Has American Anxiety About AI Changed Since 2023?

The line that went viral

On August 18, the Pew Research Center published a short note on what Americans think of artificial intelligence. One line escaped it and made the rounds of English-language newsrooms: 52% of Americans say they're more concerned than excited about AI's growing role in daily life, up from 37% in 2021.

TechSpot, Forbes, The Register, Gizmodo, Euronews, ABC Columbia. Six write-ups, and six times the same comparison: today versus 2021. A fifteen-point gap over five years.

All of that is accurate. The 52% is right, so is the 37%, and there's nothing rigged about the comparison, it's Pew itself that offers it, in its own opening sentence.

The problem isn't the numbers. It's the four years you have to skip to get from one to the other.

The chart was right above it

On Pew's page, that sentence sits right after a chart. Six stacked bars, one per survey year, with the percentages printed in white inside the bars, at headline size.

They read, for the share of people who say they're more concerned than excited: 37 in 2021, 38 in 2022, 52 in 2023, 51 in 2024, 50 in 2025, 52 in 2026.

The 2026 number is the 2023 number.

Nothing is hidden, nothing is buried in an appendix: all six figures are printed right on the image, and the page offers the full table as a one-click download. The chart's caption, though, states that Americans have grown more worried over time. That's the line that traveled. The chart stayed put.

Four years, inside the margin of error

52, 51, 50, 52. The survey covers 3,488 American adults interviewed between June 22 and 28, 2026, with a margin of error of plus or minus 1.8 points.

Put differently, the gap between the highest and lowest of those four years is two points, on an instrument that claims 1.8 points of precision. Measure the same part four times with calipers and read 52, 51, 50, 52, and that doesn't prove the part is growing. It proves the calipers work.

Across those four waves, American opinion hasn't moved anywhere measurable. It's flat.

Eight months, then nothing

The real shift did happen, and it was sharp. Between the survey run December 12-18, 2022 and the one run July 31-August 6, 2023, the share of the worried jumped from 38% to 52%. Fourteen points in eight months.

Those eight months have a name: they're the ones that followed ChatGPT's launch in late November 2022. The public discovered the tool, tried it, talked about it, and flipped.

What the table shows next is that the shift stopped right there. Since summer 2023, the needle has been stuck. Three years of new models, record funding rounds and parliamentary hearings haven't moved the dial by a single meaningful point.

That's arguably the most interesting finding in the whole document. It's also the one that never made the rounds.

Pew spells out the plateau, right there

It's worth being precise about who's responsible here, because there's no manipulation and no cover-up involved.

Pew explains that concern has risen among both younger and older Americans since 2021, but that for Americans 30 and older, that rise was concentrated in the first two years. The sentence is right there, in plain English, immediately after the one everyone quoted.

The "since 2021" framing is a choice of starting point rather than a sleight of hand, and it happens to be the only one that produces a rising line.

An institute that publishes six numbers and highlights two is doing its job. A newsroom that picks up the two highlighted numbers is doing its job too. It's just that, by the end of the chain, not much of the original document is left.

The real move is under 30

Pew's own headline, in fact, isn't about Americans in general. It's about young adults.

Among 18-29 year-olds, the share who say they're more concerned than excited follows a very different path: 31% in 2021, 29% in 2022, 42% in 2023, 39% in 2024, 47% in 2025, 55% in 2026. The climb is uneven, but it keeps going, and 2026 marks a threshold: for the first time, a majority of under-30s land on the worried side.

This age group was the most upbeat in the panel back in 2021. It has now caught up with the 30-49 group. That's a real, dated shift, one that clears the margin of error by a wide margin, and it gets little attention because it's been absorbed into a headline number that, on its own, has stopped moving.

The one number that never climbed back

There's one column nobody talks about. Those who say they're more excited than concerned were 18% in 2021. They're 15% in 2022, 10% in 2023, 11% in 2024, 10% in 2025, and 9% today.

Enthusiasm has been cut in half, and its decline is the only steady trend in the entire table.

The rest of the panel, the ones who answer "equal parts both," make up 37% of Americans. That ambivalent majority routinely disappears from the coverage, and its absence changes the picture: without it, you'd picture 48% of people feeling at ease against 52% who are worried, when in fact three-quarters of that 48% are actually split down the middle.

Jobs, meanwhile, moved seven points

The second half of Pew's headline sits in a different question, asked in the same wave: will AI create or destroy jobs in the United States over the next twenty years?

71% of adults answer "fewer jobs," up from 64% in the same survey in 2024. Seven points in two years. Those betting on more jobs hold steady at 5%, exactly where they stood two years ago: it's the "won't change much" crowd and the undecided who shifted toward pessimism.

Three indicators moving, all measured in the same week, on the same 3,488 people: under-30s, enthusiasm, jobs. None of the three made a headline.

The one line Pew flags as different

One last detail, but it matters. The survey's technical documentation carries a footnote: for the November 2021 wave, the question wasn't worded the same way. It opened by explaining to respondents what an artificial intelligence program is, before asking what they thought of it.

The starting point everyone uses as their baseline is, in other words, the one wave the institute itself flags as not strictly comparable to the rest.

That same technical document lists the sample size, the margin of error, and the fieldwork dates. None of those three appear in any of the six write-ups reviewed.

Seven days after publication, no French outlet had picked up this survey. The headline, meanwhile, had already circled the globe twice.