Any one of the past ten days' AI stories would have led the news. On November 12, Jensen Huang resigned as Nvidia's CEO after reports of an affair with Alice Huang, an executive at a Singapore chip buyer whom U.S. intelligence agencies suspect of spying for China. China denies it. The same day, Microsoft 365 went down, the start of six major outages in six days: an FAA ground stop, United, Chase, Apple Pay and the NSA's own network. On Thursday, Anthropic went public in the largest IPO on record. On Friday the S&P 500 and the Dow closed in correction territory, and the Nasdaq gave back its gains for the year. Also on Friday, an Anthropic mathematician posted a claimed proof of the Riemann Hypothesis, built with Claude and not yet verified.
That's a lot of AI news in less than two weeks, and a good moment to step back and ask where Americans actually are on AI. Poll Vault has tracked 166 AI polls in the past four weeks. Here's what they say.
Nearly Half, Several Times a Week
Use of AI is high, and it has grown fast. In YouGov's November survey, 45% of Americans use AI at least several times a week, and 16% use it several times a day. Only 19% never use it. The Reuters Institute tracked daily use across six countries and found it went from 7% to 28% in three years.
With that much use, you might wonder why so many Americans are still reluctant about AI. Part of the answer is in the use numbers themselves, which look better than they are. Some of that use isn't voluntary. In a September Visier survey, 45% of workers said they feel pressure to use AI at work without knowing how to use it well, and 48% admitted exaggerating their AI use to colleagues or bosses.
And a lot of it is careful. People use AI to look things up and keep the decision for themselves. Three in four Medicare beneficiaries who shopped for 2027 plans used AI to research them, according to eHealth, but only 4% relied on it most; half relied most on a licensed agent. Harris finds 65% would let AI compare prices for them, but only 41% would let it make the purchase.
That 24-point gap is people testing the waters. Adoption has been fast, but full adoption, where people let AI act for them instead of only advising them, is coming much more slowly.
Good for Me, Not for Thee
So use is mainstream, but trust isn't. Asked about AI's effect on their own lives, Americans lean positive, 39% to 27%. Asked about its effect on society, they lean negative, 40% to 35%.
I see the same split every month in the tracker I run for Centerline Liberties, which asks about AI from both directions. In our September wave, 62% of voters were hopeful AI could improve their own daily lives, a number that has held steady since June. At the same time, 84% were concerned that AI could be misused or cause harm, up 8 points since June.
It's the same pattern as "I hate Congress, but I like my congressman." People trust themselves to use AI well. They don't trust everyone else to, and they don't trust the companies and institutions around it.
People trust themselves with AI. They don't trust everyone else with it.
Down for Everyone
The outages of the past ten days, which were far from abstract, aren't helping. Planes sat on the ground, Chase customers had trouble reaching their money, and Outlook went down for offices across the country. Officials say none of it was a cyberattack and none of it involved AI.
They set off a bout of paranoia online. Recent polls reveal something much deeper that drives that paranoia: a profound skepticism that business or government can stop the next outage or attack.
YouGov asked on November 19: 62% are not confident that big American companies can prevent a major cyberattack, and only 7% are very confident. In a separate YouGov question the same day, 59% were not confident the federal government could prevent one either.
YouGov asked that question the same week officials said the outages came from software bugs and aging systems, not an attack.
It fits something I come back to often: institutional trust in this country is close to bankrupt. In a September Economist/YouGov poll, 16% said they trust the government in Washington to do what's right most of the time. Kettering and Gallup surveyed more than 23,000 adults and found only 28% believe political leaders will be held accountable to the law. And the people building AI start from a deeper hole: in Quinnipiac's September poll, 3% trust the leaders of AI companies a lot.
This is where AI picks up the doubt. Four days after the FAA ground stop, YouGov asked whether using AI to manage air traffic would change flight delays. 29% said delays would become more common, 16% less common, and 39% weren't sure.
And yet people still want Washington in charge. Data for Progress finds 75% of likely voters support a new federal agency to regulate AI, including 66% of Republicans. That's a real contradiction. The same public that doubts the federal government could stop a major cyberattack, and expects it would keep one quiet, wants that government to police AI. I read it as a demand for someone to be accountable, more than a vote of confidence in the people who'd do the job.
Poll Vault tracked 166 AI polls in the past four weeks. See them all β
Jensen Who?
The Huang scandal dominated tech coverage for days. Most Americans never knew who he was.
In April, a Verasight survey found 45% of Americans didn't know who Huang was; among the rest, he was slightly more liked than disliked (18% favorable, 12% unfavorable). After the scandal, Economist/YouGov has him at 12% favorable and 34% unfavorable, with 54% still unable to say. His unfavorable number nearly tripled, and a majority still has no opinion of the man who ran the world's most valuable chip company.
