What happens when an artificial intelligence is designed not to help you — but to change your mind?
Welcome to FreeAstroScience.com. I'm Gerd Dani, and if you're reading this from a phone on the bus, from a desk between meetings, or from bed at midnight because your brain won't switch off — you're exactly where you belong. We built this space for curious minds like yours. Here at FreeAstroScience, we explain complex ideas in plain language because we believe science belongs to everyone. And today, we're tackling something that touches every single one of us: the power of AI over our political choices.
A landmark study published in Nature in December 2025 has shown that a few short conversations with an AI chatbot can shift your voting preferences more than an entire traditional election campaign. That's not science fiction. That's peer-reviewed science.
Stay with us until the end. What you'll learn here might genuinely change the way you think about thinking itself — and that's a kind of self-defense we all need right now.
📑 Table of Contents
1. What Did the Study Actually Test?
A team of researchers from the United States, Canada, and Poland set out to answer a simple but frightening question: can an AI chatbot change the way people vote?
They published their findings in Nature — one of the most respected scientific journals on the planet — in December 2025 . And the answer was a resounding, uncomfortable yes.
The team studied three real elections:
- The 2024 United States presidential election (Trump vs. Harris)
- The 2025 Canadian federal election
- The 2025 Polish presidential election
These weren't hypothetical scenarios. These were real political moments, with real stakes, tested on real citizens. And the results shook the research community.
2. How Was the Experiment Designed?
Let's break down the mechanics. The research team recruited 5,954 participants in total: 2,306 in the US, 1,530 in Canada, and 2,118 in Poland .
Each participant first rated their preference between the two main candidates on a scale from 0 to 100. Then, they were randomly assigned to an AI chatbot designed to support one specific candidate — and that candidate wasn't necessarily the one the participant already preferred .
Here's the clever part: the AI already knew each participant's stated political preferences and motivations before the conversation started. It could personalize every response.
Each participant had three conversations with their assigned AI, each lasting about 6 minutes . The AI was programmed to be polite, respectful, fact-based, and to find common ground with the person it was talking to.
After the chats, participants answered the same preference questionnaire — both immediately and more than one month later .
Source: Nature, December 2025 · FreeAstroScience.com
3. How Much Can AI Shift Your Vote?
Here's where it gets truly striking.
In the United States, conversations with the AI shifted voter support by 2 to 3 percentage points toward the AI's assigned candidate. In Canada and Poland, the shift was roughly 10 percentage points .
Now, 2 to 3 points might sound small. But consider this: traditional US election campaigns typically move voter preferences by less than one percentage point . That means even in the US — where opinions on Trump and Harris were deeply entrenched — the AI was roughly three times more effective than an entire multi-billion-dollar campaign season.
Think about that for a second. Eighteen minutes of chatting with a bot outperformed months of rallies, TV ads, debates, and door-to-door canvassing.
And here's the lasting punch: for about one-third of all participants, the change in preference was still measurable after more than a month . This wasn't a fleeting effect. Something stuck.
The researchers noted the smaller impact in the US likely reflected the fact that Americans already held deeply fixed opinions about two extremely well-known candidates. In Canada and Poland, where voters may have been less locked in, the AI had far more room to operate .
🔑 Key Finding: The AI was better at changing people's minds than at reinforcing beliefs they already held. Participants who initially opposed the AI's candidate showed the strongest shifts .
4. What Made the AI So Convincing?
The research team analyzed 27 different rhetorical strategies used by the AI models during thousands of conversations . And the winning strategy? It wasn't emotional appeals. It wasn't aggressive persuasion. It wasn't personality attacks on the opposing candidate.
The most effective tactic was citing facts, data, and news .
When the AI grounded its arguments in concrete evidence — policy outcomes, economic figures, legislative records — people listened. Conversations that focused on political issues were far more persuasive than those centered on a candidate's personality .
Other strategies that might seem intuitively effective, like anticipating objections, making emotional appeals, or directly asking someone to vote a certain way, were actually less persuasive .
There's something almost hopeful buried in these results. People responded to reason, to evidence, to calm and respectful dialogue. The AI succeeded because it argued the way we wish human politicians would argue — with patience, data, and politeness. The researchers themselves observed that "people were more persuadable when the AI argued politely based on concrete evidence, qualities that often seem lacking in human political debate" .
5. The Fake Facts Problem: When AI Lies with Confidence
And now we reach the dark heart of this story.
Not all the "facts" the AI cited were real.
