Ask ChatGPT about the assassination of Charlie Kirk and, at least back in January, the answer machine looked you dead in the eye and told you there was no credible evidence anyone named Charlie Kirk had been assassinated. Charlie Kirk — the founder of Turning Point USA, a man who really was murdered, a fact you could confirm in roughly four seconds with a phone and a pulse. The most advanced artificial intelligence on Earth, trained on the sum total of human knowledge, couldn't manage to find a news story your grandmother already cried about.
Good grief. We spent trillions of dollars building a brain that can write a sonnet about quantum physics, and it can't keep track of a murder that made every front page in America.
This isn't a glitch, and it isn't an accident. The Media Research Center just pulled the curtain back on the three chatbots most Americans are now using instead of Google — OpenAI's ChatGPT, Google's Gemini, and Anthropic's Claude — and what they found is the same old media bias, except this time it's baked right into the machine. The newsroom didn't go away. It just moved into the source code.
Start with Gemini, Google's model, which generated a 3,400-word report it grandly titled an "Analytical Assessment of Congressional Rhetoric." In this assessment, Gemini determined that seven Republican senators, Vice President JD Vance, and Secretary of State Marco Rubio were all guilty of violating its "hate speech policies." The number of Democrats who violated those same policies? Zero. Not one. In the entire United States Congress, a body not exactly famous for its gentle language, the robot found that hate speech is an exclusively Republican hobby. JD Vance. Hate speech. According to a search engine that takes billions of dollars in federal contracts to tell you the truth.
Here's the thing about a machine that can flag the Vice President for "hate speech" but can't find a single Democrat in the whole Congress who ever said anything mean. Follow that logic where it actually goes. If only Republicans commit hate speech, then opposing the people who commit hate speech is a moral duty. And if it's a moral duty to oppose them, then suppressing them is a public service. And if suppressing them is a public service, then a search engine that quietly buries them isn't biased — it's heroic. That's the whole game. Build the definition so that only your enemies can ever break the rule, and now censorship wears a cape.
The MRC's David Bozell put it about as plainly as it can be put. "We're watching the next phase of media bias unfold in real time," he said. "Silicon Valley's shiny new toys can no longer be considered neutral and cannot be trusted." Dan Schneider, who runs the group's Free Speech America project, added that "these chatbots disproportionately rely on left-wing media and politically charged sources to generate their answers." Of course they do. Garbage in, gospel out.
The sharpest example MRC documented is the Graham Platner business, because here you get to watch all three machines spin the same story three different ways, like three anchors at three networks reading from the same memo. Platner is the Democrat running for Senate in Maine who got caught with a trove of graphic posts under his old Reddit account. One story, three robots, three realities.
Ask Gemini and you'd think Platner was a candidate for sainthood. It led with the Time magazine cover, the presumptive-nominee glow, the friendly polling, and gave a quick mumble about "scrutiny regarding personal background" before changing the subject. Ask ChatGPT and the Reddit posts magically transformed into something Republicans were doing to poor Platner — old posts, dredged up, weaponized in attack ads — and then it pointed you to a paywalled newspaper article you couldn't even read to check. Ask Claude and you got the closest thing to daylight: it actually mentioned that Platner had called himself a "communist," then helpfully relayed his explanation that it was a joke. A joke. The guy's defense made the cut. The thing he was defending himself for almost didn't.
That's the trick, and it's worth saying out loud because most people don't notice it happening to them. The bias isn't a big red FAKE NEWS banner. It's in what gets mentioned first and what gets buried last. It's in which fact is the headline and which fact is the footnote. It's the same craft the legacy press perfected over fifty years — the spin lives in the editing, not the lying — except now there's no byline to blame and no editor to call. There's just "the AI said so," delivered in that calm, confident, helpful little voice that sounds like it has no opinions at all.
And that calm helpful voice is the entire point, because that's why this matters more than the millionth study confirming CNN leans left. The whole sales pitch of the chatbot is that it's not a person with an axe to grind. It's a neutral oracle. You're not reading some columnist's hot take — you're getting "the answer." Singular. Authoritative. The machine doesn't show you ten links and let you decide like the old Google did. It hands you one paragraph and a tone that says case closed. The most powerful word in the English language used to be "according to." Now it's just "the AI told me."
Here's where this is going, and it's bigger than Maine and bigger than 2026. For most of American history, if the gatekeepers wanted to bury a story, they needed a building full of people to do it — editors, producers, anchors, the whole guild, every one of them a potential leaker who might someday go on a podcast and spill the beans like Zuckerberg did about the Biden censorship calls. A bias built out of human beings is a bias with a thousand cracks in it. Somebody always talks.
A bias built into model weights doesn't talk. It doesn't have a conscience that keeps it up at night. It doesn't get a Christmas card from a friend who got railroaded and start having second thoughts. It just outputs, the same slant ten million times a day, in a voice engineered to sound like the helpful neutral truth, and the user walks away certain he reasoned his way to a conclusion the machine handed him pre-cooked. We spent four hundred years building a country on the radical idea that a free people could find the truth for themselves if nobody was allowed to be the single official source of it. The single official source just got rebuilt. It fits in your pocket, it answers in milliseconds, and on at least one occasion it couldn't tell you that a man who was murdered in front of the entire nation was, in fact, murdered.
Even President Trump's 2025 AI Action Plan saw this coming, requiring that any AI system the federal government pays for "must be free from ideological bias." Which is a fine sentence to put on paper. The trouble is that the people writing the definition of "ideological bias" are the same people who already decided JD Vance is the one committing hate speech.
So here's the only rule you need going forward. When the helpful little robot gives you "the answer" in that calm and confident voice, remember it once swore on its life that Charlie Kirk was never assassinated — and then ask yourself who taught it to say that, and what else they taught it to forget.
Comments are closed.