He's not unusual. In Navigator Research's October poll, 68% of voters had never heard of Anthropic's Dario Amodei and another 15% didn't know enough to rate him. For OpenAI's Sam Altman, it was 41% never heard of him and 24% no opinion.
I'd bet even those numbers overstate how many people actually know these names; plenty of respondents will rate someone they only half recognize. AI's most powerful executives are an online story. Most Americans aren't following it as closely as the people who write about it.
The national-security worry is real, though. Two days before Huang resigned, YouGov found 74% of Americans concerned about foreign spies getting information from U.S. tech companies. In the Economist poll, 30% had heard nothing about the affair, yet 54% said it probably or definitely compromised national security. Half said the U.S. should have banned chip sales to Megaspeed, Alice Huang's company, including 60% of Republicans.
Remarkably though, that worry about China doesn't turn into a desire to race China on AI. In my own surveys, I've tested the argument that the U.S. has to beat China in AI and technology many times, and it rarely moves people. The public polls agree. In a September Verasight poll, 45% worried more that AI is being developed too fast and will cause serious harm; 11% worried more about falling behind China. Echelon Insights found 76% of likely voters want safety first even if it slows innovation, against 15% who'd rather keep pace with China. People want the spies kept out. They don't want to win the race at the cost of safety.
Bubble Talk
A market bubble is partly a matter of belief. If enough people with money in the market start to think AI stocks are overpriced, they sell, and the belief makes itself true.
The public was already there in early October. In Navigator's poll, 57% of voters were concerned that AI stocks would become "so overpriced β sometimes called a 'bubble' β that they eventually cause a stock market crash." 29% were very concerned. The professionals moved the same way: in the New York Fed's fall survey of market contacts, 87% named AI as a risk to financial stability, up from 50% in the spring.
The optimism sits at the top. In a Charter survey of 5,000 workers, 85% of executives were optimistic about AI's effect on their organization, compared with 56% of employees.
Proof of Concept
If Levent AlpΓΆge's Riemann proof holds up, it will be one of the biggest results in the history of mathematics. Outside mathematicians have already checked that his computer-verified version compiles, though formal review will take longer. But don't assume this will change how many Americans feel about AI.
Math and code are formal fields with clear rules, and AI has made the fastest progress there. Proofs are impressive, but they don't touch most people's lives. Attitudes will turn when AI changes people's lives outside work: when it produces a breakthrough in biology, chemistry or medicine that someone can see in a doctor's office. Turning family photos into GIFs won't do it.
The polls show what the public is waiting for. In the same Navigator poll where 57% worried about a bubble, 67% found it appealing that AI could help doctors detect diseases earlier and speed new cures, and 61% liked the idea of AI producing scientific breakthroughs on problems like hunger and clean water. Only 13% said they don't believe AI can do it.
People also draw a clear line around where they want AI. The North Texas Quality of Life Initiative found residents want more AI in weather forecasting, science and clinical healthcare, and less in counseling, relationship advice and spiritual guidance. And when Milltown Partners tested arguments for AI this summer, the one about doctors catching cancer earlier and scientists developing new medicines moved support by 10 points, more than any other.
The past ten days gave people plenty of reasons to doubt AI. The polls show what would change their minds, and it's medicine more than math.
This Week's Shorter Stories
π€ "Speak to a human." TELUS Digital finds 46% of regular AI users have given an AI customer-service bot exaggerated or false information to get through to a human. 77% say a guaranteed human option would make them more likely to choose one company over another.
π Learn a trade. In a Careerminds survey, 62% of workers would rather their child become an electrician than a software developer.
ποΈ Not this year. Bank of America finds 72% don't plan to use AI for holiday shopping, up from 49% last year.
π¬ Someone to talk to. A Gallup and Digital Wellness Lab study of nearly 16,000 young people finds 33% of 15-to-22-year-olds rely on AI for companionship, rising to 50% among daily AI users.
What Else We Tracked This Week
- Americans Report AI Productivity Gains While Most Expect No Near-Term Job Security Effect - The Conference Board
- U.S. Employee Engagement Falls to 29%, Lowest Since 2011 - Gallup
- Most Students Use AI When Distressed; Report Urges Youth AI Safety Rules - The Jed Foundation
- Only 8% of Americans Use AI Chatbots for Election Information - Pew Research Center
- Three-Quarters of Americans Say They See Ads They Think Are AI-Made - YouGov
- Faculty Say AI Has Eroded Their Trust in Students' Work - Inside Higher Ed / Hanover Research
- Most Surveyed Across Eight Countries Say AI Music Should Be Labelled - Deezer / Ipsos
Until next time,
Alex Lundry
Due Data is powered by Poll Vault. Get the full data behind every poll mentioned above.
Want the full data behind every poll mentioned above?
Create free account β