Estimated reading time: 7 minutes · Last updated:
American public opinion on AI leans toward worry, and the lean holds however pollsters phrase the question. Five probability-based surveys fielded in 2026 each recorded more negative than positive views of artificial intelligence. They were run by Pew Research Center, Gallup with Bentley University, Quinnipiac University, Ipsos and the Annenberg Public Policy Center. Pew's June poll is the starkest: 52% of adults were "more concerned than excited" and 9% the reverse, a gap above five to one. Max Roser set the results side by side and found the direction never flipped, as first reported by Our World in Data. The worry extends to the gravest scenario. In September, 73% told Quinnipiac they were concerned AI could threaten human survival.
Key takeaways
- Five surveys, one direction: Pew, Gallup and Bentley University, Quinnipiac, Ipsos and the Annenberg Public Policy Center all found more Americans viewing AI negatively than positively in 2026.
- Pew's gap has widened: Pew Research Center measured 52% of adults more concerned than excited about AI and 9% more excited in June 2026, a ratio above five to one, against about two to one in 2021.
- Intense views skew hostile: In an Ipsos poll for ABC News and The Washington Post, 17% called AI extremely bad for society and 2% called it extremely good.
- Survival fears are mainstream: Quinnipiac's September 2026 poll found 38% very concerned and 35% somewhat concerned that future AI systems could threaten human survival.
- Worry crosses party lines: Pew found 56% of Democrats and 49% of Republicans more concerned than excited about AI, with roughly one in ten in each party leaning excited.
Table of contents
- Key takeaways
- Question Wording Moves the Numbers, Not the Direction
- Pew Finds the Concern Gap Widening Since 2021
- Strong Feelings About AI Are Mostly Hostile
- Most Americans Worry AI Could Threaten Human Survival
- Concern About AI Is Bipartisan, With Democrats More Worried
- Could Public Views of AI Soften?
- What to be careful about
- Frequently asked questions
Question Wording Moves the Numbers, Not the Direction
Max Roser warned that poll results can swing when a question is reworded or its answer choices change. A neutral option shows the effect clearly. In Pew's survey, 37% picked the middle answer and just 9% chose the positive one. Gallup and Bentley University, polling 3,270 Gallup Panel members online from 4 to 11 May 2026, also found only 9% taking the positive view.
To limit distortion, Roser used only probability samples. A probability sample is one where respondents are selected at random by the pollster rather than volunteering to take part. Quinnipiac dialled randomly generated landline and cell numbers, while the other four used online panels whose members were recruited at random, mostly by mail. All five weight their results to match US adults and publish their full question wording.
Opt-in panels, which anyone can join, were left out. Roser showed why: one opt-in poll found 20% of Americans under 30 agreeing that the Holocaust is a myth, while Pew's probability panel put that figure at 3%.
Pew Finds the Concern Gap Widening Since 2021
Pew Research Center's poll, written up by Monica Anderson and William Bishop and published on 16 September 2026, surveyed 3,488 adults on the American Trends Panel from 22 to 28 June 2026. Its core question asks whether wider use of AI in daily life leaves people more concerned, more excited or a mix of both. Concern led with 52% against 9% for excitement.
Pew first posed the question on the same panel from 1 to 7 November 2021. At that point, concern outnumbered excitement by about two to one. That earlier wording was preceded by a brief definition, and instead of "artificial intelligence (AI)" it spoke of "artificial intelligence computer programs". The answer choices stayed the same, and Pew reports both years as one trend.
That change in wording is worth keeping in mind. But a move from roughly two to one to more than five to one in under five years is large. It looks more like opinion hardening than a phrasing artefact, though Pew's data alone cannot prove which explanation is correct.
Strong Feelings About AI Are Mostly Hostile
Ipsos polled 2,545 adults on its KnowledgePanel for ABC News and The Washington Post from 9 to 13 October 2026, asking whether AI is good or bad for society. At the extremes, 17% called it "extremely bad" and 2% "extremely good".
The imbalance matters as much as the totals. A public that is simply unsure would show strong views roughly balanced on each side. Here, most people with a firm opinion of AI hold a hostile one, and genuine enthusiasts make up a small minority.
Only one question in Roser's set produced an even split, and it asked about AI's effect on respondents themselves rather than on society. Even on that question, Ipsos found intense negative views well ahead of intense positive ones. Self-interest, in other words, softens the verdict without reversing it.
Most Americans Worry AI Could Threaten Human Survival
Quinnipiac University's September poll moved past everyday benefits and harms. Live interviewers reached 1,202 adults by random digit dialling between 24 and 27 September 2026. They asked how concerned people were that future AI systems might endanger human survival.
