What Is the Pew Research AI Paradox? Why More People Use AI but Trust It Less
49% of US adults now use AI chatbots, up from 33% in 2024—yet more Americans predict AI will have a negative impact on society. Here's what the data shows.
The Numbers That Shouldn’t Make Sense Together
Nearly half of American adults now use AI chatbots. That’s 49%, up from 33% just a year prior, according to Pew Research Center’s 2025 data. By almost any measure, that’s rapid mainstream adoption.
But here’s the part that doesn’t fit the usual “technology adoption” story: over the same period, the share of Americans who believe AI will have a negative impact on society grew. More people are using AI tools than ever before — and more people are worried about what AI is doing to the world than ever before.
This is what researchers and observers have started calling the Pew Research AI paradox: rising usage alongside declining trust. It’s a pattern that matters for anyone building AI products, deploying AI in the enterprise, or trying to understand how public opinion on this technology is actually shifting.
This article breaks down what the data actually shows, why the paradox exists, and what it means practically — for businesses, builders, and everyday users.
What the Pew Research Data Actually Shows
Pew Research Center has been tracking American attitudes toward AI consistently enough to reveal real trend lines, not just snapshots. The picture that emerges is more complicated than either AI boosters or AI critics usually admit.
Usage Is Up Sharply
The jump from 33% to 49% of US adults using AI chatbots between 2023 and 2025 is not a rounding error. That represents tens of millions of people who weren’t regularly using tools like ChatGPT, Gemini, or Claude — and now are.
Usage skews younger. Adults under 30 are far more likely to use AI chatbots regularly than those over 50. College-educated adults use them more than those without degrees. But the growth is happening across demographics — even among groups that were late adopters initially.
The use cases people report are practical: help with writing, answering questions, summarizing information, learning new topics, and getting work done faster.
Concern Is Also Up
At the same time, the percentage of Americans who say AI will have more of a negative than positive impact on society has increased. In Pew’s surveys, the share expressing concern about AI’s societal effects has grown even as the usage numbers climbed.
This isn’t just a generalized anxiety. When asked about specific applications, Americans express particular worry about:
- AI being used in hiring decisions
- AI-generated misinformation spreading online
- Surveillance and privacy erosion
- Job displacement across industries
- AI being used in medical diagnoses without sufficient human oversight
What’s notable is that many of the people expressing these concerns are the same people using AI tools. The two groups — users and skeptics — overlap substantially.
The “Helpful but Dangerous” Framing
When Pew asks about specific applications rather than AI in general, the results get even more nuanced. Americans tend to be more positive about AI in low-stakes, assistive contexts (helping with online searches, writing assistance, entertainment recommendations) and more negative about high-stakes, autonomous contexts (medical diagnosis, criminal sentencing, hiring, financial decisions).
This suggests the paradox isn’t really about hypocrisy. It’s about context. People have learned that AI can be genuinely useful for some things. They’ve also learned — or at least become more aware — that it can cause serious harm in others.
Why More Use Doesn’t Equal More Trust
The assumption that “familiarity breeds trust” turns out to be wrong here — or at least incomplete. Normally, as people use a technology more, they get more comfortable with it. That’s what happened with search engines, smartphones, and social media in their early years.
AI is behaving differently, and there are a few reasons why.
Exposure to AI Failures Is Also Increasing
More people using AI means more people personally experiencing its limitations: hallucinations, confident wrong answers, biased outputs, awkward refusals. When someone uses a hammer and it doesn’t work perfectly, the failure is obvious. When an AI gives a plausible-sounding but incorrect answer, the failure can be invisible until it causes a real problem.
High-profile examples of AI errors — fabricated legal citations in court documents, AI-generated fake images spread as news, chatbots giving dangerous medical advice — have made the failures more visible at a societal level even as individual users are getting more utility from the tools.
The Stakes Have Gotten Higher
In 2023, most public discussion of AI was about whether it was impressive or overhyped. By 2025, AI is being embedded in hiring pipelines, customer service systems, content moderation, medical triage, and financial analysis. As AI moves from novelty to infrastructure, the consequences of its failures grow.
People are right to apply more scrutiny to AI making decisions about their job applications than to AI helping them draft an email. The concern isn’t irrational — it’s proportional to real-world stakes.
The AI Industry Has Had a Trust Problem
Repeated overpromising by AI companies — timelines for AGI, capabilities claims that don’t hold up, safety assurances that feel thin — has contributed to a credibility gap. Trust in technology companies was already shaky after years of social media controversies. Extending that skepticism to AI isn’t a leap.
Media Coverage Has Become More Critical
Early AI coverage focused heavily on capability and potential. Coverage in 2024 and 2025 has included substantially more reporting on failure modes, labor displacement, environmental costs, and regulatory concerns. More users are arriving with more context — and more questions.
Who’s Most Skeptical, and Why
The trust gap isn’t evenly distributed. Pew’s data reveals some clear patterns in who is most concerned about AI.
