Trust in U.S. government AI is not high – but it is negotiable.
37% of Americans would trust government AI vs. 40% who would distrust it.
Roughly 1 in 5 Americans categorically reject government AI across at least 1 of 4 independent trust signals; 7.1% reject it in every framing tested.
AI familiarity predicts trust nearly 5 times more sharply than political party.
Two consensus answers stand out across the dataset: 70% put accountability for AI failure on the government and 70% want a human in the loop on consequential AI decisions.
Only one government AI use case clears majority support – basic chatbots at 51%.
71% of respondents named at least one condition that would increase their trust.
Executive Summary: In May 2026, ZeroClick Labs surveyed 350 U.S. adults to gauge the public trust in AI entirely built, trained, and run by the government. The results revealed a relatively balanced split between trust (37%) and distrust (40%). Notably, the strongest predictor of openness toward government AI was not political affiliation but whether the respondent uses AI personally. In addition, 71% of respondents named at least one specific condition under which their trust could be earned, signaling that for most Americans, trust in government AI is not an outright refusal but a negotiation.
“Americans wouldn’t trust an AI created by the government!” – If asked the question, most people would probably align with this statement. In reality, the trust and distrust camps are evenly split – 37% to 40% – with the neutral 20% sitting in the middle, according to our recent survey.
While there is a real mistrust reflected in our data, it is conditional, with the majority ready to accept a government-created AI with proper safeguards in place. The picture that emerges from the data is not one of blanket rejection, but of a public whose trust is, in effect, negotiable – contingent on transparency, oversight, and accountability.
To measure American public trust in government-built, trained, and operated AI in 2026, identify conditions under which said trust can be earned or lost, and detect demographic/behavioral patterns that shape it.
| Total respondents | 350 |
| Fielded | Pollfish, May 4, 2026 |
| Geography | United States only |
| Female / Male | 193 / 157 |
| Age range | 18–64, mean ~46 |
| Region split | South 126 · Northeast 99 · Midwest 66 · West 59 |
| Top political affiliations | Republican 32% · Democrat 30% · Independent (any) 30% |
| Education | Bachelor’s+ 47% · Some college 26% · HS or less 27% |
| Weighting | Pollfish census weights applied (range 0.53–4.12) |
| Segment | N | % Trust | Net trust |
| By AI usage frequency | |||
| Frequent AI users (>= weekly) | 206 | 55.8% | +27.2 |
| Occasional users (monthly/rarely) | 83 | 13.3% | −37.3 |
| Never use AI | 56 | 5.4% | −60.7 |
| By party affiliation | |||
| Republican / Lean-R | 143 | 44.1% | +7.0 |
| Pure Independent | 38 | 39.5% | +7.9 |
| Democrat / Lean-D | 142 | 32.4% | −11.3 |
| Other / no party | 27 | 22.2% | −29.6 |
| By age | |||
| 18–29 | 30 | 40.0% | +3.3 |
| 30–44 | 119 | 46.2% | +7.6 |
| 45–59 | 144 | 32.6% | −9.0 |
| 60+ | 57 | 28.1% | −14.0 |
| By education | |||
| College+ | 163 | 45.4% | +11.0 |
| Some college / Associate’s | 90 | 24.4% | −32.2 |
| HS or less | 78 | 35.9% | +3.8 |
Self-reported political affiliation: Standard limitation.
Surprisingly, our study revealed no absolute answer to the question “How much would you trust an AI system designed, trained, and operated entirely by the U.S. government?”. The distribution between the “trusting” and “distrusting” camps is almost even: roughly 37% would trust government AI somewhat (26.00%) or strongly (11.14%), and roughly 40% would distrust it somewhat (20.29%) or strongly (20.00%). The rest, who neither trust nor distrust, sit right between them at 22.57%.
The surprise comes from the fact that the results of several studies, all of which measured broader AI sentiment, reveal historically high distrust toward AIs in public/government sectors:
Despite the slightly higher trust readout our survey produced, in the eyes of many Americans, government AI warrants not just an extra layer of suspicion but a whole different frame of reference, and categorical rejection patterns spread across the dataset confirm it.
The structural trust deficit goes much deeper than what preliminary results show. Across multiple independent questions, roughly 1 of 5 respondents gives a categorically rejectionistic answer regarding government AI, across four distinct signals:
What’s crucial to note is that the rejectionist tendencies have nothing to do with political inclination or a response to a specific policy. It’s a stable feature of how the public feels about government AI in 2026 - and the proof lies in the fact that these signals aren’t isolated. They stack:
What these results show: For roughly half of Americans, distrust in government AI is present in some form. For 7.1%, it’s a default state. The government is not entering this conversation from a position of trust. At best, it’s entering neutral. Most likely, it’s entering behind.
