What happens when citizens of a country start trusting a US-engineered AI chatbot for political advice over their own local government?
A few months ago, a survey found that Gen Z Americans trust social media more than government institutions for political information. Some found this alarming and a crisis for democracy, but at least in such a case, the influencers are Americans and humans. They are subject to the same laws, elections, and consequences as their audience.
Now, imagine the same trust, in a country like Indonesia or Nigeria or some other from the global south, where this trust is not directed to an influencer but to an AI chatbot, built by a company in San Francisco. A chatbot has no citizenship, no actual memory, and no accountability to any election. It is a bot whose answers about your government, your politics, and your rights were shaped by an overwhelmingly Western, English, and American data.
This isn’t hypothetical but happening—right now.
ChatGPT crossed 100M followers faster than any product in history. A significant portion of these users are in the developing world. Here, it is often more responsive and more available. It is also more articulate than any government service they’ve encountered. Such an appeal towards an AI chatbot is obvious in countries where politicians are distrusted and bureaucracies are slow. Public information is also scarce or unavailable. When such a bot is asked something, it answers clearly and right away.
The problem, now, is what it is answering with.
The internet isn’t a neutral place, and large language models are trained on the internet. Studies have consistently shown that AI training data is dominated by English-language content. Whatever political assumptions baked into this content are distinctly Western: Liberal democratic frameworks, Individualist conception of rights, a particular understanding of how “God Governance” looks like. Now, when a citizen somewhere in a rural area in a developing country asks an AI bot whether their government’s media restriction is justified, they’re not getting a neutral answer. The answer they’re getting is filtered through a worldview that is built somewhere else, for someone else.
This matters because political advice is not like asking for a recipe. When an AI model tells you that a government policy is authoritarian, or that your leaders are corrupt, or that your rights are being violated, it is shaping how you understand your own country. If that AI is built by a corporation with its own interests, and trained on data with its own biases—answerable to no one in your country, it’s like you have handed a foreign entity enormous influence over your domestic politics.
Companies like Anthropic, Google, OpenAI—they’re private. They have shareholders. They are not embassies or aid organizations. They have their own political pressures inside the United States. And yet, they now function as a kind of diplomatic presence in countries where their governments have ever set foot. This is by virtue of their products being trusted.
History has a word for when a foreign power shapes how a population thinks of its own rulers. It used to require radio and printing presses; now it requires a good user interface and a free tier.
All this doesn’t necessarily mean AI is deliberately running influence operations. However, when a model frames western values as default common sense, and local governance traditions as deviations that require explanation, it does the political work without anyone deciding to do it. That is what makes it hard to see and easy to dismiss.
What is new is people trust these bots and believe in them. The conversation format creates an intimacy that a newspaper or a website doesn’t. You asked a question. It gave you an answer. It felt like a conversation. That is how they’re designed, and it has real consequences for how people weigh what a machine says and what people from their institutions say.
The answers, however, is not to ban these tools. They provide real value in many areas and people will keep using them regardless. It is the governments in the global south who should be asking whether political topics should trigger transparency notices about the limits of what the model knows about local context there. Also, should there be local AI development that reflects local values and languages, funded as seriously as they fund any other information infrastructure?
These arguments aren’t anti-technology, but the same arguments every country makes when a foreign media company or social platform starts shaping its public discourse. It’s the technology that is new and not the power dynamic.
Your government may be slow, corrupt, and frustrating. That is a legitimate grievance, and it deserves to be addressed. But the solution probably isn’t outsourcing your political judgment to a chatbot that was trained to sound trustworthy and built to retain users—made by people who have never heard of your town.