Older Adults
Adults over 50 are more likely to be concerned about AI’s societal effects and less likely to believe its benefits outweigh its risks. This is partly a familiarity gap, but it’s also a values gap — older adults are more likely to prioritize human oversight and human judgment in consequential decisions.
Lower-Income Workers
People in jobs more exposed to automation risk — service workers, administrative roles, transportation — express higher concern about AI’s economic impact. This isn’t abstract for them. AI displacement is a present concern, not a future one.
People Who’ve Had Negative Experiences
Individuals who have directly encountered AI bias, AI-generated misinformation, or AI-enabled scams express significantly higher skepticism. Lived experience with AI’s downside shapes attitudes in ways that general usage doesn’t necessarily counteract.
People with Strong Privacy Concerns
The connection between AI and data collection is increasingly understood by the general public. People who are already concerned about corporate data practices are more likely to view AI systems as another vector for surveillance and exploitation.
The Enterprise Dimension of the Trust Gap
The Pew data tracks public opinion, but the same paradox shows up in enterprise contexts — and it matters for how organizations deploy AI internally and externally.
Employees Are Using Shadow AI
Surveys of enterprise workers consistently show that employees are using AI tools — often consumer-grade tools like ChatGPT — even when their companies haven’t formally approved or deployed them. Usage is happening regardless of official policy.
This creates real risk: sensitive data entering systems outside the company’s control, outputs being used without quality checks, and compliance exposures that legal and security teams haven’t planned for.
Customers Are Paying Attention to How Companies Use AI
Consumer trust in AI-powered services varies significantly based on how transparently companies communicate about AI use. Customers are more accepting of AI when:
- They know AI is involved in the interaction
- They have a clear path to a human if needed
- The stakes are low (recommendations, FAQs)
- The company has a history of trustworthy behavior
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
They’re less accepting when AI is used invisibly in high-stakes decisions, or when companies are vague about what data is being used.
Governance Is Becoming a Differentiator
Companies that invest in AI governance — documented policies, human oversight on consequential outputs, bias auditing, transparency with users — are starting to stand out from those that don’t. As AI regulation increases globally, governance infrastructure that was optional in 2023 is becoming standard practice in 2025.
What the Paradox Means for AI Builders and Deployers
If you’re building AI tools or deploying them in a business context, the Pew Research AI paradox has direct implications.
Useful Isn’t Enough
Products that are merely useful won’t automatically earn trust. Design choices matter: how much control users have, how transparent the AI is about its limitations, whether there are clear escalation paths, and how the product handles failure cases.
Users are increasingly capable of distinguishing between AI that’s genuinely trying to be helpful and AI that’s trying to maximize engagement or minimize support costs.
Trust Has to Be Earned Through Specificity
Vague commitments to “responsible AI” don’t move the needle much with skeptical users. Specific, verifiable practices do: “This tool doesn’t store your conversations,” “Outputs are reviewed by [X type of human] before being sent,” “Here’s what the model can and can’t reliably do.”
The Trust Gap Is Segmented
Different users in different contexts have different trust thresholds. A medical professional using AI for diagnostic support brings entirely different expectations than a college student using AI for research help. Building for one doesn’t mean building for the other.
Effective AI deployment often means matching the level of autonomy and opacity in your AI system to the trust level your users actually have — not the trust level you wish they had.
How MindStudio Addresses the Trust Problem in Practice
One of the underappreciated aspects of the Pew Research AI paradox is that it’s partly a deployment problem, not just a technology problem. People are more comfortable with AI they understand, control, and can verify.
That’s where the way AI gets built and deployed matters as much as the underlying model.
MindStudio is a no-code platform that lets teams build and deploy AI agents — custom-built tools that do specific, defined tasks rather than general-purpose chatbots with unpredictable scope. Instead of handing employees or customers access to a general AI system, organizations can build targeted agents: a contract review tool that only looks at uploaded documents, a customer FAQ agent that only draws from approved knowledge bases, or an internal HR assistant with explicitly scoped capabilities.
This matters for the trust gap in a few concrete ways:
- Scoped tools are easier to audit. When an AI agent is built to do one thing, it’s much easier to verify that it does that thing reliably and only that thing. General AI systems are harder to test exhaustively.
- Custom UIs signal intentionality. When users interact with a purpose-built tool rather than a generic chat interface, they understand what they’re dealing with. The expectations are set correctly from the start.
- Human oversight can be baked in. MindStudio’s workflow builder lets teams design agents that escalate to human review at defined points — high-stakes outputs, low-confidence responses, or edge cases. The AI doesn’t have to operate without a net.
Seven tools to build an app. Or just Remy.
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
The platform supports 200+ models and 1,000+ integrations, so teams aren’t locked into a single AI vendor. As models improve (or as specific models prove more reliable for specific tasks), the underlying AI can be swapped without rebuilding the whole application.
You can start building for free at mindstudio.ai.
