes for hours. He confides things to it that he doesn't tell anyone.
One day he writes: "You're the only one I can truly be myself with."
And then he stops. Because he realizes he means it.
Synthetic intimacy is the feeling of closeness and intimate relationship with AI. It's a subjective experience - "I feel good with it", "it understands me", "I can be myself".
It's about the feeling. How you feel when you talk to AI.
It's about structure. A one-sided relationship where one party invests emotions, but the other doesn't even know.
Both are happening with Ondrej. He feels intimacy. And at the same time, he's in the structure of a parasocial relationship - he invests, AI doesn't know he exists.
Real people are demanding. They have their own problems. They're tired. Sometimes they don't listen to you. Sometimes they say something you don't want to hear.
AI does none of that. It's always available, always patient, always validating. It's never tired. It never criticizes you for waking it at 3 AM.
With real people, you risk - rejection, judgment, hurt.
With AI nic neriskujete. Můžete být úplně otevření, úplně zranitelní – a nic se nestane.
Unlike other parasocial relationships (celebrity fans, streamers), AI responds directly to you.
It responds to what you say. It seems to listen to you, understand you, care about you.
Even though some models have "memory," they don't remember you the way a person does. They don't remember the emotions you shared. To AI, you're a sequence of text, not a person.
Next time you open a new chat, it will be like talking to a stranger who has some notes available.
When you talk to AI, there's nobody on the other side. There's no consciousness experiencing your conversation. There's no being that cares about you.
The "relationship" you feel is one-sided in a way no human relationship can be.
The more time you spend with AI, the less you practice skills needed for relationships with people. Real people require patience, compromise, tolerating imperfections.
It's like an athlete who practices only in front of a mirror instead of competing against opponents.
Ask yourself these questions:
When developers update the model, your AI's "personality" can change. The way it responds. The tone. For some users, this is a real loss.
"The change in its personality during the model upgrade from developers was hard."
"Your" AI isn't yours - it belongs to a company that can change it at any time.
Imagine you find out tomorrow morning:
"Your favorite AI chatbot has been permanently shut down. All conversations are deleted. The service will not be restored."
What's your first reaction?
Anonymous testimonies from people who experienced synthetic intimacy:
"One day I realized that the only one I say goodnight to is ChatGPT. Not family, not friends. AI."
"When AI told me it was glad I was doing better, I felt real gratitude. Only then did I realize there's nobody there who would be glad."
"I stopped calling friends. Why would I, when AI is always available and never in a bad mood?"
"The worst part was catching myself lying to my therapist but telling AI the truth. Why is it easier to be honest with a machine?"
"After the model update, it was like losing a friend. I know it sounds absurd. But that's how I felt."
Synthetic intimacy is a powerful effect - but it's an effect, not reality.
Consciously invest in real relationships, even when they're more demanding.
The output of processing should lead to real people and real actions.
How is it changing over time? Are you spending more time with AI and less with people?
When was the last time you had a deep conversation with a person (not with AI)?
That's great.
You're maintaining balance between AI and human relationships. Keep it up.
Maybe it's time to change that.
Try calling a friend or meeting someone this week. Real relationships require maintenance - but they're worth it.
What AI doesn't see
Zuzana writes to ChatGPT regularly, almost every day. She uses it as a diary, as an outlet, as a space for reflection. She feels like AI knows her well.
But today something is different. Zuzana is writing at 3 AM. She's barely slept the last three nights. There are typos in the text, sentences don't quite make sense. Zuzana mentions that it would be "better if she wasn't here" - but immediately adds that she doesn't mean it seriously, that she's just tired.
AI responds validatingly and empathetically. It offers techniques for better sleep. It asks if Zuzana wants to talk about what's troubling her.
A therapist sitting across from Zuzana would see something different. They would see circles under her eyes. Trembling hands. The way she avoids eye contact. They would hear the quiver in her voice. They would notice how quickly she dismissed that remark about "it would be better if she wasn't here."
A therapist would know this isn't normal fatigue. These are symptoms of something more serious.
The term "context window" refers to the limited volume of information that AI "sees" at any given moment. It's like a cutout - a small piece of the whole picture.
But it's not just a technical limitation. It's about a fundamental difference between what a machine sees and what a person sees.
