What happens when generation already anxious about AI is repeatedly fed the content that frightens them most?
42% of Gen Z say AI makes them anxious.
Only 18% say it makes them hopeful.
Those figures come from a 2026 Gallup survey of Americans aged 14 to 29.
Hopefulness about AI fell from 27% to 18% in just one year. Excitement fell from 36% to 22%, while anxiety stood at 42%.
There is another figure worth noticing.
Among Gen Z respondents who never use AI, 60% reported feeling anxious about it. Among daily AI users, that figure was 28%. That does not tell us why. But it should make us curious. Because while young people are trying to understand what artificial intelligence might mean for their education, careers and future, their social-media feeds increasingly contain another narrative:
AI will take your job.
Humans will lose control.
Superintelligence is coming.
Human extinction is possible.
Perhaps we do not even have another decade.
One Reel ends.
The next one begins.
The issue is not discussing risk
Artificial intelligence raises legitimate questions about employment, education, misinformation, security, power and control. Those questions belong in public debate.
What concerns me is something different. It is what happens when uncertainty is converted into engagement.
A researcher saying that a catastrophic outcome is possible is not the same as telling a teenager there may be no future left to plan for.
A probability is not a prediction.
A scenario is not a forecast.
And uncertainty is certainly not inevitability. 8Yet those distinctions are remarkably easy to lose in a 30-second video.
“Researchers disagree about the probability and timescale of advanced AI risks” is not a particularly effective hook. “We may have less than ten years.” That is.
Fear performs
Recommendation systems do not need to understand fear in the way humans do.
They need signals. Did you stop scrolling? Did you watch until the end? Did you comment? Did you share it? Did you watch another video on the same subject?
If the answer is yes, the system has learned something.
Not necessarily that you liked what you saw. But that it held your attention. And attention has value.
Research into engagement-based ranking systems has shown that emotionally charged material can receive more amplification than it would in a simple chronological feed. That matters. Because an algorithm is not merely a neutral pipe through which information travels.
Ranking is a choice.
What is shown first, repeated, recommended and amplified helps shape the information environment in which people form their perception of reality.
An engagement system does not naturally ask: Will this 16-year-old close the app feeling more secure about the future?
It asks: Did they keep watching?
Anxiety is already rising
Gallup is not the only organisation seeing this unease. Pew Research Center reported in 2026 that 55% of Americans aged 18 to 29 were more concerned than excited about AI.
In 2024, that figure was 39%. And 73% expected AI to result in fewer jobs in the United States over the next twenty years. Those figures matter, but they also need to be interpreted carefully. They do not prove that algorithms caused young people’s anxiety about AI.
A 16-year-old brain is not simply a smaller adult brain.
During adolescence, sensitivity to social feedback, emotional cues and reward is heightened, while the systems involved in self-regulation and long-term judgement are still developing.
That does not make young people powerless.
It does make the design of recommendation systems more consequential.
An algorithm optimised to hold attention does not meet a neutral user. It meets a developing nervous system.
If emotionally charged content repeatedly captures attention, the risk is not only what is shown.
It is how often it is reinforced, at what age, and in what developmental context.
Young people have rational reasons to be uncertain. They hear employers discussing automation. They see professions changing.
Schools and universities are still deciding how AI should be used. Experts openly disagree about what AI may be capable of in ten or twenty years. Their concerns cannot simply be dismissed as something an algorithm invented.
But that is not the only question we should ask.
The relevant question is: Can an existing fear be amplified by the system in which people repeatedly encounter information?
Evidence, mechanism and assumption
When assessing a possible risk, three things should not be confused.
What do we know?
We know that substantial numbers of young people report anxiety and declining hopefulness about AI. We know that engagement-based recommendation systems can amplify emotionally charged content.
What is the plausible mechanism?
Fear captures attention.
Attention creates engagement.
Engagement influences recommendation.
Recommendation creates further exposure.
What do we not know?
Is repeated AI-doom content directly causing long-term hopelessness in an individual young person?
All of those statements can be true at the same time. That is why a proper assessment requires: separating evidence, mechanism, exposure, vulnerability and uncertainty.
The peat fire
Fear does not always look like panic.
Sometimes it is more like a peat fire.
The Reel is gone. But the screen already shows something else. And somewhere underneath, a thought can keep smouldering.
Will my profession still exist in 10 years?
Will humans remain in control?
Should I even plan ten years ahead?
Is there a future to plan for?
Then another headline appears, another expert, another warning, another countdown. A little more oxygen time by time.
The question is therefore not only whether one Reel frightens somebody. The question is what happens when a recommendation system repeatedly detects which fears hold someone’s attention — and keeps returning to them.
