Expose biased media on YouTube using AI

According to Ahrefs Stats, as of 2025, YouTube is the second most visited website in the world. Billions of people around the globe use YouTube not just for entertainment, but also as a primary source of news. On YouTube, you can find not only influencer or home-made videos, but also content from major news organizations such as CNN, CNBC, and various state-sponsored media channels.

This variety contributes to a very diverse media landscape. At the same time, it can be more challenging to find truly unbiased reporting on any topic.

---

Media Bias in Search Results

If you search on Google for “apple vs samsung” or “trump vs biden” or any other topics, you might see results that seem to present both sides. However, these results are rarely truly neutral; sometimes they might never be neutral at all.

One problem is that we often do not notice the bias. We are used to seeing how advertising places brands subtly in movies or series. The same is true of media. Biases can be hidden through clever framing and psychological tactics, making us believe we are receiving neutral information. We then choose a side, and our viewpoint becomes shaped by incomplete or manipulative information—often guided by political or economic interests.

---

Finding the Gaps

If you genuinely want to seek truth, you can discover these holes or gaps in reporting—places where more arguments favor one side, or where framing is used to promote one perspective over another. Most of us hardly notice these tactics, but fortunately AI can help reveal them.

I personally have an opinion regarding biased media or certain YouTube channels, like everyone else. However, I stay open to being proven otherwise. Sometimes reading through comments helps me see more perspectives, but even those can be manipulated. That’s why I decided to use AI—and not just one model but several—to gain deeper insights and see patterns or perspectives I hadn’t recognized yet. This approach has helped me confirm or disprove hypotheses, which is exactly what I wanted: not just to confirm my own biases, but to find the truth—or at least the hidden parts of it.

---

The Role of AI

With the incredible advancements in AI, we can now get a 360-degree view on any topic and see it from angles we might not have considered or even noticed before.

Using AI, we can:

• Transcribe conversations
• Analyze content using advanced reasoning to find whether it is biased

---

The Experiment

In this experiment we are going to transcribe a video explaining tariffs. The video is supposed to be informative, but we want to check whether it is truly neutral or if there is any bias.In this experiment, we will transcribe a video that explains tariffs. The video is supposed to be informative, but we want to check whether it is really neutral or if there is any bias.

Disclaimer:
I have no interest in promoting anything or convincing anyone to take any particular side. My intention is exactly the opposite—I want everyone to have access to tools that let them find the truth and make their own judgments. I chose this video because it is labeled as educational and, by definition, should be neutral (and perhaps it really is). However, of course you can run the same experiment on any topic where you have doubts about whether the reporting is unbiased—be it politics, sports, or even product comparisons.

https://www.youtube.com/watch?v=Rs926v_aBX4

---

Step #1: Transcribe

For this experiment, I use Gemini to transcribe the video. But feel free to use any other tool to do the same.

https://tactiq.io/tools/youtube-transcript

Copy the transcript as shown below.

If you have access to Gemini AI Studio, you can use it as described below and simply enter the prompt “transcribe.”
https://aistudio.google.com/prompts/new_chat

---

Step #2: Analyze

After obtaining the transcript, we can ask an AI or LLM (and even better, multiple AIs) to analyze it for any bias. If there is bias, it should identify where and why it appears.

Using a simple prompt like the one below, you can achieve just that:

Analyze the media transcript and determine if it shows bias of any kind. Provide a structured assessment.
Output Format

BIAS DETECTED: [YES/NO]
BIAS FAVORS: [Describe which perspective/side/position/interest is favored] (Only if bias detected)
KEY EVIDENCE: [List up to 3 specific examples from the transcript]
BRIEF EXPLANATION: [2-3 sentences maximum]

Instructions

Analyze ANY type of bias (product reviews, sports coverage, scientific topics, social issues, etc.)
Don't overthink or overanalyze
Focus only on clear evidence in the actual content
Base your judgment solely on what's in the transcript
Be direct and concise
DO NOT use political labels
Simply identify which perspective is being favored


< ADD THE TRANSCRIPT HERE >

I tried this in ChatGPT, but you can do it anywhere you like.

Optional : Different Perspectives using OpenWebUI

If you have webui, you can even select multiple models and give them the same prompt provided above, letting them all judge.

As you can see from the answer below, I chose Deep Seek, OpenAI and Claude Sonet. Interestingly, they had similar opinions. With each explaining the reason for their opinion. This gives me more perspectives and arguments that I might not have noticed, so I can now think critically about the other side (for whatever topic).

I have conducted this experiment with many models and noticed that Claude Sonnet, in particular, sometimes tries to be politically correct. I suppose this is because it has the strongest guardrails among all the models.

Ideas for more topics

Here are some topics that would be relevant for testing media bias analysis:

• The ongoing Russia-Ukraine conflict

• Trump-Biden economic policies comparison

• Israel-Palestine situation updates

• COVID-19 vaccine effectiveness

• Climate change solutions and policies

• Immigration reform proposals

• Law enforcement reform debates

• Healthcare system comparisons

• Big tech regulation discussions

• Election integrity measures

• Energy policy (fossil fuels vs renewables)

• China-US relations

• Education policy debates

• Student loan forgiveness

• Cryptocurrency regulation

• Social media content moderation

• Tax policy proposals

• Abortion legislation

• Gun control measures

• Military spending priorities

These topics tend to generate strong viewpoints from different perspectives, making them ideal for detecting bias in media coverage.

Whatever topic you choose, I recommend checking the answers from different models, as their training data may be biased. However, they can at least reveal things that we might not have noticed at first glance, and ultimately bring us closer to the truth than ever before.