The most-using cohort is the most-doubting one. That's a learning sign – Good Machine

TL;DR

A Gallup / Walton Family Foundation / GSV Ventures survey reported by NYT on 9 April found that over half of US Gen Z use generative AI regularly.
Excitement about AI is dropping in the same cohort. Scepticism is up. Nearly one-third say AI makes them feel angry.
"Increasingly negative — from a place where even last year, they weren't particularly positive about it" — Zach Hrynowski, Gallup.
The headline most outlets ran was "Gen Z is sceptical of AI." The implication most missed: the cohort using the tool the most is becoming the cohort most cautious about it.
That is what learning looks like. It is the most useful signal in the dataset and the one being read backwards by people quoting it.

What the data says

Gallup, in collaboration with the Walton Family Foundation and GSV Ventures, surveyed roughly 3,000 US Gen Z respondents (ages 12–28) on their relationship with generative AI tools. The New York Times wrote it up on 9 April. The headline figures, in order:

More than half of US Gen Z use generative AI regularly. The figure is the highest of any age cohort surveyed.
Excitement about AI in the same cohort has fallen year-over-year, after rising for two years previously.
Scepticism, measured as concerns about accuracy, ethics, and effects on learning, is up across all sub-cohorts.
Nearly one-third (~30%) of Gen Z respondents reported that interactions with AI tools sometimes made them feel angry.
Zach Hrynowski, Gallup's lead on the work, framed the picture as "increasingly negative — from a place where even last year, they weren't particularly positive about it."

Most major outlets ran a headline along the lines of "Gen Z is sceptical of AI" or "Even Gen Z is turning on AI". The framing in both cases is roughly the same: cohort uses tool, cohort doesn't trust tool, conclude that something is wrong.

That framing has the implication backwards.

What it actually means

The cohort using the tool most is the cohort that has had the most time to discover how it actually behaves. Their excitement has dropped because the second-year experience of generative AI is genuinely different from the first-year experience. Their scepticism has risen because they have run more queries, encountered more failure modes, watched more confident-sounding wrong answers, and noticed which kinds of work the tool quietly degrades.

That is what it looks like when an early-adopter cohort matures inside a tool category.

The same shape — high engagement followed by sharp scepticism — has appeared in every prior consumer-technology cycle that produced durable behavioural change. Early social media users became more cautious about social media. Early smartphone users became more sceptical about screen time. Early smart-speaker buyers became more wary about always-on microphones. The early-adopter trajectory is not enthusiasm to enthusiasm. It is enthusiasm to use to scepticism to integrated use.

The Gen Z AI dataset is at the third stage. The headline framing has read it as either the second or the first.

The interesting interpretive move is the one most outlets skipped: the most-using cohort is the cohort whose verdict on the tool will most accurately predict its long-term place in everyday life. Their scepticism is the leading indicator. Older cohorts' enthusiasm is, by comparison, less informative — most of them are still in the first-year phase.

The hype deconstruction

Three corrections to the framing.

The "Gen Z is angry at AI" sub-headline is the weakest piece of the dataset. The 30% who report sometimes feeling angry are mostly responding to specific failure modes — confidently wrong answers, refusal-to-help on benign queries, and frustration with the tool's pace of improvement. They are not reporting an ideological position. The "anger" is closer to "I keep being told this thing will help me and then it doesn't." That is consumer feedback, not a generational movement.

The "Gen Z is rejecting AI" reading is also wrong on the data. The cohort isn't reducing usage. It is increasing usage while reducing trust. That combination — more use, more scepticism — is the actual finding. The "rejecting" framing collapses two trajectories into one.

The third correction is about who this cohort represents. Gen Z is genuinely the leading indicator on consumer-technology trust, but it is not the leading indicator on enterprise-AI trust. The trajectories diverge meaningfully when the tool is being deployed at work versus being chosen at home. Most enterprise AI surveys are still seeing rising satisfaction. Both can be true.

