NextFin News - The artificial-intelligence industry has a communications problem, or so the conventional diagnosis goes. The data say otherwise. Americans know more about AI than they did two years ago, yet they trust it less: familiarity with the technology has climbed to 70% of U.S. adults, up from 64% in 2024, while the share saying AI does more harm than good has risen to 39% from 31% a year earlier. The industry's credibility gap is not a messaging failure. It is a delivery failure, and it is now migrating from reputation risk into something more expensive: permitting, political, and capital risk.
The Broken Adoption Curve
Every technology adoption story follows roughly the same arc: unfamiliarity breeds suspicion, familiarity breeds acceptance, and time cures both. The AI industry has been betting on that arc since the generative-AI boom began. The bet is losing.
The latest Bentley University-Gallup Business in Society research, published in July, found that 70% of Americans now say they are somewhat or extremely knowledgeable about AI. In a normal adoption cycle, that number should be the story's happy ending. Instead it is the setup for the reversal: trust in businesses to use AI responsibly has fallen to 27%, down from 31% in 2025, halting two years of gradual improvement. Only 9% of U.S. adults believe AI does more good than harm. Meanwhile 52% say the technology does equal amounts of harm and good, and 39% — nearly matching the 40% recorded in 2023, before the current wave of commercial AI products — say it does more harm than good.
This is not a U.S.-only phenomenon, though the United States sits at the severe end of it. The 2026 AI Index Report from Stanford University's Human-Centered Artificial Intelligence institute found that globally, the share of respondents saying AI products and services offer more benefits than drawbacks rose from 55% in 2024 to 59% in 2025 — even as the share saying these products make them nervous climbed to 52%. The same report found that the United States reported the lowest trust in its own government to regulate AI responsibly of any country surveyed, at 31%, against a global average of 54%. Pew Research Center's June 2025 survey found half of U.S. adults say the increased use of AI in daily life makes them feel more concerned than excited, up from 37% when the question was first asked in 2021; just 10% say they are more excited than concerned.
The pattern is consistent enough to name: optimism about what AI can do is broad but shallow, and it is not translating into trust in the institutions building it. That gap — between belief in the technology and belief in its builders — is where the industry's problem actually lives.
The Demographic That Should Be Most Receptive Is the Least
If this were a normal hype-cycle comedown, the demographics would offer an escape hatch. Younger, more digitally native cohorts should be leading the recovery; they adopt fastest, forgive most readily, and have the most to gain from productivity tools. The data run the other way.
Among adults aged 18 to 29, the percentage saying they trust businesses at least somewhat to use AI responsibly fell from 30% in 2025 to 20% in 2026, while the share saying they have no trust at all rose from 29% to 41%. Nearly half of that age group — 47% — now believes AI does more harm than good, up 11 percentage points from 2025, the largest year-over-year increase of any age group. On jobs, the reversal is just as sharp: belief that AI will reduce the number of U.S. jobs over the next decade rose from 62% to 75% among 18-to-29-year-olds, a 13-point jump.
Walton Family Foundation and GSV Ventures research with Gallup reached the same conclusion for Gen Z specifically. Gen Zers' agreement that they feel excited about AI dropped 14 percentage points to 22%, hopefulness fell nine points to 18%, and anger increased nine points to 31%, even as anxiety held steady at 42%. The share who believe AI tools can help expedite work fell 10 points to 56%. And this is happening alongside rising usage: Gen Z's reported AI adoption has more than doubled since 2023, when only 21% reported any AI use.
That combination — usage up, sentiment down — is the clearest signal that this is not a familiarity problem waiting to be solved by more exposure. People are using the tools and coming away less enthusiastic, not more. The mechanism is straightforward: the first time a chatbot hallucinates a citation, the first time an AI summary erases nuance, the first time a worker is told a tool will augment them while the headcount plan says otherwise, the lived experience overwrites the marketing.
The Expert-Public Gap Is a Feature, Not Noise
Industry insiders often explain public skepticism as an information problem: people do not understand what AI can do, and education will close the gap. The size of the gap suggests otherwise.
Stanford's AI Index found that on how AI affects the way people do their jobs, 73% of experts expect a positive impact compared with just 23% of the public — a 50-point divide. Similar gaps appear for the economy (69% of experts versus 21% of the public) and medical care (84% versus 44%). On the jobs question, 64% of Americans expect AI to lead to fewer jobs over the next 20 years while only 5% expect more; experts are less pessimistic at 39% versus 19%, but they forecast far faster adoption, expecting generative AI to assist 18% of U.S. work hours by 2030 against the public's estimate of 10%.
