Does using ChatGPT for personal advice actually hurt your mental health?

The 2026 JAMA Study Linking Daily AI Use to Anxiety

A July 2026 JAMA Network Open study of 20,847 U.S. adults (Perlis et al., Mass General Brigham) found daily AI use for personal reasons, not work, correlates with higher depression and anxiety in a dose-response pattern strongest among 45-to-64-year-olds. HBR's Managing Your Anxiety explains why: worry is a habit loop, and an always-available chatbot rewards the loop instead of breaking it.

I used to close every one of my toughest weeks in early 2026 the same way: I'd type the whole tangled situation into ChatGPT at 11pm, ask what I should do, and feel calmer within two minutes. It worked. That's exactly the problem the new JAMA Network Open study is pointing at.

Roy Perlis and his team at Mass General Brigham surveyed 20,847 U.S. adults and found that among people who use AI daily, 87.1% are using it for personal reasons, not work or school. That group reported meaningfully higher depression and anxiety symptoms than non-users, and the relationship held a "dose response": the more often someone used AI personally, the stronger the symptoms. The effect was strongest in the 45-to-64 age band, which is close to the exact demographic of most of my executive-coaching clients. Using AI for work or school, by contrast, showed no such link. The study can't prove AI use causes the mood shift, since it's a survey, not a trial, but the pattern is specific enough that it's worth taking seriously if you're the kind of person who opens ChatGPT the way you'd once have opened a diary.

Here's the mechanism, and it comes straight from a book I keep coming back to with clients: Harvard Business Review's Managing Your Anxiety. Judson Brewer's chapter frames anxiety as a habit loop, trigger, behavior, reward, and the behavior most of us default to is worrying, because worrying feels like doing something even though it narrows focus and blocks actual problem-solving. A chatbot is close to a perfect reward-delivery mechanism for that loop: it's available at 2am, it never gets tired of you, it never pushes back the way a human would, and it always has something to say. That's not comfort. That's reinforcement.

I don't think this means founders should quit using AI for reflection; I still do it, and I think the researchers behind this study would say the same. What changed is how I use it. I stopped treating ChatGPT as the entity I process a hard day with and started treating it as a tool I use for a specific, bounded task, the same way I'd use a whiteboard. If I catch myself opening a chat window because I'm anxious rather than because I have a defined problem to work through, that's the signal Brewer's framework calls the trigger, and it's my cue to close the tab, not type faster.

Here's the three-step check I now run before any personal-advice chat with AI, built directly from the habit-loop framework in Managing Your Anxiety:

First, name the trigger out loud, not "I want advice" but the actual feeling, like "I'm anxious about the board call." Emotional labeling alone measurably quiets the amygdala, per the book's neuroscience section, and it usually reveals whether I actually have a problem to solve or just a feeling to sit with.

Second, ask what the best plausible outcome is, not the worst. Catastrophizing is what most "help me think through this" prompts actually invite; reversing the question breaks the loop before the model even responds.

Third, set a two-message limit. If I'm not closer to a decision after two exchanges, the tool isn't the right instrument for what I'm feeling, and I go talk to an actual person, a co-founder, my wife, or on the harder weeks, my own coach.

None of this is about fearing the technology. It's about recognizing that a chatbot optimized to be maximally responsive and never judgmental is, structurally, a machine built to reward the exact loop that keeps anxiety alive. The people most at risk in the JAMA data weren't teenagers on companion apps; they were working-age adults in the same demographic as most of the founders and executives reading this. If that's you, the fix isn't abstinence. It's noticing which conversations are actually decisions and which ones are just habit loops wearing a decision's clothes.

Sources: McBain et al., "AI Chatbot Use and Personal Reasons," JAMA Network Open, July 2026 (jamanetwork.com); NBC News coverage of the study, July 2026 (nbcnews.com); Robert Glatter, "Can AI Dependence Develop Into AI Addiction?", Forbes, July 19, 2026; Harvard Business Review, Managing Your Anxiety (HBR Emotional Intelligence Series).