Researchers have discovered that the same eleven specific words keep showing up in AI-generated creative stories — across different models, different prompts, different companies — and the statistical odds of that happening by chance are astronomical. If you thought artificial intelligence was a neutral tool that just spits back what you ask for, think again. Something is being baked into the code, and nobody outside the black box can tell you exactly what or why.
Forbes reported that when researchers asked large language models to write creative short stories, the same set of eleven nouns appeared prominently and consistently — even when prompting entirely different LLMs to compose from scratch. The finding defies the assumption that AI randomly assembles language. In theory, each new story should be unique. Two independent models shouldn't keep reaching for the same words. But they do.
The Forbes analysis traces the likely cause to three factors: the algorithms themselves, the data scanned during initial training, and the tuning that happens after the model is built. That last piece — the tuning — is where the ghost in the machine lives. Post-training adjustment is where engineers decide what the model should emphasize, avoid, or privilege. It's where bias gets embedded, and it's invisible to the user.
Forbes framed the discovery as a "mystery" and a "weighty and mysterious matter," soft-pedaling what it actually concedes: that not even top AI experts can explain precisely why these systems do what they do. The flow of calculations involves billions of values. The article admits we're expected to take it on faith that "if it seems to be working, and everything is okay" — then it's okay. That is not a standard any American should accept for technology that increasingly mediates what information people see, read, and produce.
Here's the stake: if eleven words can be quietly promoted across models without users' knowledge or consent, what else is being promoted — or suppressed? The same tuning infrastructure that nudges certain nouns into your story can just as easily bury topics, steer narratives, or flag wrongthink. This isn't speculation. The Forbes piece itself connects the phenomenon to the broader problem of AI hallucinations and unpredictability, noting that some experts believe these behaviors are unavoidable. If the system's outputs can't be fully explained by the people who built them, the people relying on them are flying blind.
The second source provided — an Omaha World-Herald ad placement for a Nebraska public school pride campaign — contained nothing relevant to this story. That a paid promotional blurb for government schools sits alongside reporting on algorithmic manipulation is its own quiet commentary on the state of media.
No lawmaker ordered these eleven words into your stories. No judge signed a warrant. That's the point. Censorship and narrative control no longer need the state when a handful of engineers can hardwire preferences into systems used by hundreds of millions of people. The First Amendment restrains the government. It doesn't restrain a tuned model.
The question isn't whether AI has bias. It does — by design. The question is who decides what that bias is, and whether Americans will be allowed to see inside the box before it decides what they're allowed to say.