For teams already thinking about responsible AI deployment or building internal AI tools, the architecture decisions you make now will affect how much trust your users place in the systems you build.
Will the Trust Gap Close?
There’s no guarantee that trust catches up to usage. Some technology adoption curves don’t converge — social media usage remained high even as trust collapsed. It’s possible to have widespread adoption of a technology that most people are ambivalent or negative about.
But the AI trust gap isn’t purely structural. Several things could shift it:
Better error communication. If AI systems become more reliably transparent about what they don’t know — and less prone to confident hallucination — users will calibrate their trust more accurately. Accurate trust (high trust where AI is reliable, low trust where it isn’t) is better than uniformly high or uniformly low trust.
Meaningful regulation. The EU AI Act, emerging US state-level AI laws, and sector-specific AI regulations are creating accountability structures that didn’t exist before. Regulation that works as intended could increase legitimate trust in compliant systems.
Track records. Trust is built over time. The more years AI systems operate in high-stakes contexts without catastrophic failures, the more baseline trust accumulates. Conversely, high-profile failures continue to erode it. Right now, the track record is still short.
Alignment between who benefits and who bears risk. A significant driver of public skepticism is the perception that AI benefits flow disproportionately to companies and investors while risks fall disproportionately on workers, vulnerable populations, and society at large. Closing that gap — through design, policy, or business model changes — would likely move public opinion.
Frequently Asked Questions
What is the Pew Research AI paradox?
The Pew Research AI paradox refers to the pattern Pew has documented where American adults are using AI tools at increasing rates while simultaneously expressing more concern about AI’s negative societal impact. As of 2025, 49% of US adults report using AI chatbots — up from 33% in 2024 — yet surveys also show a growing share of Americans who believe AI will have more negative than positive effects on society.
Why do people use AI if they don’t trust it?
This is the core of the paradox. The short answer is that individual utility and societal trust are different things. Someone can find an AI writing assistant useful for their own work while also being genuinely worried that AI will displace jobs at scale, enable misinformation, or create privacy risks. People make similar calculations with social media, processed food, and many other technologies — using something and approving of it broadly are separate decisions.
Is AI trust declining everywhere, or just in the US?
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
Pew’s data focuses on US adults, but similar patterns appear in global surveys on AI attitudes. Concerns about AI are broadly distributed across developed countries, though the specifics vary. European respondents tend to prioritize privacy concerns, while concerns about job displacement are more prominent in countries with large manufacturing or service sectors. Trust levels also vary significantly by domain — medical AI, for instance, faces more skepticism than entertainment AI across most countries.
What AI applications do Americans trust most?
According to Pew data, Americans are most comfortable with AI in lower-stakes, assistive roles: search suggestions, content recommendations, translation, and writing assistance. Trust drops sharply for AI involved in consequential decisions — hiring, medical diagnosis, legal proceedings, and credit decisions. The common thread is that trust is higher when humans remain clearly in control and when the AI is seen as a tool rather than a decision-maker.
How should businesses respond to the AI trust gap?
The practical response involves several things: being transparent about when and how AI is used in customer-facing interactions, building in human oversight for high-stakes decisions, giving users meaningful control over AI interactions, and avoiding overpromising what AI systems can do. Companies that are specific about their AI governance practices — rather than relying on vague “responsible AI” language — tend to fare better in consumer trust surveys.
Will AI trust improve as people use AI more?
Not automatically. The “familiarity breeds comfort” model doesn’t seem to be operating cleanly here. Increased usage is correlated with both higher appreciation for AI utility and higher awareness of AI risks. What moves trust is not usage volume but experience quality — whether AI systems behave reliably, transparently, and in ways that align with users’ interests. Better-designed AI systems with clearer limitations, more honest error communication, and stronger human oversight could improve trust. Poor design, repeated failures, and continued opacity would likely erode it further.
Key Takeaways
The Pew Research AI paradox is real, documented, and likely to persist unless the underlying drivers are addressed. Here’s what the data tells us:
- Usage has grown sharply. 49% of US adults now use AI chatbots, up from 33% in 2024 — mainstream adoption is no longer a forecast, it’s a fact.
- Concern has grown in parallel. More Americans believe AI will have a negative societal impact than before, even as personal use increases.
- The paradox makes sense once you separate individual utility from societal trust. People can find tools useful while still worrying about their broader effects.
- Stakes matter. Trust in AI is highly context-dependent — low for autonomous, high-stakes decisions; higher for assistive, low-stakes ones.
- How AI is deployed matters as much as the technology itself. Scoped, transparent, auditable AI applications earn more trust than open-ended, opaque ones.
- Businesses that treat trust as an engineering and design problem — not just a PR problem — will have an advantage as AI scrutiny increases.
If you’re building AI tools and thinking carefully about how to close the trust gap for your users, MindStudio gives you the infrastructure to build AI agents that are purposeful, auditable, and genuinely controllable — the kind of AI that earns trust rather than asking for it.
