What people are really saying about AI right now — and why the story keeps flipping

Everyone has an opinion on artificial intelligence. The interesting part is how fast that opinion is changing, and how differently it’s changing depending on where you live.

Ask ten people what they think about AI and you’ll get ten different answers — but ask the same person twice, six months apart, and you might get two different answers as well. That’s the real story buried under the daily flood of AI headlines: public opinion on artificial intelligence isn’t settling into a steady consensus, it’s oscillating, and the swings are happening fast enough that a survey from early 2025 is already stale history.

Most coverage of “what AI is doing” focuses on the technology itself — new models, new lawsuits, new funding rounds. Far less attention goes to what AI what people say about it once they’ve actually used it, because that data is messier, more contradictory, and harder to turn into a punchy headline. But it’s arguably the more consequential story, because sentiment is what eventually shapes regulation, adoption curves, and which products succeed.

The number everyone quoted in 2024 has already flipped

For a couple of years, the dominant narrative was that people were slowly warming to AI as they used it more. That trend has reversed. A joint Bentley University–Gallup poll found the share of US adults who say AI does “more harm than good” climbed from 31% in 2025 to 39% in 2026, while only 9% now say it does more good than harm. That’s not a rounding error — it’s a double-digit swing in a single year, on a subject that touches nearly every industry represented in daily news coverage.

The 2026 AI Index Report from Stanford HAI tells a more layered version of the same story: globally, the share of people who think AI offers more benefits than drawbacks actually rose from 55% to 59% between 2024 and 2025 — but the share who say AI products make them nervous rose right alongside it, to 52%. In other words, people aren’t choosing between excitement and anxiety. Increasingly, they’re holding both at once.

The familiarity paradox: more use, more skepticism

Here’s the twist that most wire coverage skips entirely. The conventional assumption in tech marketing has always been that exposure breeds comfort — get a skeptic to actually try the product, and resistance melts. AI sentiment data is quietly breaking that rule.

Research!America’s tracking since 2020 shows that as familiarity with AI has grown, the share of Americans who view it as a risk has climbed, while the share who see it as a benefit has fallen — and the drop is sharpest among 18-to-24-year-olds, the cohort using these tools most often. Quarterly tracking from SurveyMonkey backs this up at a granular level: daily or weekly AI use among Americans jumped from 30% to 38% year over year, and concern rose alongside it, from 30% to 33%.

People aren’t turning against AI because they don’t understand it. They’re turning more cautious the more they do.

That’s a genuinely different story than “public is scared of the unknown.” It suggests the honeymoon period — the phase where a new technology gets the benefit of the doubt simply because most people haven’t formed a strong opinion yet — is closing. What replaces it is a more informed, more transactional relationship: people keep using the tools because they’re useful, while simultaneously trusting them less.

Singapore and Southeast Asia are the outlier, not the norm

This is the part of the story that gets flattened in most “AI sentiment” coverage, which tends to default to US and European polling as if it represents the world. It doesn’t. The same Stanford HAI data shows that in Malaysia, Thailand, Indonesia, and Singapore, more than 80% of respondents say AI will profoundly change their lives within three to five years — and that expectation skews toward excitement rather than dread. The 2026 Ipsos AI Monitor finds the same regional split: people across Asia and Latin America are, on average, more likely to say AI has more benefits than drawbacks, while North America and Europe lean nervous.

For businesses and content publishers based in this region, that gap matters more than a single global average ever could. A US-calibrated assumption that your audience is AI-wary, and that every mention of automation needs a reassurance caveat, may simply be wrong for a Singapore-based readership that is statistically among the most receptive to AI-driven products in the world. Content and product messaging built around American anxiety data risks talking past a Southeast Asian audience that’s already three steps ahead in adoption.

Who’s actually driving the numbers

Averages hide the more useful detail: AI sentiment splits sharply along lines that rarely make it into a headline stat.

• Gender. Navigator Research puts overall US favorability at 49%, but that masks a real divide — 57% favorable among men, dropping to 63% among men under 55, against a net-unfavorable view among women.
• Age. Data for Progress found voters under 45 net favorable by 25 points, while voters 45 and older are net unfavorable by 10 points — almost a mirror image within the same electorate.
• Race. The same survey found Black voters favorable by 29 points and Latino voters by 10, while white voters were narrowly unfavorable — a pattern that gets little attention outside specialist polling outlets.
• Usage itself. Across nearly every dataset, the single strongest predictor of someone’s attitude isn’t demographics at all — it’s whether they use AI regularly. Frequent users trend more favorable even as their concerns rise in absolute terms.

Where the skepticism is landing hardest: infrastructure, not chatbots

One place the anxiety is turning concrete rather than abstract is data center construction. SurveyMonkey’s sentiment tracking found that opposition to new AI data centers centers on tangible local concerns: environmental pollution (71%), fears of higher utility bills and falling property values (65%), and a lack of transparency in community planning (59%). Nationally, Stanford’s data shows 41% of Americans think federal AI regulation won’t go far enough, against 27% who think it goes too far — a gap that has widened, not narrowed, even as more people use the technology day to day.

That’s worth pausing on: the public debate isn’t really “AI, yes or no” anymore. It’s shifted toward the physical footprint of AI — the power draw, the water use, the warehouses full of servers going up near residential areas — which is a very different (and much more local) fight than the one dominating most tech coverage.

What this means if you’re building something people will read or buy

The practical takeaway for anyone writing, marketing, or selling in this space is that “AI sentiment” is not one number you can quote and move on from. It’s at least three separate stories happening at once: a global optimism trend, a US-specific reversal driven by rising familiarity, and a sharp regional divide that puts Southeast Asian audiences closer to the optimistic end than almost anywhere else tracked.