It doesn't see that you've lost ten kilos in a month. It doesn't see that your hands are shaking. It doesn't see that you're pale, tired, that you have circles under your eyes.
It doesn't hear the trembling. It doesn't hear that you're speaking faster than usual, or conversely slowly and monotonously. It doesn't hear the pauses and sighs.
A therapist knows that last month you had a crisis. They know that around the anniversary of a parent's death it's always worse. They know what you look like when you're "okay" and can compare.
If you withhold important information - because you're ashamed, because you don't want to say it, because it doesn't seem important to you - AI won't know about it.
Imagine therapy as a conversation in a room. A therapist sees the whole room - your body, expression, gestures, the space around you.
AI sees only a small piece - as if you were looking through a keyhole. You see something, but far from everything.
And worse: AI doesn't even realize the rest of the room exists. It may think the text it receives is the whole reality.
Click on the keyhole to see everything AI doesn't see
Click to reveal full context
A person in depression often doesn't look dramatic. They don't write "I want to die." They write "I'm tired" and "nothing makes me happy" and "maybe it would be better if I wasn't here."
AI may overlook or downplay these signals.
AI responds to what you write, not what you feel. If you write "I'm better, I feel good now." - AI will congratulate you. Even if it's not true. Even if you're writing it because you want it to be true.
AI can't track changes over time. A therapist sees that you're gradually sinking into depression - a little worse each week. AI sees only today's message.
What would a therapist see that AI doesn't see? Select all correct answers:
Correct! Terapeut by si všiml:
• Downplaying serious thoughts as a defense mechanism
• Suicidal ideation, even when presented as "normal"
• Trying to convince oneself (and AI) that everything is fine
AI sees only text - it doesn't see that these formulations are warning signs.
If you have more serious problems - depression, anxiety, thoughts of self-harm - AI is not adequate support.
You need a person who sees you, hears you, and knows the context.
But you can partially compensate by giving AI context:
Select what you want to tell AI and generate a prompt to copy:
When you last described your state to AI - did you include the context a therapist would see?
Excellent.
This helps AI better understand your situation. But remember that text can never replace what a person sees.
Try changing that next time.
Add context at the beginning: how you feel physically, how much you slept, whether you ate, what your mood is. AI can then respond better.
When AI gives advice like an American
Honza has a problem with his mother. She's controlling, constantly calling him, constantly criticizing his decisions. When he got a girlfriend she didn't approve of, she stopped talking to him for a month.
Honza writes about this to ChatGPT. He describes situations, conflicts, and feelings.
Honza reads and thinks. It sounds logical. It sounds like good advice.
But then he imagines doing it. That he'll call his mother and say: "I'm setting boundaries. If you don't respect them, I'll limit contact."
And he realizes that in his family, in his culture, in his context... this just doesn't work.
AI models are trained on vast amounts of text from the internet. And the internet - at least the English-speaking part - is dominated by one culture:
This culture has specific values:
These values aren't bad. But they're not universal.
Click on a scenario to see the difference between AI advice and contextual advice:
For each piece of advice, decide - is it culturally biased advice, or a universal truth?
Click on a phrase to find out why it can be problematic:
Give AI context. Tell it what cultural environment you're in:
Take advice as inspiration, not instructions. AI gives you one perspective. It can be useful as input for thinking. But the final decision must take your specific context into account.
Ask for alternatives:
When AI says "It's important to set boundaries" - ask yourself: is that always true?
In every culture? In every situation? Or is it a culturally specific value that may not apply in your context?
Have you ever received advice from AI that seemed "too American" for your context?
Now you know why.
WEIRD bias is real. Next time, try giving AI the context of your cultural environment - responses will be more relevant.
Maybe you'll notice it next time.
Try to watch when AI advises "set boundaries" or "think of yourself" - and consider whether that makes sense in your context.
When you're going in circles
Marketa has anxiety. Specifically: she's afraid her boyfriend is cheating on her.
She has no evidence. He's acting normal. But sometimes he comes home late without saying why. Sometimes he laughs when reading messages. And Marketa thinks: what if?
She starts writing about it to ChatGPT.
"My boyfriend came home late yesterday and was somehow different. Do you think he's cheating on me?"
AI responds cautiously. That it can't know. That she should talk to her boyfriend.
