Almost every public measure of how Americans feel about artificial intelligence asks one question: are you more concerned or more excited? Pew puts concern ahead of excitement 50% to 10%. Ipsos finds two thirds of the country nervous. The Economist and YouGov find more than 70% saying AI is moving too fast. Read together, they tell a single story of a nation souring on the technology, and that story is built on a question that hides the finding that matters.
Recon Analytics asks the two halves separately, every week: which specific concerns about AI do you hold, and which specific hopes. When the questions are split, so is the country. In the first half of 2026, across a panel that has answered more than 137,000 times, 70.7% of American adults hold at least one specific fear and at least one specific hope at the same time. A single concerned-to-excited scale forces those people to a meaningless midpoint and reports the blend as a darkening mood. The blend is real arithmetic and a false description. Americans are not of one mind about AI. They keep two ledgers at once, and the two ledgers behave nothing alike.
The fear ledger is close to universal and remarkably even. Nearly 9 in 10 adults (88.4%) name at least one concern, and the worries sort into three areas: personal autonomy and privacy, held by 78.4%; truth and trust, held by 55.4%; and livelihood, held by 34.3% and led by job loss at 30.1%, the single most-named fear in the country. What is striking is how little any of it varies by who is answering. A working-age household earning $20,000 and one earning $190,000 fear AI-driven job loss at statistically identical rates. People who use AI every day carry almost exactly as many fears as people who have never touched it. Fear is the same in the city and the countryside, the same at 30 as at 70, and it has not climbed: for ten straight months the average number of fears an American holds has not moved off 2.4 of the fourteen the survey tracks. Whatever is happening to American AI sentiment, rising fear is not it.
The movement is all on the other ledger, and it runs one way. Hope is not universal; it is unequal, and it sorts by who uses the technology. The share of Americans who hold no hope at all for AI runs more than twice as high in the lowest income bracket as in the highest, and more than ten times as high among people who have never used AI as among its heaviest users. Cross income with race and the deficit acquires a face that complicates every easy story: at every income level, White households report the most hopelessness, and low-income White households are the most hopeless income-and-race group in the country, at 36.6% holding no hope. The pattern runs opposite to what many would expect: at the same income, Black, Hispanic, and Asian households hold more hope for AI than White households. It is a portrait of the people AI’s benefits have most visibly passed by: lower-income, rural, older, disproportionately White, and not using the tools.
That population is not warming as the technology improves. It is hardening. Among never-users, the share holding no hope climbed from 46.7% in August 2025 to 56.6% in May 2026, nearly ten points in ten months, even as adoption pulled the most persuadable holdouts out of the non-using pool. The national average sits calmly in the low 20s and hides the split. This is sorting, not souring: a hopeful, using majority pulling away from a hardening, non-using minority. The word hopeless is a measurement here, not a judgment. These Americans are not failing to imagine a benefit; they are reporting, accurately, that nothing in AI has yet been delivered to them worth imagining. Their hopes, where they exist, are modest and concrete: the most common hope in the country is that AI will save people time and stress (31.0%), well ahead of curing disease, boosting the economy, or solving climate change. The public is not waiting to be sold a transformation. It is waiting for something small that works.
This is why the industry’s instinct misses. The companies that shape the public conversation spend the marketing dollars meant to grow users and usage on reassurance: safety commitments, responsible-AI pledges, trust centers. But fear is not the gate. Over half of never-users say they would refuse AI even if it were guaranteed risk-free and built for them, and three quarters of those refusers hold no hope. Among the hardened, 82.3% say nothing at all would build their trust in an AI tool. Reassurance reaches people who already hold hope, the ones most likely to try AI on their own anyway. It does not move the people whose absence defines the gap, because risk was never their constraint.
What moves them, where anything does, is a concrete benefit witnessed through someone they trust. Being shown how AI helped a real person cuts the no-hope rate roughly in half, and a recommendation from family, a friend, or a coworker is the only trust lever that registers with the hardened at all. For the AI industry, that points to a different use of the same dollars: hands-on education and community clinics that walk people through the tools on their own problems; marketing built on tangible, connectable benefits (e.g., saving time on everyday shopping, or sending an AI agent to hunt for a better deal) rather than abstract capability; and basic prompting literacy, because a first session that works is the difference between a user and a refuser, while a first session that flails confirms every suspicion the refuser walked in with. The benefit worth showing is the mundane one the hopeful already name first, not the civilizational promise that fills the keynotes. No single company has the incentive to do this work, because the near-term revenue sits with the hopeful majority and the hardening minority is a shared cost. That is the shape of a collective-action problem, and whether the industry treats it as one is a choice the data can frame but not make.
The trajectory is testable, and it splits in two because adoption and hardening can both be true at once. If the no-hope share among never-users stays above 50% at the end of 2026, the hardening is real and durable. The pool itself is draining toward roughly a fifth of adults by year-end; if it keeps draining, there is no permanent adoption ceiling, only a shrinking hardened core, and the problem becomes political rather than commercial. The deficit becomes a lasting fact, for adoption and for the politics of AI, only if that pool stops shrinking while its hopelessness holds. Either way, the number to watch is not how afraid Americans are. On that, the country already agrees and has stopped moving. The number to watch is how many Americans can still name a single thing they hope AI will do for them, and which Americans they are.
—
Source: Recon AI Pulse consumer survey (Recon_Data_AI), 2026-to-date cross-section n=137,448, monthly trend from August 2025; figures bounded to the January-May 2026 window. External figures (Pew, Ipsos, Economist/YouGov) cited as published context. Recon figures reported to one decimal place. Full methodology, the income-by-race detail, and the underlying charts in the companion report.