Gen Z Uses AI More Than Anyone. They Trust It Less Every Year.

Ask a nineteen-year-old what they think about AI and you will probably hear something sharp. Ask what they did in the last three hours and the answer is a feed an algorithm built for them, a run of videos a model decided they would want, and maybe a chatbot that helped with an assignment. None of that gets counted as using AI.

The generation with the most exposure to this technology is the one turning against it fastest. That is not a contradiction. It is the most useful signal we have right now, and almost nobody is reading it correctly.

Four questions are worth answering here, in order. Is the reaction real. Is it justified. If it is, who caused it. And is it hypocritical, given how much they use the thing.

Is it real?

It is real, it is measurable, and it is moving quickly.

Gallup, working with the Walton Family Foundation and GSV Ventures, surveyed 1,572 people ages fourteen to twenty-nine in early 2026. Just over half use AI weekly or more. That part surprises nobody.

The direction is what matters. In a single year, excitement about AI fell fourteen points. Hopefulness fell nine. Anger climbed nine points, to thirty-one percent, and is now more common than excitement. On trust, sixty-nine percent said they trust work done by a human alone. Twenty-eight percent said the same about work AI had touched.

But calling this fear gets it wrong, and the survey itself says so. The most common thing Gen Z reported feeling about AI was not anger, and it was not anxiety. It was curiosity, at forty-nine percent, higher than anything else measured.

So this is not a generation running from a technology. It is a generation leaning in, paying close attention, and thinking less of it every year. That is a different thing than fear and a more serious one. Fear is what people feel about things they do not understand. This is what they feel about something they use daily.

Then there is the detail that should stop you. Among people using AI every day, the decline was steeper. Excitement down eighteen points, hopefulness down eleven. The heaviest users are souring fastest.

You could argue the pool changed. In 2025 the daily users were volunteers. By 2026 they include everyone whose school or job made it mandatory, and Gallup's own split hints at exactly that: fifty-six percent of Gen Z still in K-12 use AI weekly, against forty-eight percent of Gen Z adults. The people with the least ability to decline use it the most.

That is probably part of it. But it is not a defense. It just moves the complaint. They may not be souring on the technology so much as on having it handed to them without being asked.

Is there a good reason for it?

Two fears are in play, and they are not the same. One gets all the attention. The other has more evidence behind it.

Start with the loud one, the idea that AI is about to become something nobody can control.

In 2025 a small team led by Daniel Kokotajlo, a former OpenAI researcher who refused to sign an agreement that would have kept him quiet even though it put nearly two million dollars of his vested equity at risk, published a detailed scenario called AI 2027. It made dated, specific, gradeable predictions. That alone made it unusual. Most people making claims this size are careful not to be checkable.

A year on, the scorecard is mixed in a way that is worth understanding. The unglamorous predictions largely landed. Coding tools got genuinely capable on roughly the expected schedule, the enormous infrastructure buildout happened, agents arrived. Some risks showed up early, including AI finding software vulnerabilities on its own, about a year ahead of prediction. What has not appeared is the engine of the entire scenario, AI accelerating its own research into a runaway loop. Independent observers grading the forecast publicly put overall progress at roughly two-thirds of the predicted pace.

The modest predictions held up. The dramatic one did not, at least not on that timeline.

Now the quieter fear, the one Gen Z actually voices and almost nobody covers: that using this thing is making them worse at thinking.

That one has more behind it.

Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about 936 real tasks and found something specific. The more confidence a person had in the AI, the less critical thinking they did. The more confidence they had in themselves, the more they did. When people failed to check the output, the reasons were mundane: it did not occur to them, they had no time, or they could not have evaluated it anyway.

A separate 2025 study of 666 people found frequent AI use correlated with weaker critical thinking, with the effect concentrated in the youngest participants. And a 2026 study of Gen Z students found the sharpest version of it. The standard chatbot produced the highest confidence in what students believed they had learned and the lowest actual learning. Effort, confidence, and results came apart from each other.

Their instinct is pointing at something real.

Here is the part that keeps this honest, because it cuts the other way.

The single most-shared study on this subject, an MIT experiment that put EEG caps on people writing essays with and without ChatGPT, is a preprint that still has not cleared peer review. It ran fifty-four participants. It has a formal published critique questioning its methods. And its own authors wrote plainly that they did not find AI made anyone stupid or caused brain rot, which is almost exclusively what it gets cited for.

We have also run this play before. In 2011 a paper in Science found that people remember where to find information rather than the information itself. The Google effect was everywhere for a decade. The most-cited finding in it later failed replication twice, and a 2024 review concluded the effect is real but smaller and more situational than the original implied.