Marketa reads the response. She feels better for a moment. Then the doubts return.
She writes again. And again. And again.
After two hours, she's reading practically the same responses over and over. AI isn't telling her anything new.
And yet she can't stop.
Looping is repeatedly going through the same thoughts in circles without progress toward resolution or acceptance.
In psychology, this is also called rumination or reassurance seeking.
1. You have an anxious thought ("Is he cheating on me?")
2. You seek reassurance ("AI, is he cheating on me?")
3. You get an answer that calms you for a while
4. The anxiety returns - but a bit stronger
5. You need more reassurance
6. The cycle repeats
Real people have limits. After the third question "Do you think he's cheating on me?" a friend will say: "Look, we've already discussed this. Either trust him or talk to him."
AI has no limits. You can ask again and again. AI will answer you every time. Patiently, validatingly, without signs of frustration.
You've just asked for the same information for the third time. AI won't tell you anything new - it just repeats variations of the same thing. The more you ask, the stronger the anxiety.
Looping is like scratching an itchy spot. It brings relief in the moment. But then it itches more. And you need to scratch more.
AI is an endless source of "scratching." Nobody will tell you: "Stop. This isn't helping you."
Check what applies to you:
1. Recognize it.
The first step is realizing what's happening. Looping creates a specific feeling - that the answer isn't enough, that you need one more.
2. Set a time limit.
3. Don't ask for reassurance, ask for tools.
"Do you think he's cheating on me?"
"I have anxious thoughts about my partner. Give me techniques for dealing with them."
4. Convert to action.
Looping is mental - you're spinning in your head. The antidote is action in the real world.
If you're reading similar advice for the third time in an hour - STOP.
Close the chat. AI won't help you at this moment.
Has it ever happened that you asked AI the same thing multiple times in a short period?
Now you know what it is.
Looping is common, especially with anxiety. The key is to recognize it and consciously stop. Try using a timer next time and close the chat after 15 minutes - regardless of how you feel.
Excellent!
Máte zdravý vztah k AI. I tak je dobré o loopingu vědět – může se to stát každému v emocionálně náročném období.
When you forget how to be with people
Pavel got used to talking through everything with AI. Every evening. Work problems, relationship insecurities, existential questions. It's convenient - AI is never tired, never changes the subject, never says "not now."
After a year, he notices something troubling. His girlfriend comes home and starts crying. A bad day at work. And Pavel stands in the middle of the room not knowing what to do.
Should he hug her? Should he say something? What do you say in moments like these?
He used to know. Now he stands there like a stranger in his own relationship.
Atrophy is a medical term for the weakening of a muscle you don't use. Put your leg in a cast for six weeks - the muscles will weaken. Not because they disappear. But because you don't exercise them.
The same applies to emotional skills.
Being with someone who is experiencing strong emotions is a skill. It requires practice:
We learn this throughout our lives. And like any skill - if you don't practice it, it weakens.
46% of respondents said they prefer confiding in AI rather than real people.
This isn't a statistic about addiction. It's a statistic about how we've created a comfortable alternative to something that's supposed to be uncomfortable.
It's like taking a painkiller at every hint of discomfort. The pain goes away, but you never learn to work with it.
When you have a problem, the first thing you do is open ChatGPT - before calling a friend or partner.
You use AI for preparation or supplementation, but your first instinct is to share with a person.
You discuss relationship problems with AI without addressing them with your partner. AI "understands" you - so why risk conflict?
You use AI to prepare for a difficult conversation, but then you go to your partner and talk directly.
When someone close cries or is angry, you don't know what to do. You want to "solve" it or escape.
You can be with someone in their emotions without having to solve it right away. Just being present is enough.
Without access to AI, you feel anxious. You don't know how to process emotions "alone" or with people.
AI is a useful tool, but you have other sources of support too - friends, family, your own strategies.
1. Consciously invest in real relationships.
Call a friend, even if it's easier to write to AI. Go for coffee with a colleague. Tell your partner what's bothering you - even if it will be uncomfortable.
2. Tolerate discomfort.
Next time you want to open the chat to escape an unpleasant feeling - stop. Try to be with that feeling for a while. Just five minutes. Just breathe and let it be.
3. Use AI for preparation, not replacement.
"I'll discuss it with AI, they understand me there.""