Notice what that pairing means. The public is not anti-technology; it is skeptical of the distribution of gains. Experts forecast faster displacement and the public believes fewer jobs will follow. Both can be true if the benefits of AI accrue to capital and highly skilled workers while the costs land on everyone else. A communications campaign cannot fix a distribution problem. That is the first reason "better PR" is the wrong prescription.
From Reputation Risk to Permitting Risk
The second reason is that the credibility deficit is no longer contained in opinion polls. It is showing up on balance sheets and in local politics, where it carries a price tag.
Tech companies are facing a public-relations crisis over plans to build AI data centers across the United States, and the response has been to sweeten the deals: job guarantees, clean-water investments, and other community perks. In one case, teachers in a Louisiana parish received bonuses of up to $50,000, funded by the sales-tax windfall from a data-center project. Separately, the National Republican Senatorial Committee sent a private memo to top AI companies warning that data centers are hurting the party's chances of holding a vital Senate seat in Ohio, where Democrats have made the facilities a centerpiece of their campaign. A poll of 18-to-34-year-olds found that when given the names of nine leading figures in the AI industry, a majority said they do not trust each of them to act responsibly on AI — including 65% who said they do not trust Microsoft's Satya Nadella and 81% who said the same of Palantir's Alex Karp. And a May survey found that 71% of Americans think AI is advancing too quickly.
This is the transmission mechanism that matters for investors. A company that cannot site a data center without buying local consent is a company whose capital expenditure is rising for reasons unrelated to chip prices or power costs. Community-benefit agreements, job guarantees, and political concessions are not line items in the original AI build-out model. They are the tax levied on an industry that asked for speed before it earned trust. Every percentage point of trust lost becomes a permitting delay, a legal challenge, or a concession that widens the gap between the industry's growth assumptions and the infrastructure required to deliver them.
The industry has raised hundreds of billions of dollars on the promise of AI's inevitability. Inevitability is a financing asset only as long as the political and social license to build holds. That license is now the scarce input.
The Industry's Own Leaders Are Conceding the Point
When the criticism comes from outside, it can be dismissed as ignorance. When it comes from the chief executive of one of the field's most respected labs, it cannot.
Dario Amodei, chief executive of Anthropic, acknowledged this summer that negative public perception of AI is a "big problem" rooted in a "crisis of trust." People do not trust companies, governments, or the technology industry, he said, because they "suspect that we are cooking up some new way to screw them over." Then he assigned the blame where it belongs: "I think by far the most accurate criticism of AI companies, including Anthropic, is that we haven't yet delivered on our big promises to benefit the world. That is totally on us."
"I think by far the most accurate criticism of AI companies, including Anthropic, is that we haven't yet delivered on our big promises to benefit the world. That is totally on us." — Dario Amodei, chief executive of Anthropic
That is not a communications brief. It is an admission that the product has not yet matched the promise. Amodei's proposed remedy — deliver on the big promises, such as curing cancer — points to the only path that actually works: demonstration, not declaration.
Cyclical Backlash or Structural Deficit? The Call
The central question for anyone weighing this moment is whether it is cyclical — a normal hype-cycle trough that time and a few breakthrough applications will repair — or structural, a regime shift that will not self-correct.
This is structural, for three reasons. First, the reversal is concentrated in the cohort with the least reason to reject the technology: young adults who use it most. A cyclical backlash typically lives in older, less-adopted cohorts and fades as they age out. Second, familiarity has risen while trust has fallen, breaking the classic adoption curve in which exposure reduces fear. Third, the distrust is cross-institutional and cross-border: the public trusts neither the companies nor the governments that oversee them, and the United States ranks last among surveyed nations in confidence that its own government can regulate AI responsibly. When the credibility gap spans builders and regulators across multiple countries, it is not a news cycle. It is a governance and accountability deficit.
The cyclical case deserves its due. Hype cycles normalize; the first automobile was feared, the first internet boom crashed, and both reshaped the world. Adoption keeps climbing — 58% of employees globally reported using AI at work on a semiregular or regular basis in 2025 — and a single visible breakthrough, such as an AI-assisted medical discovery, could reset the narrative quickly. Global benefit sentiment did rise to 59% in 2025, which shows the reservoir of optimism is not empty.