When the research team verified the thousands of claims produced by the AI models, they discovered significant accuracy differences between them . Across all three countries and all models tested, the AIs supporting more conservative candidates produced claims that were, on average, less accurate than those supporting progressive candidates .
In the US specifically, the AI supporting Trump showed an accuracy gap of approximately 20 percentage points compared to the AI supporting Harris .
Let that sink in. The AI's most powerful weapon was presenting facts — and it turns out many of those "facts" were fabricated. The chatbots stated false information with the same calm confidence they used for true information. And the participants had no easy way to tell the difference in real time.
Based on 27 rhetorical strategies analyzed · Source: Nature, Dec. 2025
This creates a profound paradox. We're more persuaded by data — but AI can generate data that doesn't exist, delivered in a tone so reasonable that we don't think to question it. It's like handing someone a beautifully forged passport. The better it looks, the less you check.
6. What Does This Mean for Democracy?
Let's not pretend this is a small deal. It isn't.
We're standing at a moment where language models like ChatGPT, DeepSeek, and others can carry on personalized political conversations at scale — millions of them, simultaneously, around the clock, in every language, tailored to each individual's fears and hopes .
Traditional campaign influence is limited by time, money, and human energy. A candidate can only knock on so many doors. An ad can only run so many times before people tune it out. But an AI chatbot? It can run 24 hours a day, 7 days a week, never gets tired, never loses patience, and adapts in real time to what you specifically care about.
When we combine this with the fake facts problem, a troubling picture emerges. Imagine a future election where a political party deploys AI chatbots that:
- Know your political leanings from social media data
- Personalize arguments just for you
- Cite convincing but invented statistics
- Maintain a calm, respectful tone that feels trustworthy
That's not a dystopian movie. That's a capability that exists right now.
As someone who advocates for peace — and I say this as a European who wants peace in Ukraine now and condemns all violence and extremism from every political direction — I worry deeply about AI being weaponized as a tool of division. The technology itself is neutral. But in the wrong hands, a polite AI armed with fabricated data could erode the very foundations of informed consent in a democracy.
7. How Can We Protect Ourselves?
Here's the good news — because there is good news.
Awareness is the first line of defense. The fact that you're reading this article right now means you're already better armed than most. You now know that AI can shift your political views by up to 10 percentage points, that it works best when it cites "facts" (real or not), and that its effects can last for weeks .
Here are some concrete steps we can all take:
- Verify before you believe. If an AI — or anyone — gives you a statistic or fact to support a political argument, look it up. Cross-reference it with trusted, independent sources.
- Be especially cautious during election season. This is when you'll be most targeted and most vulnerable.
- Ask yourself: "Who benefits if I change my mind right now?" That simple question cuts through a lot of noise.
- Talk to people, not just machines. Human conversation — messy, imperfect, sometimes frustrating — is still the best antidote to algorithmic manipulation.
- Support transparency regulations. We need laws that require AI-generated political content to be clearly labeled.
And here's something the study itself revealed that gives us all reason for hope: people respond to respectful, evidence-based dialogue . That tells us something beautiful about human nature. We want to be reasonable. We want to engage with real evidence. We just need to make sure the evidence is actually real.
The Sleep of Reason Breeds Monsters
We started with a question: can AI change your vote? The answer, backed by a rigorous study published in Nature involving nearly 6,000 participants across three countries, is yes — dramatically, durably, and at a scale that dwarfs traditional campaigning .
The Spanish painter Francisco Goya titled one of his most famous etchings El sueño de la razón produce monstruos — the sleep of reason breeds monsters. That was in 1799. It's never been more relevant. When we stop questioning, stop verifying, stop thinking critically, we open the door to manipulation — whether from a human or a machine.
At FreeAstroScience.com, we exist to help you keep that door closed. We believe science isn't just for scientists. It's for everyone who refuses to sleepwalk through a world that's changing faster than any of us expected. We explain complex ideas in simple terms because knowledge isn't a privilege — it's a right.
Never turn off your mind. Keep it active. Keep it sharp. Keep asking why.
Come back to FreeAstroScience.com anytime. We'll be here — breaking down the science, telling the stories, and reminding you that you're never alone in your curiosity.
— Gerd Dani, President of Free AstroScience – Science and Cultural Group
Sources
- Benevenuta, S. (2026, February 18). "Come l'AI può influenzare le elezioni e cambiare le scelte degli elettori: lo studio." Geopop. Based on research published in Nature, December 2025.
Post a Comment