Since we’re talking about an AI created by the US government, most people would immediately assume that partisanship would be the key driver of trust. Most people would be wrong. According to our findings, the factor that dictates the openness toward government AI is familiarity, which can be confirmed if we cross-tab Q2 and Q3 to calculate the net trust score:
| Table 1: Q2 × Q3 Crosstab | |||||
| Usage group | N | Trust | Neither | Distrust | Net |
| Frequent (≥ weekly) | 206 | 55.83% → 56% | 15.53% | 28.64% → 29% | +27.18 → +27 |
| Never | 56 | 5.36% → 5% | 28.57% | 66.07% → 66% | −60.71 → −61 |
The net gap is immediately obvious, sitting at 88 points between frequent users and non-users. By comparison, a partisan gap is only 18 points, suggesting that familiarity with AI tools is a nearly five times stronger trust signal than political inclination.
While the opinions diverge across the majority of our queries, two “consensus answers” leave zero room for doubt, and they pertain to the burden of accountability in case of system failure and the mandatory presence of a human in the decision-making process.
If there’s one thing the supermajority (70%, combined) agrees on, it’s this: the burden of legal accountability if government AI causes serious harm is on the government - alone (38%) or jointly, with the AI developer (32%).
In stark contrast, only a negligible ~3% accept that “no one [is responsible], if the system followed approved rules.” The sentiment is clear: a system in which neither the agency nor the vendor owns the failure is NOT a position the American public is willing to entertain.
A system in which neither side owns the failure is a definitive no-go for the majority of Americans.
Once again, the answer is a decisive yes. The majority of respondents (~70%) want real human involvement in the decision-making process, either for preemptive review (~29%), human-only final calls (18%), or post-decision issue resolution (~23%).
Even without taking other results into account, the directional answer is more than clear: Americans don’t want AI making decisions that could have real-life consequences - not without the human in the loo
The above point is further reinforced by the fact that the “pro AI decision-making” segment is by far the smallest (excluding the indecisive 6.57%) - less than 9%. Even the strict “no AI in the loop” cohort beats it by a full 5%.
Simply put, no matter how you look at it, there’s no version of this question where this specific use case - automated decision-making - clears majority support. Which brings us to a much bigger question.
Only one: answering basic questions about government services. That’s right - the lowest-stakes interaction, the most benign use case and a very practical one is the only that managed to win a majority public support - and even then barely, at 51%.
Every other application doesn’t even come close. Even the first two runner-ups, identifying risks during natural disasters or public emergencies and detecting fraud or improper payment, only sit at ~37% each, after which the distribution keeps collapsing, albeit gradually.
No ambiguity here: Americans are willing to accept government AI for routine, low-stakes administrative work. However, when it comes to decisions that affect their individual rights and liberties (eligibility, surveillance, policing) - that’s where they draw a hard line.
It must be noted that our survey isn’t the only one to reach this conclusion. Cornell’s review of recent public-opinion research spotted the same division: support for AI on mundane workflows, fierce opposition for use cases with personally consequential outcomes.
An interesting thing to note here concerns the “absolute opposition” - a segment that wouldn’t trust government AI no matter what. They sit prominently at ~19%, suggesting that roughly 1 in 5 Americans aren’t a persuasion target, but a constraint for widespread adoption. But while this segment leaves no room for negotiation, the majority isn’t so categorical in their position.
Trust could increase if the government maintained AI safety, data security rules, public disclosure, responsibility in AI use, accountability, and freedom of choice. At a glance, this seems like quite a lot. In reality, these demands are not at all unreasonable.
Translated into policy terms, they amount to privacy law, audit regimes, real penalties, opt-out rights, transparency standards, and due process. In other words, the procedural infrastructure that the government routinely builds for other regulated activities.
Note that this sentiment is not fringe - roughly 71% identified at least one stipulation for trust increase, suggesting that the trust deficit is conditional - not categorical - for the majority of Americans.
Ultimately, the widespread public acceptance of government AI doesn’t hinge on which party is in power, who developed the system, or even who’s controlling it. It hinges entirely on the government's ability to demonstrate that its AI is safe, fair, and applied in ways that tangibly improve the lives of American citizens.
Americans don’t want to stop government AI - they want it governed.
Jordan Parkes, the CEO and founder of ZeroClick Labs, has been building and scaling digital marketing strategies since 2012, leveraging performance-driven SEO and data-based digital marketing solutions to guide the growth of hundreds of companies across the U.S. and Europe.
AI doesn’t grant trust. It cites those who earn it.
ZeroClick Labs is here to ensure your brand earns it.
Rely on us to run Generative Engine Optimization (GEO) services that align your content portfolio with AI visibility criteria – turning static resources into repeat citations across ChatGPT, Perplexity, Gemini, and AI Overviews.
“Our agency had no idea how to approach AI visibility. ZeroClick only does this one thing so they actually know what works. Worth every penny just to not waste time figuring it out ourselves.” - Jay