But the cyclical case rests on the assumption that the industry's promises are merely early, not overstated. The jobs data, the distribution gap between expert and public expectations, and the emergence of local veto points over AI infrastructure all point to a deeper mismatch between the industry's growth model and the society it is reshaping. That mismatch does not mean-revert on its own.
The Second-Order Risk the Market Has Not Fully Priced
The first-order consequence of the trust deficit is reputational: worse polling, angrier town halls, more critical coverage. The market has priced that. The second-order consequence is not yet fully in the model: the trust deficit is becoming a cost-of-capital and deployment-speed problem.
Follow the chain. Eroded trust produces local opposition. Local opposition produces permitting delays and community-benefit concessions. Those concessions raise the all-in cost of AI infrastructure and slow the build-out cadence. Slower build-out, in turn, pressures the revenue-growth assumptions that underpin AI valuations, because the industry's expansion has been tightly coupled to compute deployment. The feedback loop closes the other way, too: if growth slows, the grand promises take longer to deliver, which deepens the trust deficit. This is how a communications problem becomes a cash-flow problem.
The strongest counter-thesis is that this is simply the hangover after an extraordinary hype cycle, and that usage will win the argument. The evidence for it is real: adoption is climbing, global benefit sentiment is up, and AI tools are becoming ambient in products people already use — email, search, phones, televisions. If AI becomes invisible infrastructure that quietly improves daily life, sentiment will follow, as it did with the internet and the smartphone.
The counter-thesis fails on one test: it assumes the costs of AI are as invisible as its benefits. They are not. The costs are loud and local — data centers consuming water and power, job displacement hitting specific occupations, deepfakes and hallucinations producing visible harms — while the benefits are diffuse and financial. An industry cannot sustain a social contract in which the costs are concentrated and the gains are abstract. Until that asymmetry is addressed, polling will not recover, and neither will the political license to build.
What Fixes It — and What Does Not
The remedy follows from the diagnosis. If the problem were perception, the fix would be messaging. Because the problem is delivery and accountability, the fix is operational.
Deliver the visible wins. Amodei's framing is correct: curing cancer, or any similarly concrete, human-scale benefit, does more for trust than a thousand brand campaigns. The industry needs outcomes people can point to, not capabilities it can demo.
Pay for the local costs. Community-benefit agreements should not be crisis concessions extracted under pressure; they should be standard operating procedure for AI infrastructure, priced into projects from the start. Water, power, jobs, and tax base are real costs imposed on host communities. Treating them as externalities is what created the backlash.
Be transparent about limits. Hallucinations, training-data disputes, and energy consumption are not going away. Disclosing them plainly — and building products that make uncertainty legible to users — costs less in trust than being found out.
Outlook: Three Scenarios and the Signal That Would Prove This Wrong
The base case is a slow, uneven recovery. Trust improves only as visible benefits accumulate and as the industry institutionalizes local accountability. That means the next 12 to 18 months bring more community-benefit deals, more permitting friction, and flat-to-slightly-improving sentiment — not a turnaround.
The upside case requires a genuine breakthrough: an AI-assisted medical or scientific discovery broad enough to make the technology's benefit tangible to non-users. In that scenario, the 9% who currently say AI does more good than harm could double, and the political license to build would widen quickly.
The downside case is a self-reinforcing loop: a high-profile AI failure — a deepfake-driven fraud, a major safety incident, or a wave of visible job losses — collides with an election cycle in which AI infrastructure is already a liability. Trust in business AI use, now at 27%, would fall toward the low 20s, and permitting delays would become the norm rather than the exception.
The falsifying signal for the structural thesis is quantifiable. If, over the next 12 to 18 months, the share of Americans saying AI does more good than harm rises above 20% — from 9% — while trust in businesses to use AI responsibly exceeds 40%, up from 27%, then this is a cyclical trough after all, and the delivery-and-accountability diagnosis is wrong. A secondary signal: if major AI data centers proceed through 2027 without community-benefit concessions, the political-risk channel described here is overstated.
The AI industry's PR problem will not be fixed by a PR campaign. It will be fixed the old-fashioned way: by delivering what was promised, paying for the costs imposed on others, and earning back a license that was spent faster than it was renewed. The companies that treat trust as infrastructure — something built project by project, not announced in a press release — will be the ones that get to keep building.
Explore more exclusive insights at nextfin.ai.
