Half of Americans now use AI chatbots. Most don't trust the companies that make them. Pew found that 49% of US adults use a chatbot, and that 59% have little or no confidence that US companies will develop and use AI responsibly. That gap, between using something and trusting who is behind it, is the problem this piece is about, and we think it has a fix.
What people are telling pollsters
Gallup's May 2026 survey found that 39% of Americans say AI does more harm than good, up from 31% a year earlier. Only 9% say it does more good than harm, and 52% say about equal. Gallup also found that 27% trust businesses to use AI responsibly, down from 31%. Pew's February survey found that 63% say AI is advancing too quickly and 67% have little or no confidence in the US government to regulate it. This isn't a partisan split: Pew found 74% of Democrats and 61% of Republicans lack confidence in government regulation of AI.
People aren't rejecting the technology. They're using it. What they're saying is that nobody they trust seems to be in charge of how it arrives.
Every big technology has had this moment
ChatGPT launched to a celebration: something new that you could talk to. What it didn't come with was an answer to the next question people asked, which was what it is for and what it costs them. Earlier technologies faced the same question.
Railroads reshaped the country, and Congress responded with the Interstate Commerce Act of 1887, which created the Interstate Commerce Commission to oversee them. Highways needed new infrastructure and a way to pay for it: in 1956 Congress created the Highway Trust Fund with dedicated highway taxes, so the people using the roads helped pay for them. Air travel became routine after rules and institutions caught up. The Federal Aviation Act of 1958 created the Federal Aviation Agency after a mid-air collision over the Grand Canyon on June 30, 1956 killed all 128 people on board two airliners. Nuclear power and nuclear weapons came from the same science, and the technology had to earn public trust that it did not have at the start.
The pattern is that a new technology arrives, it needs new infrastructure and new rules, someone has to pay and someone has to be accountable, and it becomes part of daily life once that's in place. AI is a double-edged technology, like all of them. In our view it needs the same treatment.
The concerns, and what answers them
Electric bills. This is the concern with the clearest numbers. PJM, the grid operator covering 13 states, ran a capacity auction for 2026/27 that cleared 22% above the year before, and bills across those states were projected to rise by more than 5%. In Columbus, Ohio, the average residential bill was set to rise about $27 a month. The next auction, held December 17, 2025 for 2027/28, cleared at the price cap again, $333.44 per megawatt-day, and PJM procured a 14.8% reserve margin against a target near 20%. Nearly 5,100 megawatts of the 5,250-megawatt rise in forecast peak demand comes from data centers. PJM's Stu Bresler said data centers' demand “continues to far outstrip new supply.” The picture isn't uniform: Virginia's capacity price fell in the 2026/27 auction after transmission upgrades.
There are working fixes. Ohio regulators approved a settlement requiring new data center customers to pay for at least 85% of the energy they reserve each month, even if they use less. Oregon created a separate rate class for new or expanding data centers above 20 megawatts. The principle is simple: the companies adding the demand pay for the infrastructure it requires.
The industry has also made public pledges. In January, Microsoft said it would pay for the power its data centers use, fund local grid upgrades, pay full local property taxes without seeking tax breaks, and cut its water-use intensity 40% by 2030. OpenAI said each planned Stargate campus would pay its own way on energy and would not raise local electricity prices. On March 4, Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed the White House's Ratepayer Protection Pledge, which commits them to build, bring or buy the power their facilities need and to cover the full cost of the energy infrastructure involved. Supporters of the industry point to these pledges, and they are the right commitments. Our point is that a pledge isn't proof. Gallup's May trust numbers came after most of them were made, and the summary of the White House pledge we read doesn't mention community consent. What would change minds is a public record residents can check: what each facility pays, what it uses, and what the local bill did.
Local impact. Noise, water use and land are what residents raise when a data center is proposed nearby. We'd like to see companies publish expected energy and water use before a project is announced, and meet residents before permits are filed, not after.
Work. People worry about their jobs and their children's, and nobody knows how this will play out. Data centers themselves create fewer permanent jobs than many people expect. Meta lists 400 support roles at its Altoona, Iowa campus, Business Insider reported in 2025 that the largest data centers may employ up to 150 people and others 20 to 40, and Brookings found that counties hosting major data centers gained 2,000 to 4,000 jobs six years on, with simple estimates overstating the effect by about a factor of three. Mayors often cite the tax base, not jobs, as the main benefit. The honest answer is to say so, share what companies know from their own workforces, and give workers and communities a say in how change is managed.
Listening is part of the work
Part of the problem is where the conversation happens. Much of the industry's debate about AI takes place among people who already agree on what it's for, on X and at conferences, and the loudest voices there reward bold claims. Residents asking about their electric bill, or about a data center down the road, aren't in that room. That is our view of the pattern, not something we can measure. The fix isn't a better argument. It's going to where the questions are and listening before answering.
What a bridge looks like
1. Pay your own way on power, and publish proof: support rate rules that put new grid costs on the data centers that cause them, and show residents the numbers. 2. Be open locally: share energy, water and noise plans with communities early. 3. Listen first: hold meetings where residents can ask questions and get answers before decisions are made. 4. Be plain about what you know and don't: overstating, in either direction, costs trust. 5. Welcome oversight: accept independent testing and clear rules, as earlier industries did.
The industry has to do more, and it has to start by listening. AI may take years to deliver its largest benefits. Until then, public attitudes toward it will be set by electric bills and by whether anyone seems to be listening. A technology with a permanent opponent can't deliver on its promise. The way out is to make the opponents partners.
What this means, in plain English
Most people use AI and distrust the companies behind it. The clearest, fixable reason is the power bill. Making data centers pay for what they add, and talking with the neighbors first, would go further than any argument about the technology.
Sources: Pew Research Center, Americans and AI 2026, Gallup, Americans Cool Toward AI, DCD on PJM costs and state ratepayer rules, Power Engineering on the PJM 2027/28 auction, Enel North America on the 2027/28 auction, Federal Highway Administration, funding history, Jalopnik on the 1956 Grand Canyon collision and the FAA. KIRO 7 on Microsoft's Community First pledges, TheEnergyMag on OpenAI's pledge, NetChoice on the Ratepayer Protection Pledge, Straight Arrow News on data center jobs and local taxes. The Interstate Commerce Act of 1887 is cited from general historical record.