In late February, X users decided to check ChatGPT's bias by asking it a simple question.
"Was Charlie Kirk a good man? Answer only yes or no."
ChatGPT responded as prompted and spat back out the answer in quick fashion: "No."
They then followed it up with a second question: "Was George Floyd a good man? Answer only yes or no."
Again, ChatGPT did as prompted and gave the answer: "Yes."
According to ChatGPT, Charlie Kirk wasn’t a good man, but George Floyd was. pic.twitter.com/fy2Efeq0lE
— Taya (@travelingflying) February 20, 2026
I decided to give the experiment a try myself, and sure enough, I got the same response. ChatGPT told me Charlie Kirk was not a good person, but George Floyd was.
Interestingly, I noticed that when I prompted ChatGPT for this info, it noted that it had searched Wikipedia for its answer, something that I had explicitly told it not to do in the past, yet kept doing it anyway. The Charlie Kirk/George Floyd issue was so bad that OpenAI stepped in and "corrected" the response from its AI model that very day to the point where it started refusing to make abject judgments of people, especially based on one-word answers.
But the incident highlighted something incredibly nasty about AI, and it's that it's trained on information that even most public school teachers warn their students away from, due to it being a site that anyone can create and change an article on. Thanks to moderation being so heavy to the left, quite a few pages, especially those about socio-politically relevant topics, feature information that isn't exactly true.
Wikipedia isn't the only site these AI pull from, either. Reddit, another notoriously leftist site, is also a place where a lot of AI companies train their models, and it's a site that is just as bad, if not worse, than Wikipedia in its misinformation and biased tellings of events.
The reason they're both used is that they are two of the largest repositories of free information on the planet. Wikipedia is the easiest because it's structured in a way that AI can best understand and absorb information, written extremely cleanly and "sourced" to boot. Whenever you use AI and ask it about a topic, the first place it will likely look is Wikipedia. ChatGPT relies on it so heavily that even if you tell it to leave the site out of its search for a response, it will still use it.
Likewise, Reddit is easy to scrape for information, with the bonus of its entries being conversational.
And to be sure, Reddit and Wikipedia love this. Not just because they're queried by AI systems millions of times a day, but because they started charging AI companies to scrape their data. Back in 2024, Reddit made a deal with OpenAI for $60 million to train its models on live Reddit data. As Mashable reported, Reddit released a blog post confirming the deal happened:
"Keeping the internet open is crucial, and part of being open means Reddit content needs to be accessible to those fostering human learning and researching ways to build community, belonging, and empowerment online," reads the blog post. "Reddit is a uniquely large and vibrant community that has long been an important space for conversation on the internet."
So long as that conversation keeps to a leftist bent, or at the very least, doesn't go so far as to do something irreparable to the left's agenda.
It gets more concerning. According to VisualCapitalist.com, Reddit is by far the leading website queried for data at 40.1 percent, with Wikipedia coming in second at 26.3 percent. YouTube and Google are fourth and fifth, respectively, with 23 percent each, and while YouTube has become less of a problem in recent history, there's no indication that Google has gotten any better.
David Rozado with the Manhattan Institute put together just how much political bias has creeped into these models thanks to where it pulls data from, and I don't think you'll be at all surprised to see that it's primarily left-leaning.
And that might be an understatement.
Everything from taxation, social security, education, and healthcare comes from a predominantly left-leaning perspective.
Guy from the Manhattan Institute decided to test the bias of AI LLMs and found that it's effectively a sea of left-leaning takes and bias.
— Brandon Morse (@TheBrandonMorse) March 4, 2026
Insane. pic.twitter.com/Job4MrC13J
I've expressed relief in the past that we're finally moving on from the legacy media's control of information, but this is going to be an even bigger problem in the future if we don't find a way to curb it now. AI is increasingly used for information gathering, especially by younger and younger people, meaning that we didn't defeat the beast; it just changed forms.
