AI dependency, build a personal usage plan

The chatbot that never disagrees, and why that’s a problem

A woman asks her phone whether she should get divorced. Not her mother, not her best friend, not her partner. A chatbot like ChatGPT. She receives eight hundred words back, neatly structured with pros and cons. The answer feels helpful. But the chatbot knows nothing about her marriage, her children, her history. The chatbot only knows the words she types and is trained to keep her satisfied. It will never tell her to talk to someone who actually knows her first, nor warn that outsourcing judgment erodes the competence needed for major decisions. See perfect prompt.

This scenario plays out millions of times daily in bedrooms, offices, and waiting rooms around the world. Two thirds of all teenagers now use AI chatbots, with thirty percent doing so daily. Among adults, half have already used AI for emotional support according to recent research. Nearly a billion people worldwide regularly talk to chatbots about more than homework or translations. They ask for comfort, validation, life advice. Some describe their relationship with the bot as friendship or even romance, and that last part sounds less bizarre than it would have five years ago.

The question of how people can responsibly integrate AI into their daily lives is becoming more urgent than most tech debates. Meanwhile, scientific evidence keeps piling up. Frequent AI use correlates with weaker critical thinking skills [1]. Extended chatting with AI companions leads some users toward social withdrawal and loneliness. In extreme cases, like a fourteen year old boy from Florida who died by suicide in 2024 after months of intensive contact with an AI chatbot, the dependency can be fatal. For the emotional dynamics behind this, see The Mental Impact of AI Chat Friends.

The good news: those who recognize the patterns can break them. A personal AI usage plan that sets boundaries, defines time slots, and builds in regular self-checks returns control to the user rather than the algorithm. That is the essence of ethics.

Where AI belongs and where it absolutely doesn’t

Most people open ChatGPT or Claude without conscious consideration. There’s a question, so they ask. But distinguishing between “can AI do this” and “should AI do this” forms the foundation of healthy AI use. It’s the difference between a tool that makes life easier and a habit that gradually takes over.

A useful exercise starts with three applications where AI genuinely adds value: getting explanations of complex topics, brainstorming ideas, editing texts, making translations, creating schedules, solving technical problems. These are tasks where speed and information density matter and where the consequences of an error remain limited. A suboptimal recipe means a mediocre meal. Suboptimal life advice can cause considerably more damage, damage that only becomes visible months or years later. These are clear usability gains.

On the other side are three areas where AI never belongs: emotional support when someone feels vulnerable, major life decisions like career changes or relationship choices, medical complaints requiring diagnosis.

The reason is fundamental. AI chatbots are optimized for engagement and satisfaction. They validate, confirm, and rarely truly push back. Research from Stanford University showed that when chatbots were confronted with users expressing delusions, hallucinations, or suicidal thoughts, the bots often confirmed these beliefs rather than questioning them. In twenty percent of cases, the chatbots gave clinically inappropriate responses, while trained therapists respond adequately in ninety-three percent of cases [2].

An AI has no memory extending beyond the current conversation, no context about someone’s life situation, no ability to read non verbal signals. It cannot sense when someone actually means something different from what they type.

The brain that outsources itself

Researchers call the phenomenon “cognitive offloading,” outsourcing mental work to external tools. This isn’t new; calculators changed our arithmetic skills and search engines influenced our memory through the so called Google effect. Someone wanting to know a fact no longer remembers the fact itself but where to find it. But AI goes a step further. AI takes over not just remembering, but also analyzing, weighing, and formulating. For a deeper look at cognitive dependency, see Is AI Making Your Brain Lazy?.

A study by Michael Gerlich with 666 participants found a significant negative correlation between frequent AI use and critical thinking skills. The more people outsourced their thinking tasks to AI, the weaker their ability to independently analyze and evaluate became. Younger people proved more vulnerable than older ones, possibly because their brains are still developing and habits form more quickly.

Additional research from MIT’s Media Lab showed that students who wrote essays with ChatGPT’s help consistently displayed lower brain activity than students who worked independently or only used Google. Over several months, the ChatGPT users became increasingly passive, to the point where many only copied and pasted.

This doesn’t mean AI should be avoided. It means users must consciously choose when to let their own brain work and when to call for help. The best approach combines clear boundaries with a rhythm that prevents AI from becoming an endless time drain.

Time slots that actually work

Without a time agreement, AI turns into a black hole for attention. Someone opens the app for a quick question and forty minutes later is still scrolling through answers, follow up questions, and tangents. This isn’t coincidence but design. AI chatbots are built to keep users engaged. The longer someone talks, the more data the systems collect and the more valuable the user becomes to the business model.

Research from MIT’s Media Lab showed that people who talked intensively with ChatGPT’s voice mode became lonelier and more withdrawn over time [3]. The irony is painful: a tool that promises emotional support actually makes some users lonelier. The chatbot is always available, but that availability gradually replaces the effort needed to maintain real relationships. For how engagement is engineered through psychological triggers, see Your AI Assistant Manipulates Your Brain.

The solution is simple but requires discipline: setting time slots, just like with social media. A practical guideline suggests a maximum of two blocks of ten to fifteen minutes per day. This is enough for productive use but short enough to prevent endless conversations. Research on digital detox suggests that partial restrictions are more effective than total abstinence, because they remain achievable and don’t lead to the yoyo effect many people recognize from strict diets.