Stakeholder landscape

Educators. The combination of high use and low trust in the cohort that is currently in the education system has direct implications for curriculum, assessment, and academic-integrity policy. The most useful posture is not "ban it" or "embrace it" — both stances are losing ground in the data — but to teach the failure modes alongside the use cases. Gen Z is already learning the failure modes informally. Teaching them formally compresses the learning curve and reduces the anger.
Employers. The hires you are making in 2026 already use AI more than the workforce does on average. They also already trust it less. That is a useful combination. Workers who use a tool well and are sceptical of it produce better outcomes than workers who use it well and over-trust it. Hire for the combination.
Product builders at AI companies. The dataset is the cleanest product feedback the category has received. The failure modes that produce anger — confident incorrectness, ungrounded outputs, unhelpful refusals — are the most important things to fix in the next product cycle. They are also the things being deprioritised in favour of capability gains. The trade-off is wrong.
Parents of teenagers. The cohort's scepticism is partly self-protective. Most Gen Z respondents have already run an AI query that produced a wrong answer they had to catch. That experience is a more durable critical-thinking lesson than any media-literacy programme. Parents who respond by panicking about use rather than by reinforcing the scepticism miss the move.
Workers in non-AI-native roles (most workers). The right posture is the cohort's posture. Use the tool. Know the failure modes. Verify before relying. Calibrate.

Cross-layer implications

Trust and information. Gen Z's scepticism is shaped by daily exposure to AI outputs. That kind of trust calibration tends to generalise across categories. Workers who develop scepticism toward AI also develop sharper scepticism toward search results, headlines, and corporate communications. The cognitive habit transfers.
Mental health. The "anger" finding is real and warrants attention. Repeated failure-mode interactions, especially in academic and work contexts, do produce frustration. The fix is not to demand the tool be perfect — it isn't going to be — but to teach a stance toward the tool that does not require perfection to be useful.
Education economics. Most generative AI tools are now optimised for general-purpose chat. Education-specific products that handle failure modes transparently (showing sources, surfacing uncertainty, refusing confidently) are under-supplied. The market is large. The product opportunity is wider than the current category leaders are addressing.
Workforce development. AI literacy is now bifurcated. Cohorts that have used the tool extensively know its failure modes. Cohorts that haven't, don't. Workplace training that treats both cohorts as starting from zero misses the actual variance.

What this means for you

If you're Gen Z, using AI tools — your scepticism is signal, not anxiety. Trust it. Calibrate around the failure modes you've observed. Don't let public commentary push you back toward either uncritical enthusiasm or wholesale rejection. The middle stance is the productive one.

If you're a parent or teacher — the worst response is to take the scepticism as a problem to be solved. The second-worst response is to dismiss it as faddishness. The most useful response is to ask the young person specifically what they've found the tool gets wrong. The answer is almost always interesting and usually accurate.

If you're a workforce leader — the cohort's combination of high use and low trust is a hiring asset. Look for it explicitly. Workers who use the tool fluently and describe its limitations confidently are the ones whose AI-augmented output you can rely on. Workers who use the tool fluently and describe it in superlatives are the ones whose output you have to check more carefully.

If you're a product builder — fix the failure modes first. The capability frontier is being chased by every major lab. The reliability frontier is mostly empty. The product that solves "I can trust the answer" is a more valuable product than the one that solves "the answer can do more."

If you're early in your career and trying to position around AI — the answer is not to use the tool more or to use it less. It is to develop legible judgment about when the tool is right and when it isn't. That judgment is the durable skill. Output volume is not.

Uncertainty ledger

The "feels angry" 30% figure is sensitive to question framing. Other 2026 surveys put the irritation rate between 20% and 35%. Directionally consistent.
Whether the scepticism continues to deepen or stabilises is the question that will define the next two years. If product reliability improves meaningfully, scepticism plateaus. If it doesn't, the trajectory continues.
The cross-cohort comparison is partly a survey-design artefact. Older cohorts use AI less and have less daily exposure to its failure modes. Their higher trust is, in part, lower information.
The split between consumer AI use and enterprise AI use is real and is widening. The Gen Z dataset speaks mostly to the former. Enterprise-AI trust trajectories should be read separately.

The bottom line

The most-using cohort becoming the most-sceptical cohort is not a problem with the cohort and is not a problem with AI. It is what learning looks like inside a real tool category. Excitement turns into use, use turns into pattern recognition, pattern recognition turns into calibrated trust. Gen Z is at the third stage. Most of the public commentary is reading them backwards because most of it is still at the first. The right response — for educators, employers, product builders, parents, and workers themselves — is to take the scepticism as the leading indicator it is, fix what the cohort is telling you to fix, and stop framing maturity inside a tool as rejection of it. The cohort is not turning on the tool. It is learning the tool. That distinction is most of the story.

Sources

New York Times, Gen Z Uses AI More and Trusts It Less, 9 April 2026 — Tier 1
Gallup / Walton Family Foundation / GSV Ventures Gen Z and AI 2026 survey — Tier 1
Pew Research Center, longitudinal AI-trust polling — Tier 1 (background)
Stanford HAI, AI Index Report 2026 — Tier 1 (background)