Of those asked, 38% said they were very concerned and 35% somewhat concerned, a combined 73%. Another 14% were not so concerned, 11% were not concerned at all and 1% gave no answer or did not know.
Quinnipiac released the results on 30 September 2026, alongside support for independent safety standards for AI companies. The pairing suggests that many people connect their fears to a wish for oversight. The poll does not show which specific measures respondents have in mind.
Concern About AI Is Bipartisan, With Democrats More Worried
Pew's partisan breakdown found 56% of Democrats and 49% of Republicans more concerned than excited, a gap of about 7 percentage points. Pew built its September release around Democrats overtaking Republicans in worry about AI, including over its effect on jobs.
The more striking figure is how few partisans are enthusiastic. In each party, roughly one in ten said excitement outweighed concern. That leaves AI companies without a natural political constituency of supporters to rely on.
Roser drew a pointed lesson from the data. He argued that the industry will need to treat these worries far more seriously than it has so far if it wants the public to trust it. The fifth survey in his set, the Annenberg Public Policy Center's poll of 1,320 adults run from 16 June to 19 July 2026, points the same way as the others.
| Survey | Fieldwork | Adults | Method | Finding cited |
|---|---|---|---|---|
| Pew Research Center | 22-28 June 2026 | 3,488 | Random-recruited online panel | 52% concerned, 9% excited |
| Gallup and Bentley University | 4-11 May 2026 | 3,270 | Random-recruited online panel | 9% positive |
| Quinnipiac (March) | 19-23 March 2026 | 1,397 | Telephone | 55% expect more harm than good in daily life |
| Quinnipiac (September) | 24-27 September 2026 | 1,202 | Live-interviewer telephone | 73% concerned about survival threat |
| Annenberg Public Policy Center | 16 June-19 July 2026 | 1,320 | Random-recruited online panel | More negative than positive |
| Ipsos for ABC News and The Washington Post | 9-13 October 2026 | 2,545 | Random-recruited online panel | 17% extremely bad, 2% extremely good |
Could Public Views of AI Soften?
The case for
- On AI's effect on them personally, Ipsos found Americans split 50% good to 49% bad, so people's own experience is less hostile than their view of AI's effect on society.
- Quinnipiac found 86% support for independent safety standards for AI companies, which points to a concrete measure that could address some public concern.
The case against
- Pew's concern-to-excitement ratio rose from about two to one in 2021 to more than five to one in 2026.
- Strong views run heavily against AI, with 17% calling it extremely bad for society against 2% extremely good in the Ipsos poll.
- Concern spans both parties, with only around one in ten Democrats or Republicans more excited than concerned.
What to be careful about
- Pew's 2021 question used different wording from the 2026 version, so part of the widening gap could reflect the change in phrasing.
- The middle option drew 37% in Pew's survey, so the measured positive and negative shares depend heavily on the answer choices offered.
- The figures cover US adults only and cannot be read as a measure of opinion elsewhere.
- Polls based on opt-in panels can produce badly skewed results, as the gap between 20% and 3% on the Holocaust question showed, so they are not comparable with these surveys.
The bottom line
Together, the five 2026 surveys leave little room for the idea that American scepticism toward AI is a product of question design. Wording changed how many people stayed on the fence, from the 37% who chose Pew's middle option to the 50% to 49% split Ipsos found on personal effects. It never produced more positive than negative views of AI's wider impact. The clearest trend line, from Pew, shows concern pulling further ahead of excitement since 2021. The most striking single number, Quinnipiac's 73%, attaches to the most extreme risk. Partisanship does not explain the pattern away. For the companies building AI, the data describe a public that has not been persuaded and appears to be drifting further from it.
What to watch
- Watch for Pew Research Center's next wave of its concerned-versus-excited question on the American Trends Panel, which will show whether the ratio above five to one widens further. No date has been set.
- Watch whether AI companies or lawmakers respond to the 86% support Quinnipiac found for independent safety standards. No timetable has been announced.
Frequently asked questions
Do Americans think AI is good for them personally?
Opinion is split almost evenly. In the October 2026 Ipsos poll, 50% said AI is a good thing for them personally and 49% a bad thing. Strong views still leaned negative, with 15% calling it extremely bad for them against 3% extremely good.
Do Americans expect AI to help or harm their daily life?
Most expect harm. In Quinnipiac's March 2026 poll of 1,397 adults, 55% said AI would bring more harm than benefit to their daily lives, while 34% thought the benefits would win out.
Do Americans support safety rules for AI companies?
Quinnipiac's poll released on 30 September 2026 found 86% support for independent safety standards for AI companies. The same poll found 73% concerned that future AI could threaten human survival.
Related reading