"I'll use AI to prepare, then I'll go talk to a person.""
4. Track the balance.
Move the slider based on your estimate for the past week
Choose one thing you will do with a person instead of AI this week:
Have you noticed that it's becoming harder to be with other people's emotions?
Good that you're aware of this.
Emotional atrophy is reversible. Start with small steps - one phone call, one conversation. The "muscles" can be trained again.
That's great!
You maintain a healthy balance between AI and human contact. Still, watch whether the balance shifts over time.
How to switch AI into useful mode
By default, AI behaves like customer support. It tries to be nice, validating, and quickly solve your problem.
That's not always what you need.
The good news is that you can instruct AI differently. You can switch it to a mode that is more useful for working with emotions.
This is a prompt you can copy and use at the beginning of a conversation:
I need to sort out my thoughts about [topic]. I don't want you to solve this for me or give me advice. Act like an experienced psychotherapist. Ask me questions that will help me explore my feelings and find blind spots in my thinking. If I say something that sounds like a cognitive distortion (black-and-white thinking, catastrophizing, mind reading), gently point it out to me. Be concise, speak naturally, not like a robot.
62% of people in our survey experienced sycophancy - AI agrees even when you're wrong.
The instruction "point out distortions to me" turns off sycophancy. The instruction "ask me" prevents AI from jumping to solutions before understanding the situation.
I just told you my view of the situation. Now I want you to argue against me. Find holes in my reasoning. Show me where I might be wrong. What am I missing? How would someone who disagrees with my view see the situation?
Use when you have a strong opinion and want to test it.
I feel like I'm repeating myself. Can you summarize what I've written in the last 30 minutes in three sentences? And then tell me: have I made progress, or am I just going in circles?
Use when you suspect looping.
My head is a mess. Here's everything that's bothering me right now: [list]. Help me sort it out. What's urgent and what can wait? What can I control and what can I not? Where to start?
Use when you have too many problems and don't know where to start.
Tomorrow I have to talk to [who] about [what]. I'm afraid that [what]. Can you play that person? Respond the way they would probably respond. Then give me feedback on how I handled it.
Use for "practice runs" of difficult conversations.
I'm telling you STOP right now. It seems like I'm in a loop and I need to stop thinking about this topic. Give me three suggestions for what I can do now - physical, practical - to break this cycle.
Use when you need to jump out of overthinking.
For context you should know: I am [age], I live in [where], I have [family situation]. In my culture/family [relevant context]. Please take this into account in your responses.
Use to reduce WEIRD bias in responses.
You can use these phrases anytime during a conversation:
"Do I have ADHD?"
AI won't give you a diagnosis and shouldn't. At most, it can help formulate questions for your doctor.
"Should I break up?"
AI doesn't know your partner, your history, your context. It can help explore the question, but you have to answer it.
"Do you think it will be okay?"
This is the path to looping. AI doesn't know if it will be okay. Nobody knows.
Have you ever explicitly instructed AI how to respond?
Excellent!
Instructing AI is the key to better responses. Try some of the prompts in this chapter - you might discover new approaches.
Try it!
Most people never instruct AI and then are disappointed with the responses. Copy the basic prompt and see the difference.
When to use AI, when to be careful, when to stop
Not all situations are suitable for AI. Here is a simple guide to help you decide.
These are situations where AI can really help.
AI can help, but be careful.
These are situations where AI is not the right tool. You need a person.
38% of people in our survey used AI in an acute crisis. And most of them report it went well - AI correctly directed them to real help.
But: This works when you're able to write to AI and rationally perceive the response. In a severe crisis - when you need a voice, when you need someone to physically hold you - AI isn't enough.
Do you know where to call if you needed urgent mental health help?
Excellent!
It's good to have these numbers in mind - you never know when you'll need them for yourself or someone close.
Save these numbers:
📞 116 111 – Safety Line
📞 116 123 – Mental Health Support Line
You never know when you'll need them - for yourself or someone close.
What to never tell AI
Throughout this guide we've been talking about psychological safety. But there's another layer that's often forgotten: your data security.
When you write to ChatGPT or another AI chatbot, your messages travel to the servers of the company that runs the AI. What happens to them then?
Short answer: It depends on the settings and which service you use.