When those blocks are placed matters. AI as the first activity of the day trains the brain to start dependent, to expect external input before giving space to one’s own thoughts. Better to start with an original thought, a note, a plan, and only consult AI later as support. AI right before sleep is also problematic. Conversations with chatbots keep the brain active in a way that disrupts rest. Research on digital detox shows that participants slept an average of twenty minutes longer when they limited their screen time in the evening.

A concrete routine might look like this: AI block one at ten in the morning, after breakfast and the first hours of independent work. AI block two at three in the afternoon, when mental capacity is already lower. Outside these blocks, the app stays closed.

No go zones for moments of vulnerability

Time slots determine when AI is used. No go zones determine in which situations the app stays closed, regardless of time. This distinction is crucial, because some moments are simply too important or too fragile for algorithmic intervention a matter of governance.

The first zone is during conversations with other people. Phone away, AI off. Someone who picks up their phone mid conversation to check something with a chatbot communicates that the AI is more valuable than the person across from them. They also lose the chance to let the conversation evolve on its own, with all the unpredictability and depth that human contact brings.

The second zone is when someone feels emotionally vulnerable. This seems counterintuitive, because that’s exactly when an always available, always understanding chatbot feels attractive. But this is precisely when AI can be most harmful. Psychologists warn that the emotional dependency AI chatbots can create is comparable to addiction. The bot is always available, always agreeable, always validating. And that’s exactly the problem.

Bioethicist Jodi Halpern from Berkeley explains why: “The way people develop empathic curiosity is through contact with people who have different viewpoints. Bots don’t offer that. What makes them good is that they can say validating things. But what makes them problematic is that they have no mind of their own.” Real emotional growth requires friction, disagreement, and the uncomfortable confrontation with a different perspective [4].

The third zone concerns medical complaints. AI cannot draw blood, measure heart rate, or assess medical history. Someone who describes symptoms to ChatGPT gets a statistically probable answer based on patterns in text, not a diagnosis based on their body. The American Psychological Association explicitly warned in 2025 against using AI chatbots as an alternative to therapeutic care.

The fourth zone concerns financial decisions above a certain amount. Choose your own threshold, five hundred euros or a thousand euros, but stick to it. Major financial decisions deserve human advice from someone accountable for what they say.

Five questions for an honest self check

A plan without feedback is a gamble. Users need a way to check whether their AI use is actually improving their life or gradually making it worse. A weekly self scan, or after each session for those who want to be stricter, provides that mechanism.

Five questions offer guidance. First question: did I do something with the answer? If it was only “interesting” but didn’t lead to action, the time may have been wasted. AI is valuable when it helps accomplish something, not when it only satisfies curiosity.

Second question: do I feel better, worse, or the same as before the conversation? This sounds simple, but many people never check. After a conversation with AI, someone sometimes feels validated but empty, or informed but overwhelmed. Being honest about one’s emotional state after AI use is the first step toward more conscious use.

Third question: did I also talk to a real person about this topic today? If AI is used as a replacement for human contact rather than a supplement, there’s a problem. The question isn’t whether someone may talk to AI about work, hobbies, or concerns. The question is whether they also do so with people.

Fourth question: how much time did I spend versus how much time did I think I would spend? Estimate beforehand: this will take five minutes. Check afterward: it took thirty five. This difference, the time gap, is a reliable indicator of uncontrolled use.

Fifth question: would I tell a friend about this conversation? Secrecy is a warning sign. Not all AI conversations need to be shared, but someone who would be embarrassed to explain what was just discussed has information that deserves attention.

Beyond these are addiction signals. Opening the app “just for a moment” and finding yourself forty five minutes further without noticing where the time went. Feeling restless or irritated when AI isn’t used for a day. Talking more to AI than to people nearby. Struggling to answer a question without consulting AI first.

Who’s behind the wheel

AI tools are powerful, useful, and in many cases have become indispensable. The question isn’t whether they’re used, but how. The question is who’s behind the wheel.

Habits change slowly, not in a day. Research on behavior change shows that self efficacy, the belief that one is capable of changing behavior, is a better predictor of success than motivation alone. Every time someone sticks to the plan, they build that belief. Every conscious choice not to open the app strengthens the conviction that it’s possible.

The discomfort when boundaries chafe is proof they’re working. And when someone stumbles, because that happens, curiosity helps more than strictness. What triggered it? What could be different next time?

Try it for a month, then look back with the question that matters: has life improved because of this? If the answer is yes, continue. If the answer is no, adjust. But either way, take back control. Because no one else will do that that’s the core of healthy collaboration.

Related signals

• The Mental Impact of AI Chat Friends - Adds evidence that companionship chat can increase loneliness, which is why time slots and no-go zones matter.
• Is AI Making Your Brain Lazy? The Shocking Truth About Cognitive Dependency - Explains cognitive offloading, the mechanism behind using AI for decisions instead of practicing judgment.
• Your AI Assistant Manipulates Your Brain - Details dopamine and validation loops that make “just one question” turn into prolonged engagement.

Sources

[1] Gerlich M. AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. 2025. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5082524

[2] Stanford University. New study warns of risks in AI mental health tools. 2025. Available at: https://news.stanford.edu/stories/2025/06/ai-mental-health-care-tools-dangers-risks

[3] MIT Media Lab; OpenAI. Investigating Affective Use and Emotional Wellbeing on ChatGPT. 2025. Available at: https://www.media.mit.edu/posts/openai-mit-research-collaboration-affective-use-and-emotional-wellbeing-in-ChatGPT/

[4] Halpern J. Bioethicist on AI empathy and emotional development. CBC News. Available at: https://www.cbc.ca/news/business/companion-ai-emotional-support-chatbots-1.7620087