So the direction is probably right, the magnitude is unproven, and the loudest claims are the least supported.

Which is the same answer twice, arrived at from two fields that have nothing to do with each other. In both, the boring claim survives contact with evidence and the alarming one does not. And in both, the alarming one is the version that reaches them.

Sit with that for a second. Gen Z is being handed the least supported version of a concern that has a well supported version underneath it. That is an efficient way to make someone anxious and leave them unable to explain why.

Who is causing it?

The reflex is to blame the AI companies. It is worth being more precise, because the responsibility splits three ways and they are not equally guilty of the same thing.

The catastrophe narrative comes largely from inside the labs. Not from critics outside them. The people building these systems have spent years making public statements about existential risk, and AI 2027 itself was written by former frontier-lab staff. Whatever you make of the substance, the apocalyptic register of this conversation did not come from teenagers and it did not come from nowhere.

The pressure on their prospects comes from employers and schools. Somebody decided this was mandatory. Look again at who uses it most and it is the students, the group least able to say no.

And the daily grind, the part that wears on them hour after hour, comes from a third group entirely. Not the labs that built the models. Not the institutions that required them. The companies that take that capability and wrap it in something engineered to hold attention. The feed that decides what plays next. The chatbot with a streak counter attached. The assistant nobody asked for, bolted onto an app they already had.

That distinction matters, and there is a familiar way to think about it.

A car company that builds a sound car is not responsible when somebody drives it drunk into a fence. The car did what it was designed to do. The failure was a decision made by the person behind the wheel, and courts have drawn that line for a long time.

But the analogy has a second half that usually gets skipped. We did not make driving safer by lecturing drivers. We made it safer by regulating the design. Seatbelts, crumple zones, airbags, ignition interlocks. The driver stayed responsible for the choice, and the machine got safer anyway.

Both halves apply. The model is the car. The company that tuned a product to maximize the hours a teenager spends inside it made a business decision, and that decision is the drunk driver. And the question we eventually asked about cars is the one almost nobody is seriously asking about attention-optimized software: what should this be required to be built like?

I spend my working life on cases where somebody trusted a system that was designed to be trusted and had no practical way to check it. That is the Microsoft finding almost exactly. Not stupidity. A confident-sounding output, no easy way to verify it, and no time to try. In fraud work, the confidence of the presentation is part of the attack surface. The difference between that and an engagement loop is not the mechanism. It is the legality.

Is their reaction hypocritical?

This is the objection used to dismiss the whole thing. If they hate it so much, why are they on it constantly?

Because using something is not the same as endorsing it, and treating those as identical is how a person avoids having to listen.

Most of the exposure was never chosen in the first place. Gen Z spends three to four hours a day on social media, more than double what Baby Boomers report, and every minute of it runs through a recommendation system deciding what comes next. That is AI. Nobody opened an app and consented to it. Pew Research Center found separately that sixty-four percent of teens have used a chatbot, about three in ten daily, mostly for homework and looking things up rather than the companionship panic the headlines prefer.

Add those together and their real exposure is almost certainly higher than any survey can capture, because most of it was never labeled as the thing they are being asked to have an opinion about.

Then there is the matter of choice. You cannot opt out of the job market, and a sixteen-year-old cannot opt out of the assignment. Calling that hypocrisy is like calling someone a hypocrite for driving to a job they do not like.

And the strongest evidence that this is not incoherence: the souring is worst among the heaviest users. If they had never touched the stuff and hated it anyway, that would be the position worth questioning. Informed distrust is the opposite of a contradiction. It is what paying attention looks like.

One real tension is worth naming instead of smoothing over. Just over half use AI weekly, but only twenty-eight percent trust work that AI has touched. Some of them do not fully trust their own output. That is not hypocrisy either. That is what it looks like to use a tool you have not been given a good reason to believe in, because the alternative is falling behind.

What to do with this

If you employ young people, teach them, or are raising one, the temptation is to treat this as an opinion you need to form about AI. It is not. It is an incentives problem, and incentives are visible if you look at them.

Stop asking whether AI is good or bad. Ask what the specific product in front of you was built to take. Attention, hours, engagement, retention. That answer is usually sitting in plain view in the business model, and it is a question with an actionable answer, unlike the one everyone is currently arguing about.

Then notice what this generation is actually telling you. They are not anti-technology. They are the most curious group in the survey and among the heaviest users in the country. They simply do not trust what is being built with it, and they have been handed specific, checkable reasons not to.

They are not confused. They are early.