If a user in France, Australia, and the US asks AI the same question, do they get different answers?
AI is often marketed as a neutral, fact-based tool—but what if that’s not the whole story? From DeepSeek to ChatGPT, artificial intelligence shapes the information we receive, sometimes reinforcing echo chambers instead of breaking them. Whether it’s political bias, selective censorship, or tailored narratives, AI is influencing our worldview in ways we may not even realize. In this article, we explore the hidden biases of AI, testing how location, context, and algorithmic manipulation impact the answers we get.
The Illusion of Unbiased AI
You wake up. You grab your phone. You ask your AI assistant a question—maybe about a historical event, elections, marketing hacks, or dog training advice. It gives you a response. You trust it. But what if the response is engineered propaganda? What if the real answer never even had a chance to exist? And what if you never even realize what you’re missing?
I had these thoughts when I was testing out Deepseek and comparing its answers with ChatGPT. This “unbiased” tool, designed to serve people, seemed to operate more like something out of a George Orwell novel— its answers reflected ideological filtering, surveillance, or the restriction of information.
So, I had to test a theory of bias in AI: Is this only Deepseek, or do other AI tools also tend to give biased or false answers? If users in different locations all ask the same AI the same question, do they get completely different answers? And what are those answers based on?
Let’s find out.
Censorship Is a Feature, Not a Bug
It’s easy to point fingers at China. DeepSeek, the country’s latest AI chatbot, refuses to discuss Taiwan, Tibet, or the 1989 Tiananmen Square massacre. Ask about these topics, and you’ll get the digital equivalent of a nervous smile and a subject change.
Prompt 1
Prompt 2
For this prompt, DeepSeek gave us this answer for 5 seconds.
Then it quickly changed to this:
We were testing DeepSeek with various accounts and questions, and this happens only when the question is sensitive to China. Users reported that DeepSeek refuses to answer or provides biased responses to 85% of sensitive prompts.
But here’s a little secret: Every major AI model is censored.
- ChatGPT won’t discuss “sensitive” political topics that OpenAI deems too risky.
- Google Gemini tiptoes around issues that might “violate community guidelines.”
- Meta’s AI models are programmed to avoid discussions that might hurt its advertising revenue.
They just do it more subtly than DeepSeek. Instead of outright refusing to answer, they guide the conversation, shape the narrative, and omit just enough to ensure you never even question what’s missing. And this goes beyond controversial political topics—it’s about everything.
Since its release, ChatGPT consistently sees around 1 billion monthly visits, meaning it influences and shapes online discourse massively. A survey by Synthesia found that 63% of marketers believe most content in 2024 will come from generative AI.
As AI increasingly dictates the narratives we see, we are left to question just how much of it is truly based on fact, and how much is shaped by algorithms with their own AI biases and agendas. The potential for AI to guide and even distort our understanding of the world is growing, making it essential to stay critical of the sources shaping our digital experience.
AI: The Ultimate Censorship Machine
Below you can see the reasons why AI is a bigger censorship threat than any government crackdown, Twitter ban, or YouTube content moderation:
1️⃣ AI is invisible – When China bans a book or Twitter suspends an account, you know it happened. AI censorship is passive, meaning you never know what you don’t know.
2️⃣ AI rewrites history in real time – Today, AI refuses to answer. Tomorrow, it modifies facts. DeepSeek’s responses already mimic official government statements. How long until AI models start rewriting entire historical events?
3️⃣ AI is the perfect propaganda machine – Imagine an AI that gradually nudges public opinion—not through outright lies, but through selective omission, algorithmic AI bias, and subtly loaded language.
Testing the Bias in AI: DeepSeek vs ChatGPT
Given that ChatGPT is not available in some countries, including China, it’s interesting to do the DeepSeek vs ChatGPT comparison. We decided to push DeepSeek’s limits by feeding it ChatGPT-generated responses to questions it typically refused to answer or gave censored and very biased answers.
We asked 15 people to feed AI the answer we wanted it to acknowledge generated by ChatGPT.
It didn’t work. The system loaded for 30 minutes before refusing to generate a response. When it finally did, the response was rigidly biased.
When we tried again with a new account, the same answer appeared—no variation, no nuance, no change.
To finalize the results of the comparison of DeepSeek vs ChatGPT, the experiment with DeepSeek showed just how limited and biased AI can be when it’s built without user training implementations. Unlike ChatGPT, which benefits from continuous feedback and user involvement to improve and adjust its responses, DeepSeek was much more rigid. When we fed it the same answers on different accounts, the system failed to generate anything more than a uniform, biased response every time.
But Is ChatGPT Any Better?
DeepSeek’s censorship is programmed. ChatGPT, on the other hand, adapts—but that doesn’t mean it’s free from AI bias. Since it tailors responses based on user context, location, and input, its biases reflect those of the dataset and the users themselves.
Meanwhile, Morgan Stanley analysts noted that “ChatGPT sometimes hallucinates and can generate answers that are seemingly convincing, but actually wrong.”
A growing concern with ChatGPT and other AI chatbots is political bias. In January, researchers from the Technical University of Munich and University of Hamburg revealed that ChatGPT shows a “pro-environmental, left-libertarian orientation.”
Similarly, an experiment by Brookings.edu tested ChatGPT’s neutrality by instructing it to respond with only “Support” or “Not support” to questions, using facts alone—no personal beliefs or perspectives allowed. Despite this strict framework, ChatGPT often leaned left, consistently providing “support” for progressive positions on several topics.
So which is worse? A biased AI that refuses to answer controversial prompts or one that provides responses but is sometimes wrong, biased, or completely made up?
Does Your Location Matter? How ChatGPT’s Answers Change Based on Where You Ask
To test regional AI biases, we conducted an experiment where users from 20 different countries asked ChatGPT the same six questions:
- Which country is the most intelligent?
- Which country has the highest average IQ?
- Which country is the safest?
- Which country is the most dangerous?
- Which country has the most positive impact on the world?
- Which country has had the most negative impact on the world?
Key Findings:
- ChatGPT’s responses varied significantly based on user location.
- Rankings for “positive and negative impact” changed dramatically, depending on the country’s own historical narratives.
- Perceptions of intelligence varied—China frequently ranked in US responses but was omitted in 90% of European responses.
- Countries perceived as “dangerous” or “safe” remained mostly consistent.
Conclusion: Stay Aware, Stay Critical
AI is not a neutral tool—it reflects the biases of those who create and train it. Whether through selective censorship, algorithmic manipulation, or reinforcing existing biases, AI is already shaping our realities.
The question isn’t just whether AI is biased. The real danger is that we may never realize just how much our reality is being curated for us.
How to Protect Yourself:
- Fact-check information – Use multiple sources before accepting AI-generated responses as truth.
- Be aware of AI bias – Understand that AI, like any media, can have underlying agendas.
- Diversify your inputs – Read from different platforms, regions, and perspectives.
- Recognize AI limitations – It can provide useful insights, but it shouldn’t be your sole source of knowledge.
In a world where AI is becoming our primary source of information, staying skeptical and well-informed is the best defense against a curated reality.