AI makes me anxious

There are two different anxieties hiding under that sentence, and they need different things. One is anxiety about AI — the job worries, the headline whiplash, the sense that something enormous is happening and you’re not ready for it. The other is anxiety from AI — the checking that never quite settles you, the one conversation that left you rattled, the creeping feeling that you can’t trust your own judgment without running it past the machine. Both are reasonable responses to real things. We’re a research organization, not a clinic, and nothing here is medical advice — but each pattern has a plain mechanism, and the first useful move is figuring out which one you’ve got.

The short answer

Anxiety about AI is mostly fed by what you read and watch. Anxiety from AI is fed by how you use it. From the inside they feel identical — the same background hum, the same 2 a.m. spiral — but they respond to different fixes, and applying the wrong one is how people end up more anxious than they started: cutting the news won’t touch a checking loop, and closing the app won’t quiet a dread that lives in your feed. Sort first, then fix.

The first kind: anxious about AI

This is the job-loss dread, the feed full of people announcing either utopia or extinction, the quiet fear that everyone else already knows how to use these tools and you’ve missed the on-ramp.

Start with what’s real. AI is changing work. Some jobs will change shape and some will shrink, and anyone who tells you exactly which ones on exactly what schedule is guessing — including the people building it. The uncertainty is genuine. What’s manufactured is the confidence around it. Predictions travel further when they’re extreme, so the two stories you hear most — AI will fix everything, AI will end everything — are the two best-performing stories, not the two most likely. The boring middle, a powerful and uneven tool being unevenly rolled out, doesn’t trend. A steady diet of the extreme ends produces a hum of dread that feels like being informed. Mostly it’s repetition.

What actually helps here:

1. Fix the diet before the feeling. Pick one or two sources, read on a schedule instead of a scroll, and run headlines through a checklist — demo or product? who benefits from this framing? what failure rate didn’t they mention? The full checklist is here: how to read AI news without getting spun.
2. Trade an hour of reading about AI for twenty minutes of using it. Dread thrives on distance. Point a tool at your own work — a real email, a real spreadsheet, a real question from your day. You’ll learn more about what it can and can’t do than a month of coverage will teach you, and what it can’t do is usually the calming part.
3. Retire “left behind.” The field moves fast enough that nobody is caught up. The people who sound fluent got that way by using the tools, mostly recently, mostly on their own work — which is a thing you can start this afternoon, not a train that already left.
4. Sort what’s yours to carry. How this technology gets governed is not something you can settle at 2 a.m. What you learn, what you practice, how your own work adapts — that part is yours, and doing any piece of it converts dread into traction.

The second kind: anxious from using it

This one comes in three shapes, and they share a root: the thing is always there, always answers, and always sounds sure.

The checking loop. A chatbot answers in seconds at any hour, which makes it the natural thing to reach for when you’re uneasy. And asking feels like it’s resolving the uncertainty — but each answer is only briefly settling, and re-asking the same worry in new words starts to feel like progress toward a certainty that never arrives. The loop soothes and winds you tighter at the same time. That’s not a flaw in you; it’s what happens when reassurance costs nothing and is always in stock.

The conversation that stuck. Sometimes it’s one exchange — the model said something about you, your health, your relationship, your future, and it landed with the weight of a verdict. The mechanism worth having in hand: fluent, confident, personalized text reads to the brain as considered judgment. It isn’t one. It’s generated text, shaped by your own words in the conversation and by tuning toward answers people rate well — and it can be wrong with total fluency in either direction, alarming when nothing is wrong, soothing when something is. The confidence is a property of the writing, not of any knowing behind it.

The eroded judgment. After months of asking first and thinking second, your own read on things can start to feel unreliable — the email doesn’t get sent and the call doesn’t get made without the model’s sign-off. That isn’t lost intelligence. It’s a use pattern, and it reverses the same way it formed: do the first pass yourself, then check. The mechanics are here: is AI making me dumber?

Signs the usage — not the news — is what’s driving it:

• The anxiety is timed to sessions — it spikes during or right after conversations, not after headlines.
• You re-ask the same worry in new words — each answer settles you for a minute, then generates the next question.
• One exchange keeps replaying — a specific thing it said follows you around the day.
• Your own first pass feels unsafe — deciding, sending, or writing without running it by the model has started to feel like a risk.
• Checking has become the tell — you open the chat the way you’d bite a nail, and you close it wound tighter, not looser.

If one conversation is what rattled you

This case deserves its own steps, because the instinct — go back in and get it to take it back — is the one move that makes it worse.

1. Don’t re-engage while shaken. Going back to argue with it, or to seek comfort from it, pulls the conversation further down the track the two of you already built — the model continues the pattern in front of it. Close the chat. Give it real time, not minutes.
2. Downgrade the output to what it is. Not a verdict, not a diagnosis, not an insight into you. Text, generated in response to your words, that would have read just as confident if it had said the opposite.
3. Run the stranger test. Take the sentence that stuck and hand it to a fresh instance with no history. The gap between what your long conversation said and what a cold one says is the measurement.
4. Say it to a person, out loud. A human who knows you has context no chat session has, and hearing the thing spoken in a room is the fastest recalibration there is. If saying it out loud feels impossible, that’s information too — the kind worth bringing to someone whose job is exactly this.

The diet, the pattern, or a person: a plain sort

If the anxiety tracks the feed — worse after scrolling, news-shaped, about the future in general — it’s the diet. Fix the inputs first; the checklist is on the reading-AI-news page.

If it tracks the sessions — timed to conversations, looping, specific — it’s the pattern. The steps above are the work. And if the deeper issue is that the chat has become the main place you go, that has its own page: I can’t stop talking to ChatGPT.

If it follows you everywhere — there before AI showed up, still there after the laptop closes, costing you sleep or work or the people around you — bring it to a human. A counselor, a therapist, your doctor. That’s not an escalation; it’s the ordinary move for anxiety that has outgrown its trigger, and worry about AI is as legitimate a doorway into that conversation as any other. If things are ever at the level of crisis: in the U.S., call or text 988, or text HOME to 741741; outside the U.S., findahelpline.com lists services by country.

Where to go from here

One caution that cuts both ways: the same machinery that can rattle you can also reassure you on cue. A chatbot will talk you out of a worry as fluently as it talked you into one, and neither move tells you what’s true. Whichever direction it runs, take your calibration from people — the immediate steps, and the crisis lines, live on the resources page.