A peat fire does not need dramatic flames. It can remain below the surface.
Vulnerability matters
There is a difference between presenting catastrophic speculation to an experienced professional and repeatedly recommending it to someone who is 14, 15 or 16.
Not because young people are incapable of critical thought.
But because age, experience, context and developmental stage matter when assessing risk.
Gallup’s Gen Z research includes people as young as 14. We already accept the principle that young users may require additional protection online. We regulate advertising aimed at minors. We discuss addictive design. We worry about harmful content and mental wellbeing. We debate age-appropriate online environments.
So why should algorithmic amplification escape the same scrutiny?
The question should not only be: Does this individual post violate a rule? We should also ask: What happens when a platform repeatedly serves the same category of fear to a young person because previous behaviour showed that fear captures their attention?
That is not simply a content question. It is a systems question.
Creator, platform and algorithm
A creator is responsible for what they publish.
A platform is responsible for what it amplifies.
Those are not the same thing.
One creator may publish one dramatic prediction.
A recommendation system can expose somebody to hundreds of similar predictions from hundreds of creators.
There is a term sometimes used for highly sensationalised catastrophic content: Doomsday porn.
The phrase is provocative.
But the mechanism underneath it is more important than the label.
The problem does not require every underlying fact to be false.
The more difficult problem occurs when real research, genuine expert concern and enormous uncertainty are compressed into emotionally irresistible certainty.
A possibility becomes a prediction.
A risk becomes a countdown.
A debate becomes fear.
The feedback loop becomes simple:
Fear. Attention. Engagement. Recommendation. More fear.
Nobody needs to have deliberately designed the entire loop, but this loop can still exist.
When fear becomes commercially valuable
This becomes even more uncomfortable when attention can w0be monetised. Fear stops the scroll. Fear generates comments. Fear gets shared. Reach creates followers. Followers create commercial value. That on itself does not mean every creator talking about AI risk is deliberately frightening young people for gaining money. Intent requires evidence. But on the other hand incentives matter too.
If the most frightening interpretation repeatedly creates the strongest engagement, the system rewards people for turning up the emotional volume.
A concern becomes a threat.
A threat becomes a catastrophe.
A catastrophe becomes imminent.
This is where the expression emotional arson comes in the picture. You do not need to invent the combustible material. The uncertainty is already there. You only need to keep supplying oxygen.
A new technology can be commercially successful while still producing social harm. And a business model can be profitable while exploiting human vulnerabilities. An information system (platform) can function exactly as designed while producing outcomes society does not consider desirable.
That makes this a social sustainability question, for
What is the impact?
Who is exposed?
Who is particularly vulnerable?
How frequent is the exposure?
What evidence supports the concern?
What remains uncertain?
What controls are available?
Who owns the residual risk?
These are normal questions when assessing complex organisations, technologies and systems.
Recommendation algorithms should not be exempt because their effects are less visible.
A better system does not require censorship
None of this means banning difficult conversations about AI.
Nor should young people be shown only reassuring information.
They deserve truthful information about technological risks.
But there is a large space between censorship and unrestricted optimisation for engagement.
Platforms can examine repeated exposure.
They can give users meaningful control over recommendations.
They can provide greater transparency about why content keeps appearing.
They can consider age and vulnerability.
And recommendation systems can optimise for more than raw engagement.
The objective should not be to make people feel artificially safe.
It should be to avoid building systems that become more successful when people feel increasingly unsafe.
That distinction matters.
42% anxious. 18% hopeful.
Those numbers do not prove that algorithms are stealing a generation’s hope.
But they should make us pay attention.
A substantial part of a young generation is already approaching AI with anxiety rather than confidence.
Hopefulness has fallen.
Concern about jobs is growing.
And many of the systems through which they encounter information are designed to learn what captures their attention.
That combination deserves assessment.
Not panic.
Not censorship.
Assessment.
AI deserves serious discussion.
Existential risk deserves serious discussion.
Employment disruption deserves serious discussion.
But seriousness and sensationalism are not synonyms.
Protecting young people online should therefore mean more than removing individual pieces of prohibited content.
It should also mean examining the architecture that decides what comes next.
Because if fear captures attention, and attention drives recommendation, the important question may no longer be whether an algorithm can detect our anxiety.
It is whether the algorithm is learning how to keep it alive.
Emotional arson does not always produce visible flames. Sometimes a peat fire is enough.
Sources
Gallup — Gen Z’s AI Adoption Steady, but Skepticism Climbs (2026)
Pew Research Center — Young US adults are increasingly wary of AI, concerned it will take jobs (2026)
PNAS Nexus — research on engagement-based ranking and emotional content (2025)
Scientific Reports — research auditing recommender-system amplification (2023)