Governments in the AI era
Trust Is the Product: Why Citizens Believe, or Do Not Believe, Digital Government
Kenroy George · 2026-11-09 · 7 min read
TL;DR: Digital government initiatives rarely fail because the technology is bad. They fail because citizens do not trust them, and trust is not something you can market your way into. Trust is a property of the system itself, inferred by citizens from direct experience: does it work every time, can I see what is happening, do I control my own data, is there a human when things go wrong, and does it tell me the truth about small things. Get those five right and trust compounds across every service you launch. Get them wrong and automation only accelerates the distrust.
The mistake almost everyone makes
When a digital government service struggles with adoption, the instinct is to treat it as a communications problem. Run a campaign. Put up billboards. Get the minister on the radio explaining how convenient the new portal is.
This almost never works, and the reason is simple: citizens do not form their beliefs about government systems from advertisements. They form them from experience. Their own experience first, and then the experiences of people they know. A cousin who lost an application. A neighbor who waited three months with no update. A friend who tried the portal, hit an error, and went back to standing in line.
Trust is not the message. Trust is the product. It is something the system either has or does not have, and citizens are remarkably good at detecting which one is true. You cannot announce your way past a system that quietly fails people.
So the real question is not how to communicate trustworthiness. It is what properties a system must have for citizens to correctly infer that it deserves their trust. I think there are five, and they follow from first principles rather than from any survey.
Reliability: it works every time
The first property is the most obvious and the least forgiving. The service has to work. Not usually. Not for most people. Every time.
The asymmetry here is brutal. One lost application undoes a hundred smooth ones, because the person who lost their application tells everyone, and their story is more vivid than any success story. Nobody calls their mother to say the permit portal worked as expected. Everybody calls their mother when the system swallowed their documents and nobody can find them.
This is why reliability engineering is not a technical nicety in government systems. It is the foundation of the entire trust relationship. A citizen who submits something to the state is performing a small act of faith. If the system drops that submission even once, the faith does not come back at the same price.
Transparency of state: I can see where I stand
The second property is that the citizen can always answer two questions: where is my application right now, and what happens next.
Opacity reads as arbitrariness. When a citizen submits a form and hears nothing, they do not assume an orderly process is unfolding out of view. They assume the worst, because the worst is what opaque systems have historically delivered: lost files, forgotten queues, decisions that depend on who you know. A silent system is indistinguishable from a broken one, so citizens rationally treat it as broken.
The fix is structural, not cosmetic. Every application should have a visible status, a named current stage, and a clear statement of the next step. When the status changes, the citizen should know. This is not a dashboard feature. It is the difference between a process the citizen is inside of and a black box they threw paperwork into.
Consent and custody: my data is mine
The third property is about who holds what. In systems built for trust, citizens hold their own credentials and choose what to share, with whom, for what purpose. In systems built for administrative convenience, data flows silently between agencies and the citizen finds out later, if ever.
Here is the thing about silent data collection: citizens feel it even when they cannot articulate it. There is a background unease with systems that seem to know more than you told them, and that unease becomes resistance. People start giving incomplete information. They avoid the digital channel for anything sensitive. They keep the important business on paper, where at least they understand the rules.
The alternative is a custody model where the credential lives with the citizen, presentation is an explicit act, and sharing is scoped and consented each time. This is not just an ethical position. It is the adoption strategy. A citizen who controls the flow of their own data has a concrete, personal reason to prefer the digital channel over the queue.
Recourse: a human path when it breaks
The fourth property is what happens when something goes wrong, because something will go wrong. No system is perfect, and citizens do not actually expect perfection. What they expect, and what trust depends on, is a visible path to a human being who can fix it.
Digital dead ends are trust killers of a special kind. A citizen stuck in a queue can at least argue with someone. A citizen stuck in a portal loop, with a help page that links back to itself and a contact form that never answers, has been abandoned by design. Every person who hits that wall converts from a user into an opponent of the whole program.
Recourse should be designed with the same seriousness as the happy path. A named escalation route. A real reply within a stated time. The ability to say a human reviewed this. The systems that survive their own failures are the ones that fail toward a person instead of toward a void.
Honesty in small things
The fifth property is the quietest one. Accurate wait estimates, even when the honest number is disappointing. Admitting an outage instead of pretending the site is fine. Plain language instead of legal fog.
Citizens calibrate constantly. If the portal says two weeks and delivers in two weeks, the next estimate is believed. If it says two weeks and delivers in two months, every future promise from that system is discounted, and the discount spreads to systems that had nothing to do with the delay. Small honesty is how a system earns the right to be believed about big things.
Trust compounds, and so does distrust
These five properties matter individually, but the real stakes are in how they combine over time. Trust transfers across services. A citizen whose first digital interaction with the state worked cleanly arrives at the second service already leaning toward yes. Each service that works makes the next adoption cheaper and faster. This is compounding in the truest sense.
Distrust compounds the same way, which is why the first service a government digitizes carries weight far beyond its own use case. It sets the prior for everything that follows. This is the deeper argument behind starting with a single painful queue and executing it flawlessly, which I laid out in the queue playbook. The first service is not a pilot. It is the reputation of the entire program, concentrated in one place.
The AI-era warning
There is a temptation right now to skip ahead. AI can automate document review, triage applications, draft decisions. Why not automate first and clean up the fundamentals later?
Because automating an untrusted process does not produce trust. It produces faster distrust. If citizens do not believe the system is reliable, transparent, and respectful of their data, then adding AI just means the thing they distrust now operates at machine speed and machine scale. Every flaw gets amplified, and the citizen has even less visibility into why decisions happen.
The sequence matters. Fix custody and consent first, so citizens control their own data. Fix transparency of state, so every automated step is still a visible step. Then automate, and the AI inherits a foundation of trust instead of a deficit. Governments that get this order right will find automation welcomed. Governments that get it backwards will find their most advanced systems are their least used.
Common questions
Can a communications campaign help at all? Yes, but only as an amplifier of real experience. If the system genuinely works, communication accelerates awareness. If it does not, communication accelerates the discovery that it does not, and the backlash lands harder because expectations were raised.
How long does it take to rebuild trust after a failure? Longer than it took to lose it, and the only path is through demonstrated reliability, not apology. A public, honest account of what went wrong helps, but the real repair is a long unbroken run of the system doing what it says. There is no shortcut, which is another reason to protect the first services so carefully.
Does self custody of credentials really matter to ordinary citizens? Most citizens will never use the phrase self custody, but they act on the instinct behind it every day. They notice which systems ask permission and which ones do not. Custody is how that instinct gets a technical answer, and it shows up in adoption numbers long before it shows up in vocabulary.
If you are thinking about how to build citizen trust into your government's digital services from the first credential onward, we would be glad to walk through it with you.