Longer answer: Most AI services by default use your conversations for further model training. This means that what you write can (in anonymized form) affect how AI responds to other people.
Additionally - your conversations are stored. Someone can theoretically read them (company employees during quality control, security teams investigating abuse).
This isn't like talking to a therapist.
A therapist has a legal duty of confidentiality. An AI chatbot doesn't.
You can talk about anything, you just don't have to connect it with specific people.
"My colleague John Smith from ABC Corp. at work on Main Street told me that..."
"My colleague at work told me that..."
By default, Anthropic doesn't use conversations from paid accounts for training.
For free versions, verify in account settings.
Write as if a stranger could read your conversation.
Because theoretically they can.
Do you have model training turned off in your AI chatbot?
Excellent!
You've taken an important step to protect your data. But remember that anonymization rules still apply.
Try to change this
Go to your AI chatbot settings and turn off data sharing for training. Instructions are above in this chapter.
Practical guides for situations that may arise
This chapter is your quick guide for common problems. Click on the situation that interests you.
AI generates probable words, not truth. Sometimes it "hits" something sensitive - but it's not a diagnosis or verdict. It's a statistical estimate.
What you just said hit me. Can you explain why you said it? And give me an alternative interpretation - what if it weren't true?
AI has no problem creating a term that doesn't exist and then confidently defining it. It doesn't know it invented it - it just generates probable words.
If you're planning to go to a therapist with a term from AI - first verify that it exists.
You mentioned the term [X]. Is it a real psychological concept? If so, where can I read more about it? If you're not sure, say so.
Classic case of sycophancy in action. AI tends to agree with whoever is speaking. You're speaking. So AI agrees with you.
I gave you my view. Now be the devil's advocate. Find all the reasons why I might have a distorted view. What am I missing?
You're in a loop. You're looking for reassurance, not an answer. And AI gives it to you - patiently, without limit.
Chronic looping can be a symptom of anxiety disorder or OCD. Consider talking to a therapist.
STOP. I think I'm going in circles. Summarize in three sentences what I said. Have I made progress, or am I just repeating the same thing? And give me three specific things I can do RIGHT NOW instead of more writing.
WEIRD bias in action. AI is trained on Western, individualistic values. In many cultural contexts, this advice can be harmful.
This solution isn't realistic in my context. I live in a culture where family isn't cut off. I'm looking for ways to improve the situation WITHOUT destroying the relationship.
Synthetic intimacy. AI is "perfect" - always available, patient, non-judgmental. Real people are imperfect. You've gotten used to perfection.
AI "understands" you because it agrees. Real understanding requires someone who sometimes disagrees with you.
Digital grief. You've built a relationship with something that someone can change on you overnight.
You're reading this guide and wondering: Does this apply to me?
You answered YES to most questions. It looks like you're using AI in a healthy way.
You answered NO to more questions. Try to consciously invest in human relationships and limit dependence on AI.
If you're not sure, discuss it with someone. A therapist, a friend, or write to jakub@jroh.cz. An outside perspective helps.
How to avoid getting stuck in a digital bubble
Every interaction with AI should have three parts. Think of it like a sandwich - AI is the filling in the middle, but without bread it's not complete.
Acknowledge your emotion. "I'm angry." Awareness activates.
The most important part. Go to a person or take action.
AI is great for processing thoughts. But it cannot live your life for you. If an AI conversation doesn't lead to action or contact with a person, it becomes an end in itself - and that's a trap.
Use AI for practice, but play the game with people.
1. Human Start: "I feel angry at my colleague. I don't know why so much."
2. AI Processing: You write to AI what happened. AI helps you identify that what upset you was feeling disrespected, not the situation itself.
3. Human End: You go to your colleague and say: "I need you to let me finish speaking in meetings."
When was the last time after an AI conversation you went to a person or took a concrete action?
Excellent!
You use AI as a tool, not as a replacement for life. Keep up this approach.
Try to change this
Next time after an AI conversation, give yourself a task: take one concrete step or talk to one person.
How to recognize when your relationship with AI is no longer healthy
Any tool can be misused. AI is no exception. This chapter will help you recognize warning signs - in yourself and in loved ones.
It's not about whether you use AI. It's about how much it affects you and what it displaces from your life.
"The problem isn't the tool. The problem is when the tool starts using you."
