Every benchmark went up. Public opinion went the other way. That divergence is now a business problem.
There was a comfortable assumption inside the AI industry: adoption would breed acceptance. Put the technology in enough hands, ship it into enough products, and people would come around the way they eventually came around to the internet, the PC, and the smartphone.
That bet is not paying off.
Pew’s latest research puts the number at 52% of Americans who say they’re more concerned than excited about AI showing up in daily life. In 2021, that figure was 37%. So while the models got dramatically better, sentiment moved fifteen points in the wrong direction.
It’s not an isolated data point either. A CNBC poll of 18-to-34-year-olds — the demographic that’s supposed to be the early adopter class — handed respondents a list of nine leading figures in AI and found most of them didn’t trust those people to act responsibly. An Economist/YouGov poll put over 70% of Americans in the “this is moving too fast” camp.
The generation that grew up on tech is the one applying the brakes.
This has stopped being a PR problem
Reputational damage is easy to dismiss until it starts costing money. It’s now costing money.
The Wall Street Journal reported that tech companies building data centers across the U.S. are running into serious local resistance and having to buy their way past it — job guarantees, clean water investments, community sweeteners. In one Louisiana parish, that reportedly extended to $50,000 bonuses for teachers.
More telling: the National Republican Senatorial Committee sent a memo to major AI companies warning that data centers are becoming a liability in a key Ohio race.
When AI infrastructure becomes an electoral risk, you’re no longer dealing with vibes. You’re dealing with a variable that shows up in permitting timelines, capex, and political capital.
The uncomfortable diagnosis
The instinct in Silicon Valley is to treat this as a communication failure — people just don’t understand AI yet, so explain it better.
I’d argue the opposite. People understand it fine. They’ve done the math and don’t like the answer.
Here’s the trade the average person believes they’re being offered:
| What they were promised | What they actually got |
| Cures, breakthroughs, more free time | Summarized web pages and a chatty TV |
| Automation that raises wages | Automation that threatens their job |
| A creative multiplier | Models trained on other people’s work |
| Better tools for learning | A cheating engine that devalues their degree |
Nobody needs a clearer explanation of that bargain. They need a better one.
And notice who’s absorbing the cost. Higher local utility loads, contested land use, strained grids, careers under pressure — those land on people whose daily experience of AI is a feature they didn’t ask for, bolted onto software they already paid for.
The retro rebellion is a signal, not a fad
Watch what people are opting into instead:
- Dumbphones and point-and-shoot cameras
- Algorithm-free iPods selling at a premium on eBay
- Cassette decks and CD players
- Quilting, knitting, jigsaw puzzles, Mahjong — “grandma hobbies,” suddenly everywhere
- Run clubs and in-person meetups outperforming dating apps
Marketers love to file this under nostalgia. It isn’t nostalgia. It’s a preference for things that are finite, legible, and don’t optimize you back. That’s a direct verdict on what algorithmic products feel like to live with — and AI is now the loudest example of that category.
Even the insiders are saying it out loud
Airbnb’s Brian Chesky recently acknowledged the backlash is real, and tied it to a simple failure: the industry isn’t shipping things ordinary people actually love. His example was a doctor on demand — something a person could never otherwise afford. Not a summary button.
Anthropic’s Dario Amodei went further, describing the perception problem as a “crisis of trust” and saying people suspect the industry is engineering some fresh way to take advantage of them. His own assessment of the fairest criticism aimed at AI companies, his included: they haven’t delivered on the big promises yet, and that’s on them.
When the people building this stuff concede the point, the argument is over.
What this means if you build or market anything with AI in it
This is where it gets practical for the rest of us.
1. Stop leading with “AI-powered.” In 2023 it was a premium signal. In 2026 it’s closer to a warning label for a large slice of your audience. Lead with the outcome. Let the method be an implementation detail.
2. Ship one thing that’s undeniably better, not ten things that are marginally AI. Feature-stuffing is what created the backlash. A single capability a customer would genuinely miss beats a dashboard of novelties.
3. Be specific about who benefits. Vague civilizational promises now read as evasion. “Cuts your intake paperwork from 40 minutes to 6” is trusted. “Transforming the future of work” is not.
4. Make the trade visible. What data goes in, what the person gets back, what they can turn off. Transparency has become a differentiator precisely because so few offer it.
5. Keep a human in the frame. The fear isn’t capability, it’s replacement. Products positioned as leverage for a person outperform products positioned as a substitute for one.
The industry raised enormous sums on the premise that AI was inevitable. Inevitability, it turns out, isn’t the same as welcome.
The gap won’t be closed with better narrative. It closes when someone ships a product where the ordinary person — not the investor, not the early adopter — looks at it and says I’d rather have this than not have it.
That product is still mostly unbuilt. Which, depending on where you sit, is either the problem or the entire opportunity.
If you’re building in this space: what’s the one AI feature your customers would actually be upset to lose? If you can’t name it, that’s the roadmap.
