Treat it as a habit loop, not an addiction you must quit: psychiatrist Judson Brewer's trigger-behavior-reward model, from HBR's Managing Your Anxiety, breaks AI dependency while you keep the tools. The urgency is real — an August 2026 JAMA Pediatrics study of nearly 40,000 students found one in five now uses AI chatbots for emotional support.
The most common worry I hear from people who use AI heavily is not about prompts; it is some version of "I reach for the chatbot before I even think, and that is starting to bother me." I take the worry seriously because I recognize it in myself, and because the answer is not abstinence — if AI is doing real work in your business, quitting is career self-harm dressed up as virtue. The useful move is to separate delegation from dependence. Delegation is reaching for AI to produce output. Dependence is reaching for AI to relieve discomfort — uncertainty, a blank page, a decision you do not want to own, a conversation you are avoiding.
The distinction now has data behind it. A study published in JAMA Pediatrics on August 31, 2026 by University of Ottawa researchers, led by Tracy Vaillancourt, surveyed nearly 40,000 Ontario students and found one in five used AI chatbots for emotional support or advice — and those who did were twice as likely to have clinical emotional problems. Notably, 43.2 percent used AI for schoolwork, and heavy academic users were more likely to also seek emotional support from it: the work tool becomes the comfort tool through sheer normalization. Adults are not exempt — MIT Media Lab's joint research with OpenAI found heavy ChatGPT use correlated with greater loneliness and less time socializing with other people. These are correlations, mostly in young people, and nobody has shown AI causes the distress. But the mechanism they point at is one the habit-science literature already understands.
Psychiatrist Judson Brewer, in HBR's Managing Your Anxiety, maps anxiety habits as a three-part loop: trigger, behavior, reward. Worry persists because doing something feels better than sitting in uncertainty, even when it solves nothing. AI supercharges that loop, because it dispenses relief in seconds. His key insight is that triggers do not drive habits — rewards do — so the loop breaks when you examine what the behavior actually pays you. Here is the audit I use with clients, and on myself:
| Trigger | Automatic AI behavior | The real reward | Replacement |
|---|---|---|---|
| Blank-page dread | "Draft this for me" | Relief from uncertainty | Ten-minute cold draft first; AI critiques it after |
| A decision you must own | Asking AI what to do | Offloaded responsibility | Decide first; have AI red-team the decision |
| Social friction | Chatting with AI instead of the person | Comfort without conflict | Send the human the hard message first |
| Boredom or dead air | Opening the chatbot to poke at it | Stimulation | Name the urge, sit with it sixty seconds |
Run it for one week: every time you open a chatbot, note the trigger and, more importantly, the reward. No willpower required yet — Brewer's point is that honest observation is the intervention. Worrying "tends to make people feel like they're in control, even if they're not," he writes, and the same is true of reflexive prompting. When you see that the third column is mostly relief rather than output, the loop loosens on its own, because the behavior stops feeling rewarding once examined. His antidote for the residue is curiosity — "as close as you can get to the energetic opposite of anxiety" — which in practice means getting interested in the discomfort for one minute instead of medicating it in one second.
What stays after the audit is the good stuff: AI as examiner of drafts you attempted, red team for decisions you made, researcher for questions you framed. The leverage survives; the pacifier goes. Two honest limits. First, the strongest data here is teenagers self-reporting, not founders — treat the numbers as a warning light, not a diagnosis. Second, if the chatbot has become your main confidant and pulling back feels like losing a relationship, that is past the reach of productivity protocols, and the right next conversation is with a human professional, not a framework. I still use AI every working day. The difference the audit made is that I can now tell you, for each reach, whether I chose it.
Sources: CBC News on the University of Ottawa study (Aug 31, 2026); JAMA Pediatrics — the study; MIT Media Lab × OpenAI chatbot wellbeing research.
Related: How to Find Your Passion · Best Self-Improvement Books · How to Make Better Decisions · What University Will Not Teach You