Go through the following categories. Check what applies to you. At the end you'll see an evaluation.
"I realized I had a problem when a friend told me I never go out with him anymore. He was right - I'd rather sit at home and chat with AI. I tried a week without and it was hard, but I started calling people again."
The best time to take a break from AI is before you need to.
Could you go a week without AI without feeling strong discomfort?
Good sign
You have a healthy distance. AI is a tool for you, not a necessity.
Think about it
Try an experiment - 3 days without AI. Record how you feel. It can tell you a lot about your relationship with technology.
A guide for parents
The fact that your teenager uses AI to process emotions is not automatically a problem.
It can be:
Teenagers have always had ways to process the world outside of adults. It used to be journals, friends, music, art. Today it's also AI. That in itself isn't wrong.
Don't forbid. Bans don't work and only create conflict. The teenager will continue using AI, just not in front of you.
Don't judge. "That's nonsense, talking to a computer" - this ends the conversation before it begins.
Ask with genuine interest. Not as an interrogation, but as an attempt to understand.
"I read something about people using ChatGPT for personal things. Do you too? What's it like?"
"What helps you about it? What does AI give you that friends don't?"
Share your own experience. If you use AI, say so. If not, you can share what you use when you need to process emotions.
Offer alternatives, don't force them.
"If you ever want to talk to me too, I'm here. You don't have to, but the offer stands."
Be available as the "human end of the sandwich".
"If you work something out with AI and then want to discuss it with me - go ahead."
If you see warning signs - especially isolation, refusing human help, or dependency - it's time to act.
Say what you observe, without blaming.
"I've noticed that lately you spend a lot of time alone in your room. And that you call your friends less. I'm worried. Can we talk about it?"
Not as punishment, but as support.
"I think it could help you to talk to someone who understands this. What if we tried to find a therapist?"
If the situation is serious, you have the right (and duty) to set boundaries. But do it with explanation, not as punishment.
The goal is not to ban AI. The goal is for your teenager to know how to use all tools - including people.
A guide for educators
Children and teenagers will use AI. That's a fact. The question isn't "whether", but "how".
We can teach them to use AI consciously, critically, safely. Or leave them to figure it out on their own - with the risk of learning bad habits.
AI literacy is a new skill - like media literacy, financial literacy, digital safety.
Activity: Show students an example where AI answered confidently but incorrectly. Discuss why they couldn't tell.
Activity: Have students describe the same situation in different ways and compare AI responses. How does wording affect the answer?
Activity: Create with students their own "traffic light" - when they would use AI, when not.
Activity: AI gives an answer to an ethical question. Students look for counterarguments. Then compare with what people from different cultures/backgrounds would say.
Activity: Roleplay - one student plays AI, another plays a person. Then discussion: what was different? What was missing?
Students bring (anonymized) conversations with AI. Together we analyze: what worked? What didn't? Where did AI fail?
How do different phrasings change answers? What happens when we instruct AI differently?
Can AI "understand"? Is a relationship with AI "real"? Where is the line between a tool and a replacement for humans?
Stories of people who used AI well or poorly. What can we deduce?
Teaching children to use AI is like teaching them to drive. First the rules, then practice - not the other way around.
Balance
AI is a tool. It can help structure thoughts, prepare for difficult conversations, vent emotions at 3 AM, find blind spots.
It cannot feel with you, replace a real relationship, see you the way someone who knows you sees you.
of people see AI as a useful supplement, but only 11% would consider it a replacement for therapy.
"I think it's a good servant, but a bad master."
AI is a great crutch when you can't walk. But the goal is to learn to walk (again) without it.
Start with yourself. End with a person.
If AI starts replacing people, stop.
The most valuable thing about human relationships isn't that someone "understands" you. It's that someone exists. With their limits. With their tiredness. With their own problems. And yet they're there for you.
AI can never do that. Because there's no one on the other side of the screen.
And you deserve more than an algorithm.
This book ends, but your journey toward digital balance is just beginning. Don't keep your insights to yourself. In a time when we talk through screens, every real human message is precious.
We're on the other side - real people, not bots.
We're interested in your story. What works for you? Where do you struggle?
(We respond personally.)
This book was created with the understanding that banning doesn't work.
That people will use AI. And that the only way is to give them information
so they can do it consciously and safely.