The conventional story of technology adoption runs like this: a new technology appears, early adopters embrace it, skeptics watch and eventually join, and the technology becomes normal. The adoption is driven by enthusiasm — a genuine belief, however naïve in retrospect, that the technology is good, that it improves life, that being an early adopter is being on the right side of something. The story of the internet, the smartphone, and social media are all versions of this narrative. People adopted them because they believed in them.
The Pew Research survey of 5,119 US adults conducted in February 2026 describes a different story. Forty-nine percent of respondents now use AI chatbots — up from roughly a third in 2024, a substantial increase in a short period. Forty percent say AI will make society worse. Sixty-three percent say AI is advancing too fast. Fifty-nine percent say they have little to no confidence in the companies developing AI to do so responsibly. Sixty-seven percent say they have little to no confidence in the government to regulate it.
What the survey describes is not a technology that people have adopted because they trust it. It is a technology that half the country is using, despite expecting it to harm the world they live in. That is a condition without a clear precedent in the history of mass consumer software adoption
What past adoption waves looked like
The internet reached 50% US household penetration around 2001, by which point it had been widely characterized as a transformative force for human communication, commerce, and access to information. The prevailing public sentiment was optimistic, sometimes extravagantly so. The smartphone reached 50% US adult ownership around 2012-2013, during a period of broad cultural enthusiasm for mobile computing and the connected life it enabled. Social media platforms grew through the 2010s on the explicit promise of connection — keeping in touch with people you cared about, finding communities organized around shared interests, having a voice in public conversation. The early criticism of these technologies existed and was occasionally prescient, but it was a minority position during the period of rapid adoption. Most people who adopted them did so because they expected them to make their lives better.
This is not to romanticize those adoption waves. The internet’s optimism produced the dot-com crash. The smartphone’s promise came with surveillance infrastructure that most early adopters didn’t understand they were accepting. Social media’s connection promise arrived attached to engagement algorithms that turned out to optimize for outrage rather than community. The retrospective picture of each technology is substantially more complicated than the enthusiasm that drove its adoption. But the enthusiasm was real, and it was the engine of adoption.
AI chatbots in 2026 are being adopted at the same rate without the enthusiasm. The majority of Americans using them expect society to be worse for it. This is a genuinely new configuration of the technology-adoption relationship, and the question it raises is not whether people will keep using AI — the data suggests they will — but what it means to build an industry on adoption that is driven by something other than belief.
Why people use things they don’t trust
The mechanism driving AI adoption without trust is not difficult to identify. It is competitive pressure, or more precisely, the fear of falling behind a standard that AI is setting. About a quarter of Americans now report using AI chatbots daily; 12% use them several times a day. For this group, AI has become a tool that makes specific tasks faster or more tractable — writing, research, coding, summarizing, generating options to consider. The individual who uses an AI chatbot to draft an email faster is not making a statement about AI’s societal impact. They are making a local, rational decision about productivity. The fact that they expect the technology to make society worse is a separate belief, held in a separate register.
This separation — between what I use and what I think is good — is not unique to AI. People use tobacco while knowing it is harmful to their health. People use social media platforms whose design they find manipulative. People shop at retailers whose labor practices they would not endorse. The consistency between personal behavior and social values is an aspiration, not a steady state, and the gap between the two is especially wide when the cost of consistency is high — when not using AI means being slower, less capable, or less competitive than the people around you who are using it.
What AI has done faster than most previous technologies is produce a felt competitive necessity. The adoption of social media was partly social — not being on Facebook or Instagram had social costs in terms of connection and visibility. The adoption of AI is largely productive — not using it carries efficiency costs in contexts where it is increasingly the baseline. The feeling that you have to use it is not driven by peer pressure in the social sense. It is driven by the sense that the alternative is falling behind a standard that is being reset by people who aren’t waiting for the broader social questions to be resolved.
What the distrust consists of
The 40% who say AI will make society worse are not a monolith, and understanding what the distrust consists of matters for understanding what, if anything, might change it. The Pew data gives some texture. About half of Americans who get news from AI chatbots say they sometimes encounter information they think is inaccurate. Distrust of accuracy is one component — a belief that the technology produces unreliable outputs — and it coexists with continued use because inaccurate-but-fast is still useful for many applications.
A second component is distrust of governance. Sixty-seven percent of Americans have little to no confidence in the government’s ability to regulate AI, and 59% have little to no confidence in the companies developing it to do so responsibly. These are not small numbers, and they are not primarily about AI’s technical capabilities. They are about whether the institutions that control AI’s development and deployment can be trusted to make decisions in the public interest. The answer, across a substantial majority of the public, is no — and this is the dimension of distrust that matters most for understanding the long-term social contract around the technology.
A third component is pace. Sixty-three percent say AI is advancing too fast. This is not the same as saying AI is bad. It is saying that the speed of deployment has outrun the capacity of society — its regulatory frameworks, its ethical norms, its labor markets, its educational systems — to adapt. The concern is not that AI won’t work. It is that it will work faster than the human institutions that are supposed to manage its consequences can keep up with it.
What it means to build on adoption without belief
The companies building the AI economy — OpenAI, Google, Anthropic, Microsoft, Meta — are doing so against a backdrop of public sentiment that is substantially more skeptical than the sentiment that attended the build-out of previous technology platforms. This matters for reasons that go beyond public relations. Technology industries that develop in an environment of public trust tend to benefit from regulatory tolerance, from consumer goodwill that absorbs early failures, and from a cultural narrative that frames the industry’s growth as progress. Technology industries that develop in an environment of public skepticism face the inverse conditions: tighter regulatory scrutiny, less consumer goodwill to absorb failures, and a cultural narrative that frames growth as a problem to be managed.
The current AI industry is operating in the second environment while applying the assumptions of the first. The product launches, the capability announcements, the deployment timelines, and the investment levels are all calibrated to a world in which the public is, if not enthusiastic, at least broadly receptive. The Pew data suggests the public is broadly using and broadly skeptical — a combination that is more fragile than either enthusiasm or resistance alone, because it is held in place by felt necessity rather than genuine buy-in. Felt necessity can evaporate when alternatives appear, when the competitive advantage of AI tools narrows, or when a sufficiently visible failure makes the cost-benefit calculation shift.
The 49% adoption rate is, by the industry’s own metrics, a success story. Half of American adults using the technology within a few years of its mainstream introduction is an adoption curve that most new technologies don’t achieve. A 2025 YouGov survey found the same pattern a year earlier: widespread use, persistent distrust.
What this adoption rate actually represents — belief in the technology, or resignation to it — is a question the headline number doesn’t answer. The Pew data does, and the answer is instructive. We have reached the point where people are using a thing they distrust at the rate they once used a thing they believed in. Whether those two conditions produce the same outcomes, over time, is the question the industry